If you have ever had the warehouse swear blind you have units on the floor while NetSuite says zero, you already know the pain. It turns into the same mess of exports, emails and a snap decision that ends in cancellations or dead stock.

NetSuite should be the place you control inventory, and for most brands it eventually is. What breaks first is the assumption that control means one number. That falls apart the moment you run more than one location, or more than one definition of sellable.

Your stock number isn’t one number

There are a handful of stock buckets. Teams will mix these up and end up arguing which one's right, distracting them from infinitely more valuable work:

Physical stock is what the warehouse can actually locate. It’s specific to a bin and a status. It includes the awkward stuff like damaged or quarantined units, plus the cartons nobody can find.

Recorded stock is what NetSuite says exists based on transactions. Receipts, fulfilments, transfers, adjustments and the like move this number around. Finance cares because it drives valuation.

Sellable stock is recorded stock after you apply the rules you use to decide what can be allocated to an order. Location matters, inventory status matters, and commitments matter.

Published stock is what you let the public buy. Often you're taking what NetSuite holds and deducting from it buffer stock and the like, to reach the figure you share with your sales channels. It should trace back to NetSuite with named deductions that merch can control in NetSuite.

This is where it starts to unravel because NetSuite’s just doing NetSuite. On hand, committed, available and backordered are different buckets; if you treat them as interchangeable, you’ll end up reconciling noise and shipping problems.

It gets worse because other systems use the same labels to mean different things. It’s common for the warehouse on-hand number to include inventory that’s still in intake, while NetSuite only reflects it once someone’s actually done the receiving.

That’s comparing two different states and calling it a discrepancy.

Sometimes it's the opposite. Two brands can run near-identical setups, a few tens of thousands of child SKUs across a couple of locations, and one will sit within a handful of units of NetSuite at any point in the day while the other is out by hundreds and can't tell you why. The difference is whether the team doing the recon agreed on what they were measuring and when.

Timing matters. Treat it like part of the setup, not an annoyance.

Real-time doesn’t make it true

Real-time stock events cut lag. They don’t make the number right.

Real-time integrations can still miss weird sales order edits, especially when lines get changed after the initial order comes in. The boring fix is a scheduled full sync or comparison, so missed events don’t drift forever.

If you want SKU visibility, stop chasing a prettier dashboard and start insisting on a consistent level of detail. In a fashion or multi-location setup, the smallest unit that matters is a child SKU, by location, by inventory status, with a timestamp.

Anything more aggregated is how you convince yourself stock exists when it’s the wrong size, the wrong warehouse, or the wrong status.

Make the variance explainable

The most useful report isn’t total stock. It’s an exceptions report that tells you exactly how you got from NetSuite availability to what the channel is publishing.

Go look at your own reconciliation today. If a discrepancy shows up, can an operator tell whether to wait because an inbound receipt is still in intake, or whether they need to fix the transaction, change the status, or escalate it?

If the answer is no, you haven’t got visibility. You’ve got visibility theatre.

Start by separating the two boundaries you’re actually managing.

Warehouse to NetSuite is about whether the physical work has actually landed in the ERP. Missing receipts, transfer mismatches, and status mapping failures all live here. So do warehouse-only states. If the warehouse moves stock into locked or suspense because it can’t be found, and that state doesn’t map to a NetSuite status or location that affects availability, your website will happily keep selling ghost units. Then you get the death spiral. Short shipments, negative adjustments, stock mysteriously reappearing a week later when someone finds the carton, and nobody trusts any number.

NetSuite to channel is about whether the intended sellable quantity has been published correctly. Latency windows and missed events live here.

A single warehouse to Shopify comparison can’t tell you which boundary broke, so it can’t tell you what to do next.

Keep your fixes separate

If units exist but shouldn’t be sold, make them unavailable by status or a controlled location model. If units exist but are in the wrong place, transfer them properly. If you care about the in-transit state, use transfer orders so you can see the stock move from committed to in transit to received.

If you’re deliberately reserving stock for stores, wholesale, or a strategic drop, that’s a commercial holdback. If you’re hiding stock because you don’t trust the warehouse versus NetSuite position yet, that’s a temporary reconciliation buffer, and it needs a reason, an owner, and an expiry.

And if you’re using inventory adjustments to stop Shopify selling something, you’re using an accounting transaction to solve a storefront control problem. NetSuite adjustments change quantity, and they can change value. They post to an adjustment account. They should close a confirmed variance after you’ve done the work, not paper over uncertainty.

Counts need the same discipline. NetSuite takes a snapshot at the start of a count, and if stock keeps moving during the count window without a process that accounts for movements, you can manufacture variances that were just normal receipts and picks.

Once you’ve got this right, NetSuite becomes what ops needs it to be: the place where stock gets controlled and surfaced. And you can stop living in spreadsheet land where everyone has their own version of stock and nobody can prove they’re right.

SKU visibility is being able to explain, at a consistent unit and point in time, why physical stock, NetSuite stock, sellable stock, and published stock differ.

If you can’t explain the difference, you don’t have a stock figure. You’ve got an argument waiting to happen.

Two events in, and the format is the same. Brand show and tells from operators who have put something live, followed by small group sessions built around the challenges attendees bring when they register. Most of it comes back to practical AI adoption: what is earning its place in the operation and what is noise.

How the day runs

Lunch from midday in the ground floor lounge, which gives everyone an hour to talk before anything formal starts. Content runs from half one to half four, with the main room split into sections so the small group work can happen in parallel rather than one table at a time. Drinks from half four. There is something on in the evening afterwards for anyone staying over, and we will confirm that separately.

Who is in the room

Around 70 senior operators from brands in fashion, beauty and lifestyle, all of them brand side. They come from ecommerce, digital, product, tech, data, ops, logistics, CX, merchandising, commercial and finance, because most of the problems worth an afternoon do not sit inside a single function.

Sponsors and partners are in the room to facilitate the small group sessions and to take part in them. Every show and tell slot goes to a brand.

What people are working on

Registration asks you to pick up to three challenges from a list of fourteen, plus anything else you are wrestling with. Those answers set the session topics and the table plan, so the two minutes it takes are worth spending. A sample of what is on the list:

• Range planning and buying decisions made on incomplete data

• Getting useful output from the tools you already have

• Product data quality at the point it is created

• Stock visibility across channels and markets

• Building internal buy-in for AI adoption

• Working out where AI helps and where it is a distraction

Why Stockholm

OX grew out of a straightforward observation. The conversations brand-side operators actually need are not happening at the events that exist. The advice landscape is loud, vendor-heavy, and mostly pointed at the wrong people. What operators kept asking for was time with peers in similar brands, doing similar work, prepared to share what worked and what did not, without flogging something at the end of it.

Stockholm is the next room. Dema are back as headline sponsor, on home ground this time. Centra, Emfas and Resourced are alongside them. HighCohesion, Vervaunt, Loop, Torque, Pro Carrier and Whānau are travelling over from the UK.

Travel and staying over

Rooms are held at the venue at a discounted group rate for both the night of the 21st and the night of the 22nd. Registered attendees get the code and book direct, so hold off on booking anything else for now. There are fewer rooms held for the 21st, so if you are travelling in the day before, say so when you register.

For UK members, there is no expectation that anyone travels for this one. If Stockholm appeals and you want to be in the room, tell me. Between us and our partners there may be a way to make that work, so it is worth asking rather than ruling yourself out.

How it works

Registrations are read and approved one at a time, and we come back to you either way. OX members and everyone who came to OX1 or OX2 have first pick of the room, before opening to our wider network.

The room is for senior operators from brands in fashion, beauty and lifestyle. If that is you, or someone you know, the details are at operatorexperience.ai.

This is the demand and inventory planning gap, and it’s close to universal among the brands we speak to - particularly the ones running a core system for finance alongside a separate planning process that finance never fully sees.

The target and the buy are built in different rooms

Finance sets a revenue target, usually top-down, informed by growth ambitions and last year’s numbers. Merch turns that target into an actual buy - which products, how many, in what sizes, from which suppliers - usually in a completely separate tool, on a completely separate timeline, with no shared translation layer between the two.

Both processes are reasonable on their own terms. The problem is that “reasonable on their own terms” doesn’t guarantee “adds up to the same number,” and by the time anyone notices the gap, stock has already been committed.

Amy, who owns Product Ops internally at CT, years as a merchandiser and a head of planning before this, puts it simply: the finance number and the merch number are almost never built from the same starting assumption, and nobody’s job is to reconcile the assumptions before the numbers get compared.

It’s a data integrity problem before it’s a reporting problem

The instinct, when finance and merch numbers disagree, is to build a better dashboard that reconciles them after the fact. That helps people see the disagreement faster. It doesn’t stop the disagreement from being built into the plan in the first place, because the actual root cause sits earlier - in whether both teams are working from the same, governed version of the underlying product and sales data, or from two versions that have quietly drifted apart.

That matters even more now that more of this analysis is AI-assisted. Feed a forecasting or reporting tool inconsistent, ungoverned data, and it won’t flag the inconsistency - it’ll just produce a confident, well-formatted answer built on top of it. What you put in is what you get out, whether the thing doing the calculating is a person with a spreadsheet or a model with a prompt.

Why more reporting doesn’t fix it

The actual fix sits earlier: a shared translation between the target finance sets and the buy merch makes, built on data both teams actually trust, checked at the point of planning rather than discovered at the point of reporting. That’s a process and data-integrity change before it’s a tooling change, even if the tooling ends up needing work too.

What this actually costs

Every quarter this goes unaddressed is a quarter where a board pack gets two numbers for the same department and someone has to explain, again, why they don’t match. That’s not a data hygiene issue. It’s a planning process that was never designed to produce one number in the first place.

Fix the translation layer and the underlying data agreement, and the reporting disagreement mostly disappears on its own, because there was only ever one gap to begin with - it just showed up twice.

If you want to talk it through, book time with the team

This is one of the most common patterns we see in brands scaling past the point where informal systems should still be working. It’s rarely framed as a risk internally, because the person carrying it is usually excellent, well-liked, and at the same time entirely too busy to document how any of it actually works. It’s come up in conversation after conversation recently, with different brands, in almost identical terms.

