Most of our work in Commerce Thinking is product data, the transactions related to inventory and the operational processes between everything back-office. We are not CRM or CDP guys, but yesterday I read Chris Bishop’s piece on knowing your real customer (he’s the Ecommerce and Digital Director at Squatwolf) and most of the questions in it link back to our world.
He argues that brands know their revenue and bestsellers far better than they know who is buying from them. He puts it as “How many customers do you have? Not orders or transactions. Customers.” Then he works through how those customers behave, by cohort and by market, all the way down to contribution margin. In our experience, a brand that tries to answer those questions gets one answer per system, and the answers don’t match.
An FD, COO or ecommerce director asked for a cohort view or a customer-led forecast for the board, which will usually find that Shopify, the ERP and the CRM each think the business has a different number of customers.
Each system has its own idea of a customer
Shopify creates a customer from the email address at checkout, so one person with a work and a personal email becomes two customers. Klaviyo counts profiles, which includes everyone who signed up for 10% off and never bought. Retail sales through POS often have no customer attached at all, and some marketplaces never pass you a usable email in the first place.
Then there’s the ERP. Many integrations post web orders to one generic customer record, or as a daily summary, because finance doesn’t need tens of thousands of customer records to close the month. That keeps a system like NetSuite or SAP tidy and means the system holding the margin has no idea who generated it.
Contribution margin by customer
Repeat rate can come from the commerce platform alone. Contribution margin needs refunds, returns, landed cost, shipping and payment fees attached to the right order, and each of those comes from a different system.
If the integration skips refunds or posts them weeks late, the customers who return most of what they buy start to look like your best ones. If landed cost runs off a template nobody has updated since the last freight contract, the error carries into every cohort. Comparing this year’s customers with last year’s then tells you more about the template than about the customers.
Market splits follow how you set up the stack
Chris is right that UK customers won’t behave like customers in the UAE or Saudi Arabia. Whether you can see the difference depends on decisions made at launch, such as one store or separate expansion stores, and one entity or several. Currency adds another layer, because converting at order and converting at settlement give you two different order values for the same customer.
If a market runs through a distributor or a local partner storefront, the end customer may never reach your systems. Sometimes the global average is the only number the stack can produce.
Check the plumbing before you build the forecast
Five things to settle before anyone opens the model:
Pick one customer ID and decide which system owns it. It’s usually the commerce platform or the CRM, rarely the ERP
Find out how DTC orders reach the ERP, whether as individual customers, one generic customer or a daily summary
Check that refunds and returns post against the original order, and count how many failed to post last month
Confirm landed cost and payment fees sit at order level rather than in a month-end journal
List the channels and markets that don’t capture a customer at all, and say so on the report
If you’re the one who has been asked for the customer-led forecast, read Chris’s article and plan in these checks.




