The order system is
the source of truth.

The symptom

Somebody asks how many units of the bestseller sold last month. Marketing answers from analytics. Finance answers from the order system. The two numbers differ by five, ten, sometimes twenty per cent, and the meeting turns into an argument about which tool is broken.

Neither is broken. They are counting different things, and only one of them is counting money.

Why the two never agree

• Consent. A visitor who declines analytics cookies still buys. The order lands in the shop; the event never reaches analytics. Under EU consent rules this is not an edge case, it is a meaningful share of European traffic.
• Blocked tags. Ad blockers, privacy browsers, corporate networks and plain script failures all drop the purchase event. The order still exists.
• Modelled data. Modern analytics fills consent gaps with modelled conversions. That is a reasonable answer to a hard problem, and it is an estimate. Estimates do not belong in a margin calculation.
• Attribution, not counting. Analytics assigns a sale to a channel using a window and a model. Change the model and the same sale moves to a different channel. The order system does not have an opinion about who deserves credit — it records that an order happened.
• Cross-device and re-consent. A purchase begun on a phone and finished on a laptop is one order and frequently two sessions.
• Returns and cancellations. A refund three weeks later reduces the money and almost never flows back into the analytics view. Revenue reported in analytics tends to be gross for ever.
• Currency, tax and rounding. Analytics usually records what the front end sent. The order system records what was invoiced, in which currency, with which tax treatment.

What each one is actually for

The order system answers “what happened”

Orders, items, prices, discounts, cancellations, refunds, invoices. It is transactional, it is auditable, and it is the thing the bank statement can eventually be matched against. This is the ledger.

Analytics answers “how did people behave”

Where they came from, what they looked at, where they stalled, which step of the checkout they abandoned. It is excellent at this and nothing else does it as well. Behaviour is a sample; sampling is fine for behaviour and fatal for money.

The rule that follows is short: count with orders, explain with analytics. Any report that mixes the two into one number is a report nobody can defend.

How to check this on your own data

Take one closed month and one market. Then, in this order:

• Pull order-level data from the shop: order id, timestamp, items, quantities, gross value, discounts, currency, status.
• Remove what is not a sale: cancellations, test orders, unpaid orders left in a pending state, staff orders.
• Subtract refunds by the date of the original order, not the date of the refund. Otherwise every month looks better than it was and last month keeps getting worse.
• Pull the same period from analytics: purchases and revenue.
• Compare units first, value second. Units are cleaner — they carry no currency, tax or discount logic.
• Write down the gap as a percentage, and repeat for a second month.

A stable gap of a few per cent, in the same direction, usually means consent and blocked tags — irritating, expected, and harmless as long as you count with orders. A gap that swings between months, or reverses sign, means something is actually broken: a tag firing twice, a thank-you page reachable on refresh, a payment provider redirect that loses the event, a currency mixed into the wrong bucket. That is worth an afternoon.

What good looks like

• One documented definition of a sale, written down, used by everyone.
• Orders as the base table for every revenue and margin report.
• Analytics joined to that base for the behavioural questions, never used to restate the totals.
• Refunds attributed back to the original period.
• A known, monitored gap between the two systems — because the day it moves, something changed.

Why this is the first thing to fix

Because everything downstream inherits it. Margin by SKU, advertising efficiency, stock planning, the decision about which market to push next — all of it is built on the count of what was sold. Get the base wrong and the sophistication above it is decoration.