Efficient, right up until it isn’t

A system built and maintained by one person is, for a while, the fastest possible way to get things done. No committee, no sign-off, no waiting for a change request to clear a queue. That speed is real, and it’s exactly why nobody wants to slow it down by asking someone to document it properly.

The cost of that speed shows up later, and it shows up all at once. A holiday, a resignation, an unexpected absence, and the business discovers that an entire operational process lived in one person’s head, with the system itself as a set of clues rather than a set of instructions. The automation everyone assumes is running quietly in the background, except it doesn’t run when the one person who checks it isn’t there to check it. The workflow that grinds to a halt because it only ever ran from one laptop, and that laptop is currently at the airport.

Ownership isn’t the same as control

The brands who get this right aren’t the ones who take the system away from the person who built it. They’re the ones who separate “this person built it and understands it best” from “this person is the only route through it.” Plenty of brands are explicit about wanting exactly this: they don’t want to depend on an external partner, or a single internal specialist, for every change. They want the tools they use to actually belong to the team running the business day to day.

That’s a different ask to “hire more people.” It’s an ask about whether the knowledge is written down, whether more than one person could make a routine change without asking permission, and whether the system would survive its architect leaving on good terms tomorrow.

The audit that actually matters

A useful exercise, and an uncomfortable one: pick the three systems your operation would struggle most without, and ask how many people could rebuild each one from documentation alone if the original builder vanished tomorrow. If the honest answer is one, or none, you’ve found where the real risk sits - not in the system itself, but in the fact that it only exists properly in someone’s head.

This isn’t a case for slowing everyone down with process for its own sake. It’s a case for writing down the two or three things that would actually hurt if they disappeared with a person, rather than everything, because trying to document everything is how documentation projects die. Do the exercise this week. Pick the three systems. Ask the question. Whatever the answer is, it’s more useful than not knowing.

The system usually isn’t the fragile part. The single point of failure almost always has a holiday calendar, and no idea they’ve become one.

This is something we spoke about in detail at OX2 London this July. We’ll no doubt talk about it again at OX3 Stockholm on 22nd October. If you want to be involved in those conversations, apply for a spot below.

https://operatorexperience.ai/

That might look like doing ourselves out of work. It’s the only honest way we know how to advise a client. When it’s not the right time for an ERP, it’s not the right time, whatever that means for us commercially - and it’s a far more useful outcome than most ERP projects produce, because most ERP projects start from “we need a new system” rather than “what problem are we actually trying to solve,” and the two questions lead to very different answers.

Sometimes twenty people need a process, not a platform

At a brand of around twenty people, the more honest fix wasn’t a new system at all. It was clearer process and better-defined roles. A platform can’t fix a business that hasn’t yet agreed who owns what. Buying one at that size usually just moves the disagreement into a more expensive piece of software.

This is the sequencing question in miniature: process and people first, system second, and only once the first two have actually settled down enough that a system would be configuring something real rather than something still being argued about.

We don’t sell a system. We sell the sequencing.

We’re vendor-neutral in the sense that we don’t push, sell or promote a specific product or agenda. What we’ve got instead is pattern-matching across enough different brands’ systems to know, with some precision, what tends to work, when, and why. That means we can be specific: this system, at this point, for this reason, and here’s what needs to change first and what shouldn’t be touched at all. It’s tailored to the individual business, not to a template.

We’ve written before about the tier question - which rung of the ladder actually fits a brand at a given size, and the specific graduation signals that tell you it’s time to move up one. This is the layer underneath that conversation, not a replacement for it. Tier selection assumes the timing question has already been answered. Most brands haven’t answered it - they’ve just skipped straight to comparing vendors.

It’s almost always people and process, not technology

Across enough brands now, the pattern repeats: the real gap is people and process, not the system. That’s where we focus our own time and resource, even when it’s less exciting to sell than a shiny new platform, because bringing in new technology without fixing the underlying process is how adoption falls apart. Teams don’t use what they don’t trust, and they don’t trust a system that’s just automating a broken handoff faster.

Part of doing this properly is being honest about where roles aren’t clearly defined, where a business might need someone new to actually own a function rather than share it informally, and where a team is already stretched thin enough that a new system will add workload rather than remove it, at least at first. None of that’s comfortable to say out loud in a sales conversation. It’s necessary to say it anyway.

Deploy with real data before you commit

Where a system decision is genuinely due, the useful move isn’t a slide deck of vendor comparisons. It’s someone standing up a working version with the brand’s own data and demonstrating it back to them, before anyone signs anything.

We try to show things practically wherever we can, not just describe them, and bring people along through the process rather than present them with a decision at the end of it. That single step removes most of the scope creep that comes from making a big decision on assumptions rather than evidence.

These are frequently six-figure decisions. Getting one wrong is considerably more expensive than getting it done slowly and correctly, which is why we’d rather a client have realistic expectations going in than a fast yes that unravels eighteen months later.

The sequencing question is the real decision

ERP versus PLM versus middleware versus “fix the process first” isn’t a menu you pick one item from. It’s a sequence, and getting the order wrong is the single most expensive mistake available. Systems bought to solve a problem that was actually about ownership, definitions, or process discipline don’t solve it. They just give the same disagreement a bigger, more expensive home.

The question worth asking before any ERP conversation starts isn’t which platform. It’s whether a platform is actually what’s missing, or whether that’s just the easiest thing to buy when the real answer is harder to agree on.

Start the conversation by contacting us.

That’s something worth talking about. The symptom always turns up in the same place - the inbox - while the cause is almost always sitting further upstream, in a system or a process that nobody’s watching closely enough to catch before a customer does. Or something that hasn’t been cleanly mapped out, so no-one truly understands. Or has the time to fix.

CS is usually downstream of something else

At record store Rough Trade, the volume of “where is my order” queries in the inbox was a genuine drag on the team, until they built an inbox classifier that now resolves around two-thirds of enquiries automatically by checking order status directly. This was work we wrote about in more detail after OX2, alongside the returns platform they purpose-built.

The point isn’t the specific numbers, though. It’s the way they fixed it. They treated a CS problem as a systems and process problem first, and the CS burden dropped as a result, rather than the other way round.

Different teams catch the same gap at different times

At Bluebella, discrepancies between the storefront, the ERP system and the warehouse used to fall through the cracks and get caught reactively by whichever team happened to notice first. They built a daily dashboard that flags the mismatch before a colleague or customer has to notice and complain about it. And showcased it at OX2.

That’s the wider pattern across pretty much every brand running more than one system for order fulfilment: the gap gets found at the most inconvenient possible moment, for whichever team is closest to it that day, and it’s rarely the same team twice. One week it’s warehouse ops mid-pick. The next it’s CS mid-inbox. Eventually it’s a brand or PR issue, because enough customers hit the same gap at once.

If someone’s genuinely responsible for watching the systems upstream, most of this never reaches that point at all - and it’s easier to build that watching function now than it’s ever been.

What actually reduces WISMO volume

Not a better help centre article, for a start. What actually moves the number is closing the gap between what the customer sees and what’s true in the warehouse, before they have to ask.

That looks like proactive status updates instead of reactive lookups. Order tracking that actually shows a customer where their order is, in real time, rather than a static “processing” screen or an “order is in progress” that means nothing.

And if they do reach out, customer service systems should be properly integrated with order and fulfilment data, so customers get a real answer immediately that recognises who they are and what they ordered, and the customer service team are enabled to do their jobs properly without chasing down the ops and logistics team.

Importantly, WISMOs should not turn into a request for the customer to jump through hoops or a queue with a two-day SLA before a human even looks at it. That’s not the experience anyone’s aiming for, and it’s entirely avoidable with the right plumbing upstream.

The real cost of the tax

Every hour a CS team spends manually chasing a warehouse booking or explaining a sync error is an hour not spent on the customer interactions that actually need judgement. That’s the WISMO tax - not the query itself, but the fact that a systems problem gets paid for, every single day, out of a team’s capacity to do the part of the job that matters.

Fix the upstream reconciliation, and the WISMO volume mostly takes care of itself. Don’t just add headcount to answer the same broken question faster.

If this resonates, commercethinking.com

Here’s the workflow. A retail partner with a return agreement decides to send stock back. Sometimes an email arrives first, vague about contents. Often it doesn’t. What arrives is a box. Sometimes 400 pieces in it. No usable manifest, no link to an order, no warning to the warehouse.

The brand can’t raise a normal purchase order because they don’t know what’s in the box, and experience says it won’t match whatever the email claimed anyway. So they raise a dummy purchase order. One line on it. The warehouse receives the box against the dummy, fills it up with whatever physically arrived, and sends the completed fiction back to head office. Only then does anyone know what came home. A credit note gets drafted, checked against what the retailer claimed, and the two rarely agree.

And then, mostly, the brand just accepts it. Disputing a discrepancy on a seventy-dollar wholesale piece costs more in admin than the credit is worth.

The retailer knows this.

“They’re trying to save money, we’re trying to save money, but they’re bigger than us. They get the efficiency, we get the pain.”

Why this is a process smell, not a software gap

The instinct is to blame tooling, and it’s true the software market has ignored this. Five years of investment went into consumer returns portals while B2B returns, the majority of revenue at brands like this one, still runs on boxes and guesswork.

But before you go shopping, look at the process, because most of this is fixable inside the ERP and warehouse setup you already have. The dummy PO exists because the process was designed around a document (the purchase order) instead of an event (stock is coming back). Flip that and the workarounds start disappearing.

What good looks like.

RMA-first receiving.

Every inbound return gets a returns authorisation record the moment the retailer signals intent, however vaguely. The RMA is allowed to be wrong. Its job is to exist, so the warehouse receives against a real object instead of a fictional PO, and so head office has a record that predates the box.

Blind receipt as a designed workflow, not a hack.

Most WMSs support blind receiving properly. Receive what’s physically there, line by line, against the RMA, and let the system generate the discrepancy report automatically. That’s the same work the warehouse does today filling up the dummy, except the output is structured data instead of a returned fiction.

Probabilistic invoice matching.

The hard problem in wholesale returns is attribution. Stock comes back at season end, six or nine months after shipping, and the same SKU has shipped to the same retailer many times in between.

