Commerce data that agrees

Business context

Web, store, and marketplace channels each produced a version of sell-through. Finance closed the month on a different extract. Promotions made the disagreement worse.

Technical challenge

Establish a certified set of order, inventory, and contribution metrics that operations and finance would both use, without replacing the commerce platform mid-season.

Constraints

• Peak trading window could not host a cutover.
• Marketplace and POS vendors remain systems of record for their channels.
• Existing BI licences stay; the semantic layer has to meet the tools people already open.

Approach

• Write metric contracts for five numbers only: ordered, fulfilled, returned, on-hand, and contribution after known costs.
• Land raw extracts with lineage rather than transforming in the BI tool.
• Model inventory movements as events, even when the source only offered daily snapshots.
• Publish certified tables and retire the three unofficial workbooks that had become policy.

Technologies in play

Named only at the category level. We do not list products as if they were a certified stack for a fictional client.

• The warehouse or lakehouse the company already paid for
• Existing orchestration and BI tools
• Source extracts from commerce, POS, and finance

Implementation notes

• A war-room week documented every unofficial definition of “available”.
• Tests failed the pipeline when grain or uniqueness broke, instead of emailing a screenshot.
• A weekly operating review moved onto the certified model; slide assembly stopped being a role.

Outcomes

Illustrative only. We would not invent a percentage improvement in forecast accuracy or margin.

Lessons

• Five trusted numbers beat fifty unofficial ones.
• If finance is not in the contract review, the warehouse will be ignored at month-end.
• Snapshots can be modeled as events, but the limitations must be written down.