Interview · Supply reading

What the tools prove

Four tools delivered in production, read not as code but as supply decisions: which field problem, which lever, which measured impact.

The method, in four steps

Go see the teams: sales admin, logistics, purchasing. The real need is expressed in their words, never in a specification.

Frame the trade-off before the tool: total cost, target service level, what error costs. A tool with no quantified trade-off is useless.

Turn the need into software: the tool prepares the work, it never decides instead of the business.

The business keeps the final word on every sensitive decision. Humans arbitrate, the tool records.

Field problem: 52 trucks tracked on scattered spreadsheets, invoices filed by hand, breakdowns discovered too late. Reactive maintenance always costs more than planned maintenance.
Supply lever: Total cost of ownership per truck and per site, automatic breakdown extraction from invoices, replace-or-repair arbitrage (CAPEX vs OPEX).
Measured impact: €40,000 saved per year, 2h/week freed for the logistics manager, rollout planned across 41 companies.
What it proves in an interview: I reason in total cost, not purchase price, and I can turn a field constraint into an investment case.

Field problem: 2,500 references ordered on instinct, from Excel extracts rebuilt every week. Stockouts on what sells, overstock on what does not move.
Supply lever: Forecast measured on backtest (WAPE), p50/p90 scenarios, safety stock calibrated on the target service level. Every proposal is broken down and explainable.
Measured impact: Fewer stockouts on fast movers, less tied-up capital, zero manual rebuilding of the need.
What it proves in an interview: I arbitrate service against cost instead of suffering both, and I can defend a figure in front of management because it is measured, not promised.

Field problem: Inconsistent labels, mixed units, missing categories: every catalogue error cascades — wrong picks, skewed forecasts, e-commerce and WMS out of sync.
Supply lever: Master-data normalisation from a SAP export, reliability score per record, human validation before reimport.
Measured impact: Reliability measured before/after, fewer picking errors, consistency guaranteed across SAP, e-commerce and WMS.
What it proves in an interview: I know data is a strategic prerequisite: no WMS, no AI project holds on a doubtful master.

Field problem: Years of manual entry: duplicates, approximate addresses. Every doubtful record ends as a delivery error or lost sales-admin time.
Supply lever: Deduplication, cross-check against the SIRENE registry, human validation of corrections before SAP reimport.
Measured impact: Deduplicated and validated master, less manual rework in sales admin, a sound base for segmentation and reporting.
What it proves in an interview: I connect data quality to its end effect: the service level delivered to the customer, not an abstract score.