The ops lead’s image was perfect: “You shoot the beast, and it comes back nine months later, and you could have shot the same beast ten times.”

Deterministic matching is impossible, so stop demanding it. A good guess, most recent shipment of that SKU to that partner, FIFO across open invoices, whatever rule finance can live with, beats no match. Configure the rule once, apply it consistently, and reserve human attention for the exceptions.

Codify the economics of surrender.

If you already auto-accept every discrepancy below a certain value because disputing costs more, that’s not a failure, that’s a policy. Write it down, put the threshold in the system, and stop paying people to perform a dispute process whose outcome is predetermined. The dishonest version of this is what most brands do now: pretend to check everything, actually check nothing.

Triage sellable stock at receipt.

Most of what comes back in wholesale has nothing wrong with it. It’s sellable product that sat on a shop floor too long, and by the time it’s travelled back it’s aged into dead stock. The receiving workflow should route it to its next best destination, outlet, partner, clearance channel, at the moment it’s scanned, not after a quarterly stock review finds it.

The uncomfortable bit

None of this needs new software categories. It needs someone to admit the current process was never designed, it accreted, and to spend a few unglamorous weeks in the ERP and WMS configuration making the return an event the system expects rather than an ambush it survives.

The consumer side of returns got beautiful because it was customer-facing and venture-funded. The wholesale side stayed ugly because it’s internal and nobody’s demo ever shows it. If your process involves a dummy anything, that’s the tell. The system is telling you it was never asked to handle reality, and reality showed up in a box with 400 pieces in it.

If any part of this is running inside your business right now, dummy POs, unmatched credit notes, a returns process nobody designed, get in touch. This is the kind of ERP and WMS reconfiguration we do.

Stop feeling smug. The same thing is happening in your automation stack, and it’s further along than most operations leaders want to admit.

Here’s the pattern, and see if it sounds familiar. A problem appears. Failed orders, mistagged returns, a courier exception, whatever. Someone builds a workflow to catch it. The workflow works. This is genuinely good, and it happens again, and again, in the returns platform, in the flow builder, in the middleware, in the email tool, in a few Zaps somebody set up in 2023 and left the company without documenting.

Three years later the brand owns dozens of automations, sometimes hundreds. Each one made sense when it was built. Together they form a web nobody fully understands, where changing one rule risks breaking three others, so nobody changes anything, so the automations drift further from how the business actually works, so more automations get built to patch the drift.

That’s debt. It has all the properties of debt. It accrued from decisions that were individually reasonable. It charges interest, in this case the growing gap between what your automations assume and what your operation does. And the payments are invisible right up until they’re not, usually during peak, when the web does something nobody predicted and nobody can trace.

One conversation we had recently put a sharper edge on it. The observation was that less-automated brands are, in one specific way, better positioned right now than the sophisticated ones. They haven’t got the tangled web. When they adopt new tooling, including AI tooling, they’re building on a clear site. The mature brands have to untangle first, and untangling a production automation web while it’s running is the operational equivalent of rewiring a house with the electricity on.

How to tell if you have it

A few honest questions. Can anyone in the business produce a list of every automation currently running, across every tool? Not a partial list. All of them. If the answer is no, you have automation debt by definition, because you’re paying interest on liabilities you can’t enumerate.

Second question. When did you last delete an automation? Teams add workflows constantly and retire them almost never, because deleting feels risky and nobody owns the decision. A stack that only grows is a stack that’s rotting.

Third. When something weird happens in an order flow, how long does diagnosis take, and how much of that time is spent working out which automation touched it? If the answer involves opening four different tools and asking someone who left, you already know.

Paying it down

The fix is not a tool. Resist the vendor who says otherwise. The fix is treating automations the way good engineering teams treat code.

Start with an inventory. Every workflow, every tool, one register, with an owner and a one-line statement of what business outcome it serves. This is boring and it takes a fortnight and it’s the single highest-value exercise available to most ops teams this quarter. You will find automations nobody can explain. Those go on the kill list.

Then kill things. Set a rule: for every workflow added, one gets reviewed for retirement. The point isn’t the ratio, it’s making deletion a normal act instead of a scary one.

Then raise the bar for new ones. “Add a workflow” should face the same scrutiny as “add a system,” because that’s what it is, a small system, with maintenance costs and failure modes and an integration surface. The question to ask isn’t “can we automate this” but “is this a downstream patch for an upstream problem we should fix at source.” A remarkable share of automations exist to tidy up bad data that a mandatory field would have prevented. Fix the field, delete the workflow.

Where AI actually fits

AI is being sold as the answer here, and it’s half true. Used to understand the web, mapping what exists, tracing what touched an order, explaining what a five-year-old workflow actually does, it’s genuinely powerful, and it’s the best untangling tool available right now.

Used the other way, it’s an accelerant. AI makes building new automations nearly free, which means a team with no discipline can now generate strands for the web faster than any human ever could. Cheap creation plus expensive maintenance is exactly the equation that produced tech debt in software. We’re about to run it again in operations faster than more teams are ready for.

The brands that come out of this well won’t be the ones with the most automations or the fewest. They’ll be the ones who can produce the register, name the owner of each strand, and delete something without holding their breath. Everyone else is still rewiring the house with the power on.

If nobody in your business can produce the full list of automations running right now, book a call with us. The inventory is the fortnight of work we help brands get through.

That order is not why you’re buying a WMS. That order works everywhere, including in a spreadsheet. You’re buying a WMS for the orders that don’t look like that, and the demo will never show you those, because that’s where the systems actually differ and where some of them fall over. Loudly, at peak, in production.

We’ve been collecting real examples from operators recently, and they make the point better than any analysis could.

Example one. A customer buys a gift card and a hoodie in the same basket. The gift card fulfils instantly, it’s digital. The hoodie needs picking. The order is now partially fulfilled, and this brand’s WMS won’t push partially fulfilled orders through to the pick queue. A completely ordinary basket, wedged in the pipe, waiting for a human to notice and unjam it. The brand has asked the vendor to fix it. The vendor hasn’t.

Example two. Same brand, different edge. Their WMS integrates with their storefront but doesn’t pull stock from it. Their returns platform does pull stock from the storefront. The WMS and the returns platform don’t talk to each other at all. Three systems holding the same stock number, two integrations, a triangle with a missing side. Everything works until an item sells out, at which point there’s a window of a few minutes where a second customer can buy something that no longer exists. The failure only appears at zero stock, which is precisely when accuracy matters most, and precisely the scenario no demo ever runs.

Example three. A brand at serious volume whose returns platform requires every accepted return to be processed again, manually, in the storefront. Every return, touched twice, by a human. At their order count that isn’t a workflow, it’s an error generator with a schedule.

None of these brands bought badly. Each tool was chosen sensibly, on its merits, often on a recommendation. The problem is that “on its merits” meant the happy path.

The checklist

So here’s what we’d put in front of any vendor before signing, and we’d insist on seeing each one demonstrated live, in their system, not answered verbally. Verbal answers to edge-case questions are always yes.

On fulfilment: show me a partially fulfilled order moving through the system. Digital plus physical in one basket. A pre-order line alongside an in-stock line. A split shipment. What does the pick queue do with each?

On stock: which system is the source of truth for inventory, yours or mine, and what happens in the minutes after an item hits zero? Show me the sequence of syncs. Who learns first, who learns last, and how long is the gap?

On returns: when a return is accepted in the returns platform, walk me through every system that has to be updated and tell me which updates are automatic. If the answer includes “and then you just go into the storefront and...”, that word “just” is going to cost you a headcount.

On exchanges: show me an exchange at a different price point. Higher and lower. Where does the price difference live, who issues it, and does the stock adjust in both directions?

On exceptions: show me what happens when your system receives an order with a missing or malformed field. Does it reject loudly at the source, or accept silently and fail three systems later?

And one meta-question that tells you more than all the others: describe the last edge-case fix you shipped because a customer asked for it, and how long it took. Vendors who can answer that specifically have a working escalation path. Vendors who go vague have a roadmap and a prayer.

Why this matters more now

Edge cases used to be an annoyance. They’re becoming the whole game, because everything brands want to do next, automation, AI on top of operational data, real-time promises to customers, assumes the underlying systems agree with each other in the weird moments. A stack that’s only coherent on the happy path can’t support any of it. You’ll be building intelligence on top of a foundation that tells three different stories whenever anything interesting happens.

The demo is a film. The edge cases are the documentary. Buy on the documentary.

If you're mid-vendor-selection and want a second pair of eyes on the edge cases before you sign anything, get in touch.

A lot of people eye roll, justifiably in a lot of instances, and think “not this s**t again”.

Some start thinking about margin, speed, and getting more done with the same team.

The others wonder whether this is how their job starts disappearing. And it’s this third group I want to talk to you about here.

If we look past the eye rolling cohort and assume that when you bring up “AI” it’s thought out and not just bulls**t.

The group who are gripped by something akin to fear and insecurity. You have a real problem, and so do they.

As a business, doing nothing on “AI” isn’t an option, because it’s a fair bet someone in your category is already automating processes, improving margins, or making decisions faster while you’re still figuring out where to begin.

Similarly, as an individual you can’t do nothing. Because organisations are changing, jobs are changing, and roles are / will change. This means individuals must develop new skills, and embrace change or risk losing their jobs.

So as a leader you’ve got to do two things at once. You can’t do nothing, and you can’t disregard your teams legitimate and valid fears out of hand.

Name the fear

Start by saying what everyone already knows. The tasks AI automates best sit between departments in the spreadsheets, manual data duplication, and endless trackers that exist.

This kind of work sits disproportionately in entry-level and coordinator roles. So, in a future where AI is embedded in a meaningful way it’s likely there’s fewer entry grade roles.

Weekly trade packs, routine reporting, first-pass copy.

That’s why the anxiety shows up hard in juniors, and why it snowballs into everyone else’s mood.

Under all the operator chat is the same question about whether this replaces me or my team. That question doesn’t get asked out loud very often, and it should.

If you don’t name it, people fill in the blanks for you. And that’s when you’ll experience lag, and rejection of change.

