Replenishment was one buyer’s judgement

Overstock and stockout look like opposite problems and get treated as two projects. They are not. They are the same estimate landing on either side of correct, which means any fix aimed at one of them makes the other worse — carry more stock and expiry rises, carry less and urgent buying rises.

Drawing them as one decision rather than two flows is what reframed the brief. The question stopped being how much stock should we hold and became what should the buyer be looking at when they decide.

Five things were missing — a demand signal, a replenishment policy, shelf-life visibility, measured supplier reliability and a production-usage feedback. Individually each looks like a small housekeeping gap. Together they are one thing: no operating model connects demand to inventory to supplier to production.

This is the finding that determined scope. Fixing any single gap alone changes nothing measurable, which is the honest reason a spreadsheet upgrade was never going to be enough — and equally the reason the ERP scope had to stay narrow rather than sprawl into full digitalisation.

The client had already decided they wanted SAP. Scoring it anyway was the point — a recommendation that agrees with the client is only worth something if it could have disagreed. Enhanced spreadsheets score 35, a standalone inventory app 67, SAP Business One 96.

Note where the losing option wins: implementation effort, where the spreadsheet scores 5 and SAP scores 3. Publishing the criterion the recommendation loses on is what makes the other seven believable, and it is what an assessor looks for.

The target state is more complex than what it replaces, and pretending otherwise would be dishonest. What keeps it usable for a ten-person business is that each individual action stays simple: receive against the order, record the batch, pick the oldest, post the issue.

The dashed return line is the part that is easy to leave out and fatal to leave out. Finance feeds planning, so last month’s variance and supplier performance change next month’s reorder parameters. Without it this is a longer version of the old process, and it drifts back inside a quarter.

This is the mechanism that makes the whole case work, in one diagram. A finished batch explodes into its components, MRP subtracts what is already on hand and on order, and only the shortfall gets bought — before production starts rather than after the shelf is found empty.

The return leg matters just as much. Two kilos over on a batch is nothing; two kilos over on every batch is a line item. Posting actual consumption against the order that planned it is what drags that number into the light, attached to a date and a name, instead of dissolving it into an aggregate purchase total where no one can see it.

No, and treating them equally is how control systems die. A high-value volatile spice needs a weekly conversation. A box of labels needs a minimum, a maximum and to be left alone. Policing the bottom-right cell costs more in admin than it ever recovers in stock.

The right-hand column is the part that outlives the project: who owns the reorder parameters, who owns stock accuracy, who approves an exception. Parameters set once at go-live and owned by nobody are stale within a quarter, and then the buyer quietly goes back to estimating.

A system can be fully configured and still leave the original problem untouched. The defence is writing the acceptance test at the same time as the requirement, in business language: receive one spice lot in two batches, consume 53 kilos against a 50-kilo recipe, deliver late and watch the supplier score move.

There is no results column on that diagram, and the omission is deliberate. Designing the acceptance criteria was my work and it is finished. Running them belongs to whoever implements. A pass I never observed is the easiest claim in a portfolio to check and the worst one to be caught on.