Ecommerce Attribution: Reconciling Platform Numbers With the Bank

Platform revenue and bank deposits never match, and reconciling them is not about forcing them to agree. It is about naming the size and cause of each gap: claimed conversions, store orders, refunds, discounts, shipping and tax, processor fees, payout timing. Once every gap has a name, both numbers become usable.

Why the platform and the bank disagree

They count different objects over different clocks. An ad platform counts conversions it believes it influenced, dated to the click, inside its own window. Your store counts orders on the day they were placed. Your bank records money that settled, after fees, on the day the processor released it.

Each step introduces a legitimate difference. Discounts and gift cards reduce revenue without reducing order count. Returns land weeks after the order. Shipping and tax inflate gross revenue that was never margin. Fees and chargebacks come out before deposit. Currency conversion moves the figure again on international sales.

Then there is overlap. Run several channels and each platform may claim the same order. That is not dishonesty, it is each one answering its own question.

Which number should you manage against?

Pick one number to run the business on and demote the rest to inputs. For most ecommerce brands that is net store revenue, with contribution margin as the decision layer above it, because it survives returns, discounts and the cost of goods.

Platform reported revenue stays useful for allocation inside a channel. It tells you which campaign the model favours and which creative earned delivery, at a granularity nothing else offers. It is a poor number for deciding whether the business grew.

Bank data is the audit: slowest, least granular, hardest to argue with. Most of our ecommerce tracking and attribution work begins by agreeing which of these runs the weekly meeting, because teams that skip it end up debating definitions whenever results are questioned.

How to run the reconciliation by hand

Choose a closed month. Export store orders with gross revenue, discounts, shipping, tax, returns and cancellations. Export spend and reported conversions from every ad platform for the same window. Export processor payouts, including fees and the dates money moved.

Work downward on one sheet. Gross store revenue, less discounts, less returns and cancellations, gives net revenue. Net revenue less processor fees gives the amount that should reach the bank, adjusted for payouts that straddle the month boundary. Compare that to deposits.

Then place platform claimed revenue beside store revenue for the same window. Do not try to make them equal. Write down the difference and which known cause explains it. The first pass takes an afternoon, and it is the most valuable afternoon in the quarter.

What the gaps tell you once you can see them

Claimed revenue above store revenue points at double counting, between platforms or between browser and server events inside one. Claimed revenue far below points at signal loss: consent suppression, broken tags, or a checkout that never fires the purchase event.

A stable, explainable gap between store and bank is healthy. A moving one means returns are rising, fees changed, or a payout schedule shifted, none of which are advertising problems even though they arrive disguised as them.

Working with a womens fashion brand, the strongest single month over the eight month period was July, at $432K.

What made it usable was not the dashboard it came from. It was that the same month had already been walked down to deposits, so we knew what it contained and what it did not.

Where attribution tools help, and where they stop

Tools are good at the tedious, repetitive parts: joining store and platform data daily, holding history, modelling paths across channels, and giving everyone a shared view that updates without a spreadsheet. That is real value and worth paying for once the definitions are settled.

What they cannot do is decide which number governs. They inherit whatever the store and the platforms feed them, so an unfixed tracking fault becomes a confidently presented wrong answer. They rarely see wholesale, marketplace or offline revenue unless it is fed in deliberately. Their modelled outputs differ by vendor, which is fine as long as you know you are reading a model rather than a ledger. The sequence matters. Settle the definitions first, then automate them.

Reconcile one month manually before buying any tool to do it for you.

Jason Lu runs Plaid Testing, an executive marketing partner for ecommerce brands. He has spoken on panel at The Whalies and works mainly with apparel and wellness brands.