Then say the other bit leaders avoid. If we don’t adopt this stuff, we don’t stay static and safe, we get slower and more expensive than the brands that do. Which puts us at even greater threat.

It’s another way of saying - sooner or later, we’re f**ked if we don’t embrace this.

Repeat the rule

Give the team a simple rule and repeat it until it’s boring. Preserve headcount, multiply capacity.

That line matters because it forces you to pick an intent. The aim isn’t to rip out teams in favour of agents, it’s to go after the most painful bottlenecks and the most tedious work.

Don’t over-promise while you’re trying to reassure people. Preserving headcount doesn’t mean no role ever changes; it means you take responsibility for redesigning roles instead of using AI as a handy excuse for a redundancy round (see the countless tech bros fluffing up their share prices using this tactic).

Here’s how it shows up in plain English. We’re going to replace the dull and repetitive parts of the work. That work will be replaced. You won’t.

We’re going to invest in training so you can run the tools, judge the output, and know when not to use them. We’ll also be honest about the learning curve, because not everything we do will work first time (or ever).

Make it real in the workflow

Most AI chat inside brands gets stuck on the big existential stuff. Most operator reality is much less sexy and much more useful.

Where in my week is the highest-leverage thing AI can take off my plate?

You don’t need a big bang programme. You need momentum.

Start by taking reporting and reconciliation off someone’s plate, then focus them on increasing its quality and depth of analysis. Pick a workflow where boring, repeatable work is absorbing a big % of a person’s week, and map it end to end.

Agree what good looks like, what gets delegated to AI, and what stays with a human because it needs judgement.

If you’re trying to make preserve headcount, multiply capacity stick, you need a plan for the freed time before you free it. Talk in outputs and bottlenecks removed, not fluff.

If your merch team spends days building weekly trade packs, get the first cut done faster, then use the time to interrogate range performance properly. If finance is drowning in reconciliation, let AI take the first pass, then spend time fixing root causes instead of just clearing the queue.

If customer service is rammed with repetitive inbound, let automation handle the simple stuff, then redeploy the humans into the messy cases that actually lose you customers.

Don’t sell magic, sell standards

People will swear AI has made everything cheaper, then one failure and the whole org starts double-checking everything again.

That’s the paradox. Automations promise efficiency, but undetected failures create redundancy and duplication, and instead of freeing up capacity they exacerbate it.

AI is also in the mix before you’ve formalised it, take a walk through any office and you’ll see Claude or ChatGPT open on most desktops. It’s just used inconsistently and sometimes invisibly.

The risk isn’t that AI was used. It’s that it was used without accountability, by someone who couldn’t verify what they were looking at.

So set the boring rules early. Workflows, checks, clear lines on where data goes, and who signs off.

If you want people on side, don’t pretend quality control is optional. Tell people you expect them to use the tools, and you expect them to own the output.

Train the humans

A lot of leaders get stuck because they think the organisation needs a perfect AI strategy deck before it can act. In reality, most teams need to identify specific tasks worth automating and start building capability around those.

If you want a useful target that doesn’t spook everyone out, make it about tasks, not jobs.

If you say preserve headcount, you’ve also got to say we’ll help you get good at this. It’s a new muscle, and it won’t build itself.

And if you need help, for the love of god join Operator Experience and get multiple members of your team to the next events.

Focus on Catalysts

If preserve headcount, multiply capacity is the intent, Catalysts are how you make it real at pace.

The framework I use maps people on two axes, AI clarity and enablement ability. Four profiles emerge.

The Catalyst is high clarity, high enablement. They see where AI fits, build something useful, and the team can actually run with it.

This matters because most org charts aren’t built for AI work to spread safely. You end up with a few good workflows trapped in someone’s head, and everyone else either ignores them or copies them badly.

Catalysts fix that, because they treat AI like an operating model shift, not a tool rollout. They’re the people who can turn a single win into a repeatable way of working, which is why Catalysts are going to be the scarce resource every brand is competing for.

You’ll spot them in how they talk. They don’t say we should use AI more; they say this is the bit of the trade pack that’s a time sink, this is the check we can’t drop, and this is how we’ll run it week to week.

This is where a lot of brands mess it up. They hire for build capability while ignoring enablement, and end up with a couple of clever workflows that nobody trusts and nobody learns from.

And watch the other failure mode as well. A lone wolf who can’t bring anyone with them isn’t a multiplier, they’re a bottleneck.

So if you’re trying to lower anxiety and raise adoption at the same time, don’t just go hunting for lone wolves. Build Catalysts, because they’re the people who make AI feel safer and more useful for everyone.

• Give them a remit tied to real work, not AI theatre.

• Give them time and permission to teach, document, and standardise.

• Reward the people who make others better, not just the ones who ship the shiny thing.

And importantly, this profile is more learnable than people think. The rare Catalysts are more likely built than hired, because it’s a combination of understanding the work and bringing others along.

One last thing

Preserve headcount, multiply capacity is a sane middle path. It lets you move at the pace the market is forcing on you, without fracturing your business.

Have the conversation early, say the scary bits out loud, and then prove the intent in the day-to-day. People don’t need you to promise a risk-free future. They need you to show you’ve got a plan, and that they’re in it (if they embrace the change).

If you're heading into this conversation with your team and want a second pair of eyes on how to frame it, have that conversation with Luke first.

What happens next usually goes one of two ways, and both of them are wrong.

Way one: panic-buy enterprise. The brand decides that outgrowing the small system means it’s time for the big system, engages an implementation partner, and spends eighteen months and a serious six figures installing an ERP built for a business three times their size. The implementation eats the ops team’s year. Half the modules never get configured. The brand is now paying enterprise maintenance on entry-level usage.

Way two: denial. The brand decides the small system is fine actually, adds another app, another spreadsheet, another integration, and defers the decision until the month finance stops trusting the numbers, which is the most expensive possible moment to start a systems project.

We want to argue for the boring middle, and specifically to say something consultants aren’t supposed to say: the lighter-touch ERP is underrated, and the expensive mistake in this market is almost never buying too small. It’s buying enterprise for a business you haven’t built yet.

The evidence for the light system

Here’s a story from a recent conversation that should be told more often. A UK brand localising its US operation, the classic complexity jump, inventory in two territories for the first time. One operator, essentially working alone, ran the whole thing through a lighter-touch ERP that the industry habitually describes as a stepping stone. Full range, live operation, no enterprise anything. The brand is now growing strongly in the US on the back of it.

The lesson isn’t that small systems are secretly enterprise-grade. They’re not. The lesson is sequencing. This operator built the operation first, matched the system to its actual current shape, and planned to add complexity when volume demanded it rather than in anticipation. The brands that struggle run it backwards: they buy for the business plan instead of the business, and the system’s demands then compete with the growth it was supposed to enable.

A light ERP implemented well beats a heavy ERP implemented late, every time, and it usually beats a heavy ERP implemented on schedule too, because the schedule is fictional.

So when do you actually graduate?

This is the useful question, because “we’ve outgrown it” is a feeling, and feelings are how brands end up in way one. The graduation signals are more specific than a feeling, and most of them are visible a year before the crisis.

Multi-territory inventory is the big one. The moment stock lives in two places with different demand curves, entry-level tools start creaking, because they were built around one pool of stock and one version of the truth. If international localisation is genuinely on the eighteen-month horizon, the ERP conversation starts now, calmly, not later, urgently.

B2B and B2C on one back end is another. Wholesale brings return agreements, bulk credit notes, partner-specific pricing and paperwork the light tools handle grudgingly or not at all. If wholesale is heading past a quarter of revenue, you’re graduating.

The third signal is the finance one, and it’s the deadline rather than the warning: the month-end close starts requiring manual reconciliation between systems, and the FD stops trusting the numbers without checking them somewhere else. When finance builds a shadow spreadsheet, the system has already failed. Everything after that is overdue.

And a non-signal, worth naming because it triggers so many premature migrations: pain that’s actually data governance wearing a systems costume. A remarkable amount of “we’ve outgrown the ERP” turns out to be ungoverned fields, duplicate SKUs and integration gaps that would follow the brand into any system on earth. Fix the data first. You’ll often find the appetite for replatforming shrinks, and if it doesn’t, at least the migration won’t import the mess.

The ladder, not the leap

Think of it as a ladder with known rungs. Spreadsheets, then entry-level inventory plus cloud accounting, then the light ERP, then mid-market, then enterprise, and the skill isn’t picking the right rung so much as knowing your climbing speed. Each move costs a project, so you want to move when the signals fire, not when the vendor’s sales team does.

The brands that get this right are unglamorous about it. They’re on a smaller system than their revenue suggests, it’s configured properly, the data in it is clean, and they know exactly which signal will trigger the next move. The brands that get it wrong have the biggest system money can buy and a finance team running the business from Excel next to it, expensively.

If any of this rings true to where you are right now, get in touch, and we can talk it through.

That the agentic web was taking shape, and that brands with inconsistent, shallow product data would be invisible to it. That the work sitting inside your PIM and ERP was about to matter in a way it never had before.

That was a prediction about where things were going. It is not a prediction any more. That argument has not aged. It has arrived.

What has changed in twelve months?

A year ago, the agentic web was infrastructure being built by Microsoft and a handful of early adopters. Today it is a product decision Shopify has made at the platform level.

Shopify Catalog, which launched earlier this year, automatically standardises and enriches product data so AI models can reason about it. Shopify claims data it syndicates drives twice the conversion in AI shopping interactions. The unit that matters in agentic commerce, they have decided, is the product. Not the brand, not the campaign, but the product, and the data behind it.

MCP, which we flagged as the plumbing between AI interfaces and your actual systems, is no longer theoretical. Brands are rolling it out now. The journey from a messy reporting setup to a data layer with a natural-language tool on top, where any team member can query and analyse for themselves, is a project operators are completing in months rather than years.

The early adopters are no longer Tripadvisor and O’Reilly. They are brands in the same category and size range as yours.

The problem nobody expected AI to solve first

The assumption was that the agentic web would expose product data problems externally. That AI agents shopping on a customer’s behalf would simply skip brands whose data was too inconsistent to reason about. That the pressure would come from outside.

