en
PolskiEnglishDeutschУкраїнськаEspañol
CASE STUDY

A marketplace retailer: from spreadsheets to per-product profitability

A production deployment where reporting stopped being manual work and margin started agreeing with the marketplace settlement.

This case study is anonymised — we name neither the client nor their commercial figures.

Sector
Marketplace retail
Scope
Ingestion, data model, profitability, AI reporting
Status
In production

Where they started

The data sat in several places at once and the report was assembled by hand — slowly, and slightly differently every time. The expensive part was that nobody could defend a single margin figure.

Profitability calculated in a spreadsheet, rebuilt for each period.
Fees and delivery costs bolted on afterwards, and often missed.
Marketplace corrections changed a closed month that nobody recomputed.

What we built

Rather than another dashboard over the same data, we started with the sources and the definitions, and only then built the reporting layer.

1
Automated ingestion
Data from the marketplace, the client’s own system and their logistics provider arrives on a schedule, incrementally.
2
A profitability model
Price minus fees, cost of goods and delivery cost — allocated to individual products and orders under explicit rules.
3
Returns and retroactive corrections
When the marketplace posts a correction after the fact, the period is recomputed instead of frozen.
4
An AI-written report
A weekly summary explaining what moved and why, built from the same metrics as the dashboards.

The outcome

Reporting moved off people and onto the platform, and the margin conversation stopped being an argument about whose spreadsheet was right.

Margin that matches the settlement
Fees, delivery and returns are inside the number the board looks at — including after retroactive corrections.
Decisions at product level
It is visible which product earns after every cost, not merely which one sells.
Reporting without manual work
The weekly summary arrives on its own — the team reads and decides instead of compiling.

Why the client trusted it

Before the switch we ran both tracks side by side: the platform and the client’s own calculator, built up over years. The figures matched to the penny — only then was the spreadsheet retired. We consider this the only honest way to put a reporting layer into production.

Same model, different data.

We will show you how this looks on your sources and your KPIs.

Commerce analytics