Case 03 · Customer rebate lifecycle · end to end ownership
A rebate programme is a contingent liability that grows every time a customer buys something. Run it on email and spreadsheets and you do not find out what you owe until somebody asks for it. This one was rebuilt so the exposure was visible in the month it was created.
The programme ran across email threads, a shared drive and several spreadsheets. There was no single view of which customer sat on which tier. Volumes were taken from what partners reported rather than from what the business had measured. Contract versions existed in more than one place with no authoritative copy.
The financial consequence was the important one: the accrued liability was discovered at settlement. Finance learned the size of an obligation at the point the invoice arrived, which is the latest possible moment and the one with the least room to respond.
It was assigned to me as a reporting problem. It was not a reporting problem.
You cannot report your way out of a process whose source of truth is an email thread. The report will be accurate about the wrong thing.
Building a dashboard on top of this would have produced a confident view of unreliable inputs — the most dangerous possible outcome, because it would have looked like control.
The constraint was that the process had to be fixed first, while the programme kept running. There was no window in which rebates could be paused, so every change had to be made to a live commercial arrangement with customers who were mid-contract.
I took the lifecycle end to end rather than at the reporting edge. Contract drafting moved to a controlled template with redlines tracked and electronic execution, so there is one authoritative signed copy with a date on it.
Tier and threshold configuration moved into the ERP, where the transaction actually happens, rather than into the BI tool or a spreadsheet. Partner sales data was integrated so that volume is measured against our own records instead of accepted from a report.
On top of that, monthly accrual generation and settlement reconciliation — so the liability appears in the month the exposure is created and the settlement is checked against it rather than trusted.
Every one of these had a defensible alternative. The trade-off is stated because the trade-off is the point.
The request was for visibility. Delivering visibility over uncontrolled inputs would have created false confidence. I rebuilt the lifecycle first and reported on it second, which delayed the visible deliverable and was the right order.
Cost: weeks of work with nothing to demonstrate, which needs explaining upward as it happens.
The BI tool was the convenient place to hold tier configuration and the wrong one. Configuration belongs where the transaction is priced, so that the rebate is correct at the point of sale and not merely correct in a later report.
Cost: ERP configuration changes are slower and need a change process the BI tool did not.
Partner-reported volume is a claim. Integrating partner sales data and reconciling it against our own records turned the accrual from an estimate based on a claim into a calculation based on evidence.
Cost: a new integration to maintain, and some uncomfortable early conversations about variances.
I documented for a successor from the first week rather than at the end — runbook, configuration rationale, the reasons behind each threshold. This was deliberate: a programme only one person can run is a liability regardless of how well it runs.
Cost: continuous documentation overhead during build, when it feels least urgent.
The handover worked because I documented from week one. What I did not do well was set an explicit end date on my own ownership. Absent one, the programme stayed with me far longer than the work justified, and it crowded out higher-value projects for several quarters.
The lesson generalises beyond this project: taking end-to-end ownership of an operational programme is how you learn a business properly, and it is also how a technical role quietly becomes an administrative one. I would now agree a transfer date at the point of accepting the work, not after the work is stable.