The more immediate pressure has turned out to be internal.

Brands pointing AI tools at their own operations are discovering exactly how inconsistent their product data is, faster and more visibly than ever before.

One product tagged as boots, another as ankle boots and another as footwear. Sizes in one format in one category, a different format in another. Colour buried in a title rather than sitting as a clean attribute.

We’ve seen it on audit after audit this year. One product tagged as boots, another as ankle boots, another as footwear - same product, three vocabularies. Sizes in UK numeric in one category, S/M/L in another. Colour sitting inside a title string instead of living as its own attribute, so nothing downstream can filter, compare or reason on it.

None of this is new. What’s changed is that the moment you point an AI tool at it, every inconsistency surfaces in front of a decision someone needs to make now - not buried in a data quality report nobody reads, but in the actual output the business is trying to act on.

These were always problems. AI has made them impossible to ignore because it surfaces every contradiction at speed, in front of a decision that needs making now.

The mess was always there. The tools have just made it visible.

The bit nobody is selling a tool for

Shopify Catalog enriches what you feed it. It does not fix what you feed it with. MCP exposes your data layer to the systems querying it. It does not make that data layer coherent if it was not already.

We’d say roughly a third of the problem that tooling solves is the visible part. The rest is whether the brand can describe its own products the same way everywhere they live, and whether the operating model can keep them that way as the business changes.

That’s an ownership question. Who is accountable for product data quality across the business? Which system wins when three systems disagree? These are the questions that determine whether the tooling pays off, and they have no vendor category.

What to actually do about it now

Get your product data consistent across every system that touches it before you point an AI tool at any of them.

That means naming a source of truth for every product attribute and writing down which system wins when systems disagree. It means deciding who owns product data quality across the business, not who manages the PIM. And it means doing that work before the AI surfaces the inconsistency in a context where it costs you.

The brands that will perform well in AI-driven commerce are not the ones with the boldest agentic strategies. They are the ones that had something coherent to syndicate when the tools arrived.

A year ago, that was a prediction. Now it is just what is happening.

This is exactly the kind of gap our product ops work is built to find - where your data breaks down, which system should be the source of truth when they disagree, and what it’d take to get there.

If you want to talk through where your own catalogue would trip up an AI agent today, grab time with Luke - worth having that conversation before the tooling forces it.

Argentina's official store is built and run by Fanatics. So is England's. englandstore.com looks like the England store, sounds like the England store, and is run start to finish by Fanatics. Not just the shipping. The range, the buying, the stock, the platform, the site, the lot. The federation takes a royalty and stays out of it.

It isn't a national-team quirk either. France, Germany, Italy, Belgium and Argentina all run their federation stores through Fanatics, and so does UEFA for the Euros. Same with PSG. Same with Juventus, Inter and AC Milan. Same with Wembley's retail. Same with a quarter of the clubs in last season's Premier League: Chelsea, Aston Villa, Everton, Leeds and Sunderland. Chelsea have gone further and handed over design and manufacturing too.

That’s a problem for us, because we’ve spent seven articles arguing they’ve all got it wrong.

The awkward bit

If you followed West Park Rovers, you know the argument. Bring it in-house. Own the stack, own the relationship, stop letting a third party sit between you and the people who love your club. We showed the operational reality, the expiring catalogue, the letters and numbers nobody counts, and we finished on the fan graph, which is the whole reason to do any of it.

We stand by all of it. But look at who’s actually renting.

These aren’t clubs that couldn’t afford the alternative. PSG could buy a warehouse before lunch. The FA is not short of a few quid. These are some of the largest, richest, most commercially sophisticated sports organisations on earth, and they have all looked at running their own commerce operation and said no thanks.

Two different things are going on there, and only one of them is defensible.

Subscribe now

You can’t grow into a World Cup

Start with the defensible one.

The barrier to running your own store was never the technology. Shopify solved the website years ago. Anyone can be selling by Friday. The hard part is the operational and merchandising muscle, and the only way to build that is slowly, on small numbers, while nobody’s watching.

Every DTC brand you admire got that runway for free. They were rubbish at forecasting when they were doing forty orders a week, and it didn’t matter, because forty orders a week is a survivable mistake. By the time they were doing forty thousand, they’d made every error once already.

A national team never gets the practice, it’s already fully grown. You cannot put up a website with three shirts on it and say we’re learning, we’ll get there. You need millions of pounds of stock sat in a building before you take a single order, a 3PL that already works, a personalisation line that already works, and a merchandising team that already knows what a good buy looks like, on day one, with the whole world watching.

And a national team's commercial operation matters for six weeks every four years.

The apocalypse, but faster

We wrote a piece called The 12-Month Apocalypse about the thing that makes sports retail structurally horrible: your entire catalogue expires every year. New sponsor, new squad, new competition, and last season’s stock isn’t discounted, it’s dead.

A tournament catalogue doesn't expire in twelve months. It expires the moment you go out. England found that out in the semi-final. Argentina found it out on Sunday. The shirt that was the most wanted garment in the country on Saturday is dead stock by Monday, and no supply chain on earth can turn a run of replicas around fast enough to catch the one week that mattered. Win the whole thing and you get the single exception: Spain's shirt is the only one still worth printing. Everyone else is holding polyester with the wrong names on the back.

Moments, Not Seasons argued that clubs should build the muscle to respond to the unpredictable, because seasonal retail has a calendar and sports retail has a pulse. Fine. But a tournament isn’t a pulse. It’s one beat. No signal you can act on, no second attempt, and the thing you’re forecasting is a penalty shootout.

You can’t build a capability around an event that happens once every four years and lasts a fortnight. You’d be paying to keep a team sharp for 206 weeks of not much so they’re ready for the two that count. Nobody sane does that. You rent it, and you’re right to.

So England gets a pass. Fine.

Subscribe now

Chelsea doesn’t

Chelsea play fifty-odd times a year. So do Everton, Leeds and Sunderland. Every year. Forever. That is about as repeatable as commerce gets, and repeatability is the whole argument. When demand comes back next week, every mistake is tuition. Every spike teaches you something about the next one. The muscle you couldn’t build for a tournament, you can absolutely build across a decade of Saturdays.

So we’re not softening the position. In-housing is better long-term. Own the stack, own the relationship, own the data. Seven articles, no change of mind.

The reason it isn’t happening is that nobody at the club is listening, because nobody at the club has ever been asked to.

A football club is judged on one thing, and it isn’t gross margin on a hoodie. Retail is a royalty line that turns up in a spreadsheet once a quarter and never causes an argument. It isn’t underperforming. It’s just not anyone’s job to notice. And the licensing partner is very good at making sure it never becomes anyone’s job, because the moment it does, the deal’s in trouble.

That’s not a strategy. It’s an absence of one, held in place for twenty years by the fact that nothing ever went visibly wrong.

This isn’t a football problem

Every business has an England store.

Some part of the operation that somebody else runs, not because you sat down and did the maths and outsourcing won, but because it ended up that way early on and nobody has reopened it since. The wholesale portal. Marketplace listings. Returns. The B2B side. The international sites you launched in a hurry in 2021.

It’s rarely the bit that’s broken. Broken gets attention. It’s the bit that works fine, earns less than it should, and is run by somebody whose incentives are not yours.

The tell isn’t in the P&L. It’s that you can’t answer a simple question about it without emailing someone and waiting two days. How many of those customers came back? What drove the spike in March? Nobody in your building knows. Somebody in a building you don’t own does, and they’ll tell you next quarter.

The other tell is the terms. The deal was signed when that channel was the ninth biggest thing you did. It’s now the third. The terms haven’t moved, because nothing’s gone wrong, and nothing going wrong is exactly why nobody’s looked.

Not everything outsourced should come back in-house. England’s genuinely shouldn’t. But you should at least be able to say which category yours is in, and whether you’re renting because the maths says rent or because the question has simply never been raised in a room where anyone could act on it.

Most brands can’t answer that. That’s the same failure as the club’s, just with less bunting.

One thing you should never rent

You can outsource the factory. Sometimes you should.

But the last piece in the series was about the fan graph, and that’s where this stops being sensible.

A World Cup generates the most valuable fan dataset a federation will see in a decade. Every shirt sold over those six weeks is a person telling you something. Who they are. What made them act. Whether they came back after 2022 or turned up new last month. For most of the teams in this tournament, all of it was captured inside somebody else's platform.

The stock will be gone by August either way. The names won’t be worth anything by then. But the graph would have been worth something for twenty years.

Rent the operation if the maths says rent. Just don’t hand over the only bit that appreciates.

What OX is trying to be

We opened with the premise. OX exists to help brand leaders and operators navigate the AI era together, collaboratively, rather than each business figuring it out alone in a tab full of vendor webinars. The event is free for brands, and deliberately so: no cold sponsorship outreach, every partner in the room is someone the founders have worked with personally, in some cases for fifteen years plus.

The framing slide is worth passing on because it’s a useful mirror. Inside every brand there are passengers, waiting for someone else to lead on AI. Lone wolves, independently capable but building alone. Enthusiasts, great at bringing people with them but fuzzy on the tech, often at exec level. And catalysts, the rare combination of high AI clarity and high enablement of others. OX’s stated mission is to turn the people in the room into catalysts and help brands identify and grow their own. It’s a generalisation, and it’s also the most useful two-by-two we’ve seen on this topic.

The other structural note: OX1’s feedback was unanimous that the brand show-and-tells were the point, so OX2 handed the entire morning to three brands showing what they’ve actually built. Not slideware. The real things, running.

Subscribe now

The morning: three brands, three builds

Paul Heath, Adanola

He opened with the most complete data story of the day. Seven months ago he joined as a data leader the day before the only data engineer left, inheriting a two-year-old Power BI and Azure estate and a daily diet of “why don’t the numbers match?”

His first decision is the one worth quoting to every new data hire: don’t rip it out. The Azure stack would do the job. It needed restructuring, not replacing, so he moved it to a medallion architecture, raw, verified, cleansed, modelled, which is twenty-five-year-old thinking and works precisely because it is.

