Monday, nine o’clock. The managing director of a fashion brand asks for last week’s sales across all channels. The brand sells on its French website, which runs on Shopify, and on its Spanish website, which is still on PrestaShop. It has stores in France and Spain, equipped with point-of-sale software. It sells on Amazon and Zalando, and through department stores and multi-brand retailers that buy from it wholesale.
The management controller spends the day exporting, converting and pasting. The figure goes out on Wednesday. At the management meeting, the e-commerce manager presents a different total, taken from Shopify, and the wholesale manager a third. Each counts a sale in their own way.
Each channel has its own definition of a sale
The sources differ in their format and in what they measure.
- The date. The website records a sale at the time of the order, the till when the purchase is rung up. Amazon distinguishes orders from settlements: its settlement reports are generated automatically, usually every fourteen days, in line with payouts to the seller (Amazon SP-API documentation).
- The amount. Including or excluding VAT, shipping costs included or not, discounts, marketplace commissions deducted or not, and the currency of the country of sale.
- Returns. In fashion, they weigh heavily and arrive weeks after the sale. Zalando, for instance, announced measures against excessive returns in 2025 (FashionUnited). A week’s net figure therefore changes after the fact.
- The level of detail. What a wholesaler buys from the brand (sell-in) does not show what it resells to consumers (sell-out). When it sends its sales, it is often through an EDI message of the SLSRPT type, weekly or monthly, and quantities may be cumulative or per period, two conventions that must be identified before any calculation (xedi).
Products must also be matched: a single item has an internal reference, an EAN code, an Amazon identifier and sometimes the wholesaler’s own reference.
Comparing comparable periods
A Monday in September cannot be compared with a Saturday in September. For this reason, American retailers use a so-called 4-5-4 calendar, which divides each quarter into months of four, five and four weeks, so that the months being compared contain the same number of weekends (NRF). An extra week is added every five or six years.
Without going that far, leadership needs to know the basis of comparison: the week against the equivalent week last year, like-for-like stores (excluding openings and closures), the same scope of countries.
Real time, channel by channel
Not all sources move at the same pace. The till and the website can send their sales continuously through an API. Amazon’s sales reports arrive with a delay of about one day (Openbridge). Wholesalers report weekly, sometimes monthly.
An honest dashboard shows, for each channel, the time of its last update. The total for the current week is presented as partial until the slow sources have completed it.
What the system does
The system connects to each source through its interface: Shopify and PrestaShop APIs, point-of-sale software, Amazon API, Zalando partner reports, wholesaler EDI files. It:
- applies a single definition to each indicator (gross sales, sales net of returns, excluding VAT, in euros at the chosen rate, on the chosen date);
- matches product references from one channel to another;
- presents, on mobile, sales by channel, by country and by period, with the freshness of each source;
- answers leadership’s questions, asked in plain language: “How much of the autumn collection have we sold in Spain since 1 September, excluding wholesalers, compared with last year?”
The answer gives the figure, the definition used, the period, the channels included and the date of their data.
The check: every figure links back to a definition and its sources
A second system, separate from the one that answers, checks the answer before it is displayed:
- the answer uses only defined and validated indicators;
- the period, the scope and the freshness of the sources are stated;
- a question that would require an undefined indicator is not improvised: it is passed to the management controller;
- each day, the total for each channel is reconciled with the source’s total, and each Amazon settlement with the sales it covers; a discrepancy triggers an alert.
What remains the finance department’s responsibility
Choosing the definitions is a management decision: the date a sale is recognised, the treatment of returns, exchange rates. The system applies what has been decided. The accounts remain the reference for the close; the difference between the dashboard and the month’s accounts is explained every month.
Where to start
You will need:
- the list of channels and countries, with access to the APIs and exports;
- one month of real exports from each channel;
- the report currently presented to the management committee, with its definitions;
- the management controller, to validate the definitions and the first reconciliations.
We start with three sources, for example the website, the stores and Amazon, and then add the other marketplaces and the wholesalers. The result is measured by the time between the end of the week and the figure being shared, and by the difference from the accounts.
Other uses of AI in retail, and a detailed scenario, are described on the page AI applied to retail.