Eight contract sources consolidated — and the silent data loss stopped

The monthly contract sales report was rebuilt by hand from eight group purchasing organisations, each with its own layout and fee rules, taking around a week and a half to assemble and verify. Worse, rows that failed to match a customer record vanished without trace. Both problems are now closed.

The client

US medical-device manufacturer. Published anonymously at the client’s request — no company name and no person’s name appears in this entry.

The problem

Eight group purchasing organisations each sent sales data in their own format: different layouts, different column names, different fee structures. Every month the report was rebuilt from scratch.

Each report and each query had to be checked and verified by hand, taking roughly a week and a half of the month by the client's own account.

The serious problem was quieter than the slow one. A row that failed to match a customer record simply disappeared from the report. Nothing warned anyone, so the output looked complete and nobody knew the figure was short.

What was built

• A pipeline in n8n with Python and Google Sheets that consolidates all 8 GPO sources.
• Each contract's own fee logic applied in code — defined once per contract instead of remembered each month.
• Unmatched rows flagged for review and never dropped, so an incomplete report announces itself instead of looking finished.
• A repeatable monthly run, so the output is reproducible rather than reassembled.

Before and after

	Before	After
Monthly assembly and verification	About 1.5 weeks	Near-zero hands-on time
Sources reconciled	8, by hand, each in its own format	8, read automatically
Rows that fail to match a customer	Dropped silently	Flagged for review, never dropped
Fee calculation	Applied by hand, per contract, every month	One rule per contract, applied in code

How each figure was measured

Monthly assembly and verification

The before figure is the client's own account of the manual cycle, not an instrumented baseline. The after figure is the scheduled run plus review of whatever rows the pipeline flags.

Sources reconciled

Count of GPO sources in the pipeline. Each has its own layout, column names and fee structure, handled per source rather than normalised away.

Rows that fail to match a customer

Unmatched rows are written to a review list and counted in the run log, so the report states what it could not place instead of omitting it.

Fee calculation

Where the fee logic lives: in someone's head and a spreadsheet, or in the pipeline.

What is deliberately not here

The time saved is the smaller of the two results. A report that is wrong and looks right costs more than a report that is merely slow.

The other engagements

• Eleven data sources in one dashboard — and stale numbers labelled stale
• A seven-day reporting cycle became a scheduled evening job
• A 40-node content pipeline that runs without supervision