Seven months later: a team of seven, over 250 pipelines running around the clock with Claude-assisted automation on the failure handling, fifteen rebuilt Power BI reports sitting on a single semantic model instead of per-report logic, and, the bit merch actually cares about, a stock rebalancer that takes sales and stock data, works out cover by weeks of stock, and raises the purchase orders back into NetSuite automatically.

Then the showpiece.

When his agentic analytics budget met reality, Microsoft’s MCP endpoints wouldn’t authenticate the way his end users needed, so he asked Claude for an alternative and built a wrapper on the Claude API instead. Total effort, about five days.

The result is ANA, a conversational analytics assistant the merch team named themselves: natural language questions against the Power BI data model, answers with the working shown, suggested drill-downs that change with the question, export to Excel, all behind the normal Microsoft sign-in so nobody sees data they couldn’t already see. It routes simple questions to a cheap model and hard ones to a bigger one, and he estimates $20 per ten to twenty thousand questions. Rollout to the first ten users starts early August.

Subscribe now

His closing point was the one the room wrote down. He’d watched another business demo almost the same screen the week before and admit it hallucinates. The difference, he argued, is four ingredients: good data, a proper data model, context (the business rules a model can’t guess, like “you can’t sum a stock snapshot”), and an ontology for the synonyms, so “gross revenue” means the same thing to the machine as it does to finance. With those four, he claims 99.9%. Without them, you’ve built a confident liar.

Roberta, Technology Solutions Manager at Bluebella,

She showed the build we suspect the most people will copy: a discrepancy dashboard estate that answers one question across the stack, where do our systems disagree? Her diagnosis of the underlying problem was the best articulation of it we’ve heard.

Every team finds cross-platform issues at the worst possible moment for that team: finance at month-end reconciliation, merch just before a launch, ops when a batch of orders has already gone wrong, customer service when somebody complains. Anything that falls between platforms lands on the tech team, late. The dashboard means she finds things the day they happen instead.

The estate covers stock reconciliation across NetSuite and two Shopify instances, orders stuck mid-journey, fulfilment states, despatch speed, and a dedicated ASOS card, because every marketplace has fields and formats that match nothing else. The plumbing runs nightly in n8n: pull from each platform, merge, diff, snapshot, with the previous night’s version kept so every morning there’s a “fixed since last run” view and a briefing pushed to Teams.

Subscribe now

You only open the dashboard when something changed. The same snapshot feeds an MCP into Claude, so a wrong-looking number can be interrogated conversationally: of these orders, pull this information, why do you think this is happening?

Three details to steal.

The diagnosis layer learns by exception: each new failure pattern gets labelled once, and the dashboard names it automatically forever after. The security posture is properly thought through: credentials live in n8n and never touch Claude, access is filtered to exactly the fields needed, everything is read-only, and no customer or payment data exists anywhere in the estate, so sharing a dashboard never grants more access than the recipient already had.

And our favourite: Bluebella’s founder Emily had already built her own ASOS stock tracker, running on her laptop. Rather than replacing it, Roberta absorbed its rules into the shared estate, so the founder’s logic survived and stopped depending on the founder’s laptop being open. That is what healthy internal tooling culture looks like: it collects the intelligence the business already has.

Cost in time, since everyone asked: a few days to land the structure, which partly grew out of conversations at OX1, then half an hour to an afternoon per new dashboard.

Lawrence, managing director of Rough Trade,

He closed the morning, and for a fifty-year-old independent record shop with nine stores across the UK, Berlin and New York, this was the most radical talk of the day, and it was delivered in the calmest voice.

Three months, three problems, all customer service. First the inbox: analysis showed WISMO, where’s my order, was 60 to 65% of all inquiries. Their head of digital built an n8n flow that classifies every inbound conversation into what’s now six paths, and, crucially, once it gained the Shopify MCP connection it stopped merely tagging conversations and started resolving them: checking whether the order genuinely shipped, replying to the customer with the real status, and tagging genuinely late orders in Shopify to create bespoke picking views the warehouse clears daily. Median first response time fell from nearly a day to seconds, and roughly two thirds of conversations are now handled without a human.

Then the honesty that made the room trust everything else he said: CSAT hasn’t followed. Human-response CSAT went up considerably. Bot-response CSAT is low, and their working theory is a plain distrust of labelled automation. They label it anyway, because the alternative is worse, and they’re still working out the nuance. So is everyone.

Then returns, and a reframe worth the ticket price. Digging into the data, they found a huge share of their “returns” were actually cancellations, because two thirds of what Rough Trade sells online is pre-orders. Exclusive pressings, coloured vinyl, order it before it sells out, then minds change or another variant drops. They implemented a best-in-class returns platform and the business rejected it organically, through simple non-use, because it didn’t address the actual problem.

Subscribe now

So they built their own with AI coding tools, shaped around cancellation as the primary event. Customers always get 100% cash back, or can choose 10% extra as store credit, and the finance team’s raised eyebrow lasted until the numbers landed: three to four grand a week retained, an estimated £200,000 to £250,000 a year that would otherwise have walked out as refunds. Hosting costs about ten quid a month against the couple of hundred a SaaS tool wanted.

The app also standardises damage claims (image upload required, because vinyl buyers want pristine and the product is inherently fragile), gates non-delivery claims behind a configurable number of days so the inbox doesn’t drown, and resolves everything in one click into Shopify. Next up: using carrier data to systematically claim back lost deliveries, money most brands write off because the claims process is designed to exhaust you.

And the discipline underneath, which is the transferable bit. After a day of what Lawrence cheerfully called vibe coding, he concluded he was “being so dumb here” and the team adopted structure: Matt Pocock’s Grill with Docs skill, which interrogates a feature request, around a hundred questions in one case, until it fully understands it, maintains the context documentation as it goes, and outputs a PRD the coding agent can execute against for hours unaided.

He’d been back to a four-month-old codebase built without this and had to rebuild the context before any work was possible. Their head of ecommerce, Paddy, then used the same connected setup to build an agent that reviewed every out-of-date help centre article against current policies in Intercom and Shopify and rewrote the lot, with a plan to run it on a schedule. The job at the bottom of every to-do list, now just happening.

Two closing thoughts from Lawrence that summarise where every brand in the room is heading. First, with 85 to 90% of their tools now connected to their AI setup, the constraint has moved: it’s no longer capability or data access, it’s training and empowering non-technical teams, and navigating the culture. Rough Trade sells music as art and won’t be publicly pro AI music, while internally a small-margin business with a DIY ethos has a responsibility to use the tools.

He called the polarised debate what it is, tech-utopian on one side and doom on the other, and planted the business in the middle: it’s a tool, and the DIY tradition now scales. Second, the confession everyone recognised: a period of what he called AI mania around March, everything suddenly buildable, like walking into a bakery where everything’s free, and the resulting discipline that the question is no longer can we build it. It’s should we, and in what order.

The afternoon: three rooms

After lunch the room split into three rotating groups. In one breakout, Henrik from Dema ran a build-along: not a demo watched from chairs but the room building on the platform together, which is the difference between seeing conversational analytics and having made some.

In the other, our own Luke ran a session on transcription, everything he’s learned about going from a meeting recorder being switched on to transcripts becoming an actual working asset in the business, the pipeline from captured conversation to searchable, usable operational knowledge. If you’ve ever recorded a meeting and then done nothing with it, which is everyone, this was the session.

The main room ran the partner roundtables: Dema, Loop, Torque, Emfas, High Cohesion, Centra, ProCarrier, Torque, Resourced, Whanau and us, ten minutes a table, a whistle, and everybody moves. We won’t reproduce table conversations, that’s the covenant of the format, recorders on every table and flip it over if you want off the record. But we can report what recurred across tables independently, because that’s data: returns remains a data blind spot at every size of brand, and the wholesale side of it is a decade behind the consumer side.

The 3PL-versus-own-warehouse question is being asked everywhere, and the useful conversations were about what quotes hide, not about rates. Automation debt is arriving on schedule at the sophisticated brands. And the morning’s theme repeated all afternoon in different clothes: the constraint has moved from capability to judgement, priorities and people.

The rooftop opened at three. What was said up there stays up there.

The ask

Luke and Lauren closed the formal part with a request we’ll amplify, because it’s the right one. OX grew out of the back office, finance, ops, merchandising, customer service, systems, the functions this industry’s events chronically underserve, and the goal is cross-functional: if the day was valuable, bring colleagues from the other functions next time, because the catalyst thing only works if it spreads. We’ll share OX3 details as soon as they exist. If your job is running an ecommerce operation rather than talking about one, be in the room.

Find out more about OX

This one is the fullest version of that yet.

Three brand-side builds

The morning is Show and Tell. Bluebella, Adanola, and Rough Trade. Three operators, three real builds and they’re all in production. The through-line across all three: one workflow, one owner, one data source and one measurable outcome. None of it is a roadmap or a pitch. It all runs on the day.

Three afternoon tracks

The afternoon splits into rotating sessions across three rooms. Dema run a live build, tuned to the room each time, so you see the decisions get made rather than a finished result. Commerce Thinking on what actually happens after you transcribe a meeting, and the work of turning a transcript into something a team acts on. And the clinic, brand-side operators, partners who build software for brands, and dedicated time to work through real problems rather than sit through pitch decks. Bring a live blocker and you’ll get somewhere on it.

After the day

Rooftop drinks from 3pm. Then the Mulberry Bush from 5pm for anyone still standing.

OX2 is Thursday 16 July at Coin Street Conference Centre, London.

We are pretty much at capacity but if you want to be in the room, apply at operatorexperience.ai and we will make space.

Most brands miss that window. Not because they weren’t paying attention. Because they were too busy growing to notice it closing.

The mistakes that don't look like mistakes

We speak to brand leaders regularly who are worried about rolling out AI because of the mistakes they have already seen. We prefer to call them failed experiments. But they are worth understanding, because they follow a pattern.

The most common version is AI being used to automate tasks by people who have never done them manually first. Without that grounding, there is no frame of reference for when the output is wrong. Work gets produced, looks plausible, and passes without challenge because nobody in the chain has done it the old way and knows what to look for.

