JJ Signal · Benchmark · 7 min read
What demand forecasting is actually worth to a 50-person retailer
Across our retail engagements the recoverable number was working capital, not revenue — and it showed up in the first quarter.
Published Retail & E-commerce
For an independent retailer under about 250 staff, demand forecasting pays for itself through released working capital long before it pays through incremental sales — which means the usual business case, written around revenue uplift, understates it and points at the wrong metric.
The evidence
Drawn from our own delivery work. Each claim states its sample and its method, including where the sample is small.
- 01Stock-out reduction clustered around 30% where the sales history could be reconciled to one product identity.Four retail engagements between 2026-01 and 2026-08, 180 to 600 active SKUs each. The cluster is tight but four is four; treat it as a shape, not a rate.
- 02Cash released from slow-moving cover exceeded the first-year fee in every case; incremental revenue did not.Measured as change in holding value on lines turning under three times a year, at two quarters post go-live. Working capital is a one-time release, not an annuity, and a business case should say so.
- 03Reconciling product identity took longer than building the model, in all four.Median eleven days on identity reconciliation against six on the forecast itself. Where online and in-store SKU codes had drifted, that gap widened rather than closed.
What follows from it
If you are evaluating forecasting on projected revenue uplift, you are measuring the smaller and less certain half. Ask instead what is sitting in lines that turn twice a year, and whether your sales history can be reconciled at all — because if it cannot, the forecast is the second project, not the first.
What we would do
- 01Check first whether online and in-store share one product identity. If not, that is the engagement, and the forecast waits.
- 02Baseline holding value on lines turning under three times a year, before anything changes.
- 03Validate any model against held-out quarters it has never seen, and publish that score to the people who will use it.
- 04Ship the exception list before the forecast — the lines behaving unlike their history are useful on day one.
Get JJ Signal by email
Roughly monthly, and only when there is something measured to report. If a month produces nothing worth your attention, you get nothing.
What you are signing up to