The more serious version is AI being used to generate outputs by people who do not understand the inputs and therefore cannot verify them. A report that looks credible but is not. An analysis that reads as thorough but has not been checked. The risk is not that AI was used. It is that it was used invisibly, without accountability, by people who did not fully understand what they were looking at.

Neither of these happens because teams are careless. They happen because AI adoption runs ahead of process. And the fix is not to slow down the adoption. It is to build the foundation underneath it.

The foundation nobody builds first

The single biggest unlock for effective AI use is process documentation. Not a corporate manual. A clear, end-to-end record of who owns what, how processes feed each other, and where handoffs create inefficiency.

Fast-growing brands rarely have this. Not because they do not know it matters, but because the person who would write it has always been too busy keeping the business moving to do so. The critical path exists. It lives in someone’s head. It has just never been on paper.
AI does not fix an undocumented process. It accelerates it. Point a capable AI tool at a clean, well-documented operation and you get real leverage. Point it at an undocumented one, and you get the mess moving faster, produced by more people, with a convincing veneer on top.
Process mapping is the exercise that changes this. Not over-engineered. Not a six-month project. A lean, usable record of how the business actually runs, built before the AI is pointed at it.

The window

Startup to scaleup brands are in the best position to embed AI properly before they scale headcount. The org chart is still fluid. Cross-functional conversations can still cut across the whole business without becoming a change management project. And the decisions made now about how AI sits inside the operation will either compound well or compound badly for years.

The model worth building toward is cross-functional squads: human and agent working together, with tasks mapped clearly across the workflow. Some kept with the human. Some delegated to the agent. Some dropped entirely because they were never worth doing in the first place.

At this stage, that is a conversation. At 40 or 50 people, it is a restructure. The brands that get this right are the ones that use the window while it is still open, embedding agents into the places where monotonous, repeatable work is currently absorbing a person’s entire week, rather than hiring someone to do that work full time and then trying to unpick it later.

The question is what shape the org chart turns into.

The room where this gets worked out

The types of people we see in every business relative to AI, the enthusiasts with fuzzy deployment, the passengers waiting to be told, the lone wolves, and the rare catalysts, are not evenly distributed, and they are more likely built than hired. That is not a hiring problem. It is a leadership and process problem.

These are exactly the conversations that happen at OX. Not theory. Operational reality, from brand-side leaders working through the same things in their own businesses.

OX2 is on 16 July at Coin Street Conference Centre, London. We are pretty much at capacity, but if you want to be in the room, apply at operatorexperience.ai, and we will make space.

Most of that played out, but some of it was more complicated than it looked at the time.

What actually happened

The August 29, 2025 deadline landed roughly as predicted in terms of direction. Complications came from the execution.

Courier networks took the hit first, with processing times stretched. Duty invoices were inconsistent for weeks. Shipments were cleared without issue one day and held without warning the next. The infrastructure simply was not built for the volume of declarations that suddenly needed filing. Teams that had been operating on predictable lead times lost that predictability overnight, and it took most of the autumn to settle.

The duty arbitrage played out. Brands that had already moved inventory into US 3PLs found themselves at a structural cost advantage over brands still shipping direct. The gap between paying duty on retail value versus landed cost was real, and in categories like fashion and wellness it was often the difference between a viable US channel and an unviable one.

What the original piece underestimated was the documentation burden. Getting HS codes right turned out to be harder and more expensive than most brands anticipated. Sloppy classification created cost surprises across thousands of orders. Several brands absorbed unexpected penalties before they got their data in order. The compliance piece was not a one-week job. For some, it was a quarter.

What AI changed about this

Here is the part that did not exist in the same form when we wrote the original piece.

The documentation problem, the one that caught brands out, is now largely a solved problem if you are willing to use the tools. HS code classification, duty modelling and invoice reconciliation against what you were actually charged. These are exactly the kinds of jobs AI handles well: repetitive, rule-based and requiring patience rather than judgement.

We wrote earlier this year about the work that was never worth doing. The fifty-quid invoice error nobody caught because catching it cost more than it was worth. Cross-border duty reconciliation was the same problem at a larger scale. A misclassification across a month of US orders was worth chasing. Nobody had the bandwidth to chase it, but now they do.

That changes the calculus on whether a US channel is viable, because it lowers the cost of managing it correctly.

What it tells us about what comes next

The US experience is the closest thing we have to a preview of how the EU and UK changes will land. The pattern was consistent: the tariff itself was not the worst part, the shock to the system was. Courier networks were not built for the declaration volume. Duty invoices were inconsistent. Clearance times became unpredictable. The brands that came through it cleanest were the ones that had already moved stock into the US before August 29, not because they were lucky but because the duty arbitrage calculation had already made the case.

The same arbitrage exists for any market where you are currently shipping direct. Duties assessed on retail value versus duties assessed on landed cost is not a small difference. For the brands that ran the numbers and moved, that gap funded the 3PL setup.

What to actually do about it now

Start with the model. Work out how many of your EU and UK orders would sit under the relevant thresholds, because those are the ones that get more expensive when the exemptions close. Establish the actual landed cost impact if duty applies to every one of them. Most teams are surprised at the size of the number.

Get your HS codes in order. The US experience showed that sloppy classification creates cost surprises across thousands of orders and penalties that take a quarter to unwind. That is work to do before the pressure arrives, not after.

Review your minimum order thresholds. Orders under a certain value may no longer be commercially viable once duty and handling stack up. Know where your line is.

And seriously model the regional fulfilment case. The brands that moved inventory into the US before the deadline are operating at a structural cost advantage now. The window to do the same for the EU exists. It will not stay open.

The party ended a year ago in the US, with the EU and UK versions next. The pattern is the same, and the question is whether you use the interval to get ready, or wait until the deadline is already live.

None of it arrived as one big headline. It came in pieces. A customs tweak here, an accessibility deadline there, a product safety rule that most brands only noticed when their listings got pulled. Individually, each one is a footnote. Stacked up, they add up to the biggest change in what it takes to sell into the EU since GDPR.

And that GDPR comparison is the thing to hold onto, because it explains all of it. Every piece of this applies based on where your customer is, not where you are. A brand in Manchester or Austin does not get a pass because it has no office in Frankfurt. If a German shopper can buy from you, the EU considers you in scope. Your postcode is irrelevant. Your customer’s postcode is everything.

Which is a problem, because most brands are already far more international than they ever decided to be. The revenue went global. The operations, mostly, did not. All of the rules below land on exactly that gap.

Here is what has already gone live, what is landing next, and what any UK or US brand actually needs to do about it.

The €3 that eats your margin (live from 1 July 2026)

Start with the one people are actually talking about, because it hits the number that matters.

For years, any parcel worth €150 or less could enter the EU free of customs duty. VAT still applied, but duty did not. That exemption is gone from 1 July 2026. In its place, the EU is charging a flat €3 customs duty on those low-value parcels while it builds the proper system underneath.

Sounds small. It is not, and here is the sting. The €3 is charged per product category, not per parcel. The technical bit is that it lands on each distinct item type in the order. So a parcel with a blouse, a belt and a pair of socks is three different headings, which is €9, not €3. The more varied the basket, the more it stacks.

One thing worth knowing for later: country of origin will also affect the calculation, but not yet. During the interim phase through to July 2028, the €3 flat rate applies regardless of where the product was made. From July 2028, when the EU Customs Data Hub goes live, real tariffs based on HS code and country of origin replace the flat fee. That is when trade agreements and anti-dumping duties come back into play.

For anyone shipping low value, multi item orders direct from outside the EU, that is a real dent.

Then it gets worse before it settles. A separate handling fee of around €2 per parcel is expected from November 2026, which takes the combined charge to roughly €5. The €3 rate itself is temporary. It runs until the EU’s new customs data hub comes online around mid 2028, at which point normal product by product tariffs replace it. So the flat fee is not the destination. It is the holding pattern.

This is not the EU acting alone, either. The US killed its own $800 de minimis threshold in 2025. The UK confirmed in the Autumn Budget 2025 that it is scrapping its £135 threshold too, with reforms landing around 2029. The whole world is closing the small-parcel loophole at once.

What to actually do: work out how many of your EU orders sit under €150 today, because those are the ones that just got more expensive. Get your HS codes right, because sloppy classification now costs you money on every order. And seriously look at holding stock inside the EU. Bulk import once, clear duty once at the border, fulfil locally, and the whole per-parcel problem disappears. This is the point at which international complexity stops being a future problem and starts showing up in this quarter’s margin. The brands that move stock in before July will be selling at a structural discount to the ones absorbing €5 a parcel in August.

The product safety rule that's already pulling listings (live since December 2024)

If the customs change is the one everyone is talking about, the General Product Safety Regulation is the one everyone missed. It has been live since 13 December 2024, and plenty of brands only found out when Amazon or Etsy hid their listings.

The GPSR replaced an old directive and tightened the rules for basically every non-food consumer product sold to EU shoppers. The headline requirement is the one that catches out non-EU brands. You now need an EU-based “responsible person” before your product can be sold. That is a named economic operator inside the EU (or Northern Ireland) who holds your technical documentation and answers to the regulators. It can be an EU importer, a fulfilment provider, or an authorised representative you appoint. But it has to be someone, and it has to be someone inside the bloc.

No responsible person, no market access. Your product listings also have to display the manufacturer’s contact details, the responsible person if the manufacturer is outside the EU, product identifiers, an image, and any safety warnings. The marketplaces are already enforcing this. Non-compliant listings get delisted, and you lose the sales before you understand why.

What to actually do: if you sell physical goods into the EU and you have not appointed a responsible person, that is the single most urgent line on this whole list, because you may already be non-compliant. Fix the responsible person, then fix your listing data.

The accessibility deadline that went live in June 2025

The European Accessibility Act switched on for enforcement on 28 June 2025, and it is aimed squarely at your website.

If you run an ecommerce service for EU consumers, your site and app have to be accessible to people with disabilities. In practice, that means meeting the WCAG 2.1 AA standard across the whole shopping journey, and it is all of the criteria, not the convenient ones. Screen readers have to announce your variant pickers. Keyboard users have to get through your checkout. Price changes have to be read out when someone changes an option. The places sites usually break are the fancy bits, the carousels, the multi-step checkouts and the dynamic carts.

There is a carve out for genuine microenterprises, under ten staff and under €2m turnover. Almost nobody reading this qualifies. Existing systems get some breathing room until June 2030, but new work is expected to be accessible now, and you are supposed to publish an accessibility statement either way.

This is not theoretical. French disability groups filed formal notices against major retailers within days of the deadline and took several to court by late 2025. Regulators in the Netherlands, Germany, Sweden and others are actively surveilling. Penalties vary by country and run into the millions in the worst cases.

What to actually do: get someone to run your full purchase flow with a screen reader and see where they get stuck. That one exercise tells you 80% of what you need to fix. Then publish an accessibility statement. It is also, for what it is worth, just good commerce. There are tens of millions of EU shoppers with disabilities, and an inaccessible checkout is an abandoned cart with a legal risk attached.

The AI rules, and the part that actually applies to you (landing August 2026)

The EU AI Act has been phasing in since 2024, and most of the coverage misses the point for ecommerce brands, because most of the scary stuff is about high-risk systems like credit scoring and biometrics that you probably do not run.

The bit that lands on marketing and ecommerce teams is the transparency rule, and it applies from 2 August 2026. In plain terms, when a customer is talking to a bot, you have to make it obvious it is a bot. When content is AI-generated, including synthetic images and deepfakes, it has to be labelled. If you have leaned into AI chat support, AI product recommenders, or AI-generated creative and imagery, those are the touchpoints in scope. Worth knowing, given how many operators are figuring out AI on their own, that “we’ll sort the compliance bit later” is a common and risky default here.

One caveat worth knowing. The EU has a simplification package in flight (the so called Digital Omnibus) that reached political agreement in May 2026 and may shift some of the high-risk timelines. The transparency obligations are the part that matters for most brands, and those are the ones to plan around. The Act also reaches beyond the EU’s borders, same as everything else here, and the fines are the largest of any of these rules, up to €35m or 7% of global turnover at the top end.

What to actually do: audit where AI touches your customer. Label your bots. Label your synthetic content. It is not a heavy lift, but it is the kind of thing that is embarrassing to be caught not doing.

Packaging, and the death of the oversized box (from August 2026)

Last one. The Packaging and Packaging Waste Regulation applies from 12 August 2026, and it changes how you can ship.

The rule ecommerce teams will feel first is the empty-space limit. Void in a parcel cannot exceed 40%. The era of a lipstick in a shoebox full of air pillows is over, because that air now counts against you. There are also recycled-content requirements, restrictions on nasty substances in packaging, and, for non-EU sellers, a requirement to appoint an authorised representative to handle it. Digital labelling on packaging follows from 2027.

What to actually do: right-size your parcels and talk to your packaging supplier about recycled content and documentation. The unboxing flex has a compliance cost now.

Digital product passports, coming for your labels (from 2027)

This one is further out but worth flagging now, because the lead time to comply is long.

The EU is building towards a system where products carry a Digital Product Passport, a QR code or similar that links to verified information about materials, carbon footprint, repairability and where the thing came from. The first mandatory version is the battery passport, live from 18 February 2027 for larger batteries. Textiles and apparel are next in the queue, with the rules expected to be adopted around 2027 and compliance landing roughly eighteen months after that, so realistically late 2028 into 2029. Furniture, electronics and others follow.

There is no single deadline. It rolls out category by category. But if you sell textiles, footwear, electronics or furniture into the EU, the passport is coming for your product data, and gathering that data across a messy supply chain is exactly the kind of job you cannot start the week before it is due.

The pattern, and the point

Read all of that back, and you notice it is the same move, over and over. Your parcel, your website, your product, your packaging, your bot. Each one is now something the EU will regulate regardless of where you sit, on the single condition that a European can buy from you.

That is the mental shift. Selling into Europe used to be a shipping question. It is now a compliance question, and the truth is that the barrier to entry just went up. Which, if you look at it from the right angle, is not entirely bad news for the brands that get organised. Every one of these rules is a hurdle. Every hurdle thins the field. The cheap, non-compliant, drop-shipped competition that used to undercut you on price is exactly what most of this is designed to squeeze out.

So the brands that treat this as a checklist and get it done are not just avoiding fines. They are slowly pulling away from everyone who is still hoping none of it applies to them. That is usually how era changes actually play out: the shift looks like admin right up until it looks like a gap nobody can close.

It applies. Get the map, and start walking.

OX runs the other way round. The room is the product. The number on the door isn’t a capacity we’re straining against, it’s a decision. And right now that decision leaves 17 places.

That’s not a countdown timer engineered to make you panic-click. No fake “only 2 left!” that resets every time you reload. When the 17 are gone, they’re gone, because the entire value of OX is the specific people sitting in it. Add 40 more to hit a revenue target and you don’t get a bigger OX. You get a worse one.

What it is

Operator Experience is an invite-only community for brand-side leaders. No agencies. No platforms. A closed room of people in the same seat, under the same pressure, trying to solve the same things, without anyone in the corner waiting to pitch them.

OX2 is the second one. It runs from 10:00 on Thursday 16 July at Coin Street Conference Centre in London.

The morning is show and tell. Operators walk the room through workflows they’ve actually built for real brands. Not slides about what’s theoretically possible. The thing they shipped, why they shipped it, and what broke on the way.

Lunch at 12:00.

The afternoon is three rotating 40-minute practical sessions, and everyone does all three. Groups are mixed on purpose, so you’re sat next to someone from a different function rather than nodding along with your own kind. The day ends on the rooftop from 15:00, while it’s still light. Pub after for anyone who wants it. No obligation if you don’t.

Subscribe now

Why it’s capped

OX1 happened in Manchester. Nearly 90 brand-side operators, and the feedback came back saying the same thing over and over. The best part wasn’t the agenda. It was the person next to them.

So OX2 is built to manufacture more of that and less of everything else. More operators describing what they’ve actually made. More tables deliberately mixed so the conversation doesn’t stay in its lane. A finish time that respects the fact you have a train to catch and a life to get back to. (Or stick around in the pub afterwards, no pressure).

You cannot engineer that quality of conversation in a room of 300. You can in a room this size. That’s the whole reason for the cap, and the whole reason there are only 17 seats left.

Who it’s for

Senior people working brand-side: ecommerce, ops, supply chain, logistics, merchandising, buying, finance, product data, tech or CX, at a fast-growing fashion, lifestyle or consumer brand.

Or, if you’re in-between roles and want to share your knowledge and/or learn from other operators, please come along.

If you’re an agency or a platform hoping to slip in, this isn’t the one. That’s not snobbery, it’s the deal. OX is free to attend because partners fund it on one condition: they play by the same rule as everyone else. No pitching, no posturing, treat operators as peers. Headline partner Dema, event partners Loop, Torque and Emfas, and community partners including Vervaunt, HighCohesion, Resourced, Centra, Whanau, Pro Carrier, Commerce Thinking and Waye.

The 17

The room is small because small is the point. If you’re the right person in the right seat, this is the rare event built entirely around you, rather than around the people trying to sell to you.

There are 17 places left. When they’re claimed, OX2 is full.

operatorexperience.ai

RSVP Here

The variable that consistently predicted trust across both returns and cross-border, was support. The cohort: 24 brands, fashion-led, UK-headquartered, £10m to £100m. The data is operator data, not vendor data.

Not support in the ticket-response-time sense. Operators aren’t measuring their vendors by how quickly an automated acknowledgement lands in their inbox. What they’re measuring is something harder to systematise: genuine urgency when something breaks, commercial understanding that goes beyond technical resolution, and a willingness to adapt alongside the business rather than deflect to a product roadmap.

Our research found that 71% of operators cited support as the top factor in vendor sentiment. 68% said they would pay more for better support. And 45% had left a vendor specifically because of poor support. Those numbers are striking on their own. But the qualitative feedback is what gives them context.

The vendors rated most highly weren’t the ones with the most features. They were the ones that showed up.

That framing has limits worth naming. Not every operator arrives at vendor evaluation on equal footing.

"A lot of the big cross-border and returns platforms are positioning themselves as a benefit when in fact, legacy brands have very strong relationships with carriers, and so there is less of a benefit." That's from Katie at Strathberry.

The point isn't that support doesn't matter. It's that the baseline value proposition varies by operator, and sentiment follows from that.

Reveni scores 9.5 out of 10 in our research, and the feedback points directly to this: “Reveni have genuinely gone above and beyond for us. They’ve built custom setups, developed new integrations, and worked closely with our team throughout, without cutting corners or over-promising at any stage.” That’s not a feature review. It’s a description of a working relationship.

Swap draws similar feedback, particularly around responsiveness during high-stakes periods: “We launched Swap exchanges on Black Friday week and there were no issues whatsoever. Plain sailing during BF week, unheard of.” The value isn’t just that it worked. It’s that it worked when the stakes were highest and the margin for error was zero.

Loop sits in a different position: respected for stability, trusted at scale, with an opportunity to lean further into operator-led insight and commercial clarity. The platform works. The relationship has room to deepen.

Subscribe now

At the other end, Global-e, Narvar, and ZigZag are more frequently described as slow, rigid, or difficult to evolve. Not universally negative, but the friction is higher and the patience is lower. One operator described their cross-border provider as "the market leader, thus regarded as best in class, but not a pragmatic partner."

A managing director at a luxury footwear brand was more specific about what that looks like in practice: their cross-border provider's portal was built around returns, not exchanges. "I said, but guys, this is a refund portal. This is not an exchange portal. This is not fit for purpose."

The cost equation follows the same logic

Brands will pay premium pricing when value is visible and the relationship feels like a genuine partnership. They’ll reject lower-cost options that feel opaque or hard to measure. The vendors winning right now aren’t the cheapest. They’re the ones that make it easy to see what you’re getting.

For operators evaluating vendors, or reconsidering existing relationships, the practical implication is to weight support quality more heavily than most RFP processes currently do. A platform with a slightly shorter feature list and a genuinely responsive support model will outperform a more capable platform with a generic helpdesk almost every time.

Our full research report covers vendor perception in detail, including how support shapes sentiment across the full returns and cross-border landscape.

Read the full report