Pilot weeks 1 to 6 Baseline 20% Target ≤16%
- Baseline
- 20%
- Target
- ≤16%
- Window
- 6 weeks
Pilot delivery
A pilot with an agreed finish line
Baseline, evaluation period and acceptance conditions are signed by your operations owner before the first forecast runs.
Gate before week 1Sampled issue rate ≤1%
Signed by your operations owner
The pilot runs on a fixed clock
Six weeks agreed in advance. Every weekly run is scored on the same held-out dataset, so the finish line never moves.
-
Week 1
Setup and baseline lock
Connect the repaired feeds, freeze the held-out weekly demand dataset and lock the baseline.
Baseline 20% locked
-
Weeks 2 to 5
Weekly evaluation runs
One scored run a week. Each forecast is measured against the baseline and the target, then logged with its inputs.
- Run 1W2
- Run 2W3
- Run 3W4
- Run 4W5
-
Week 6
Review with your owner
Every run, the evidence behind it and its limits, read against the conditions signed in week 1.
-
End of week 6
Go or stop
Your operations owner decides. We record it.
Hollis Supply · 3 warehouses · 12,000 SKUs
- History builds the model
- Held-out weeks score it
- Baseline
- 20%
- Target
- ≤16%
- Scored on
- Same 12 weeks
One dataset scores both numbers
Baseline and pilot are measured on the same held-out weeks, so a lower error means a better forecast, not easier data.
- 01
The held-out dataset
Twelve weeks of weekly demand set aside before any model work. The pilot never trains on them.
- 02
Weighted forecast error
Error per SKU, weighted by volume, so the fast movers count most. Today it sits at 20%.
- 03
Same weeks, both numbers
Score the pilot on different weeks and a 16% proves nothing. We keep the set fixed.
Five conditions, signed before the pilot
The proposed acceptance conditions from the Hollis case. Thresholds agreed in advance, never results.
Proposed acceptance conditions
Hollis Supply · six-week forecasting pilot
Thresholds, not results
Weighted forecast error
Same held-out weekly demand dataset, lower is better
4 pts to close
Gate before start
≤1%
Sampled SKU identifier issue rate
8% at assessment, 40 of 500 records
Measured on
Held-out set
Weekly demand, fixed in week 1
Baseline and pilot scored alike
Evaluation period
6 weeks
Forecasting pilot
Starts after the gate holds
Decision
Go or stop
Operations owner
Approves the evaluation scope
Go or stop, called by your owner
The week 6 review ends in one decision. Both outcomes are planned before the pilot starts, so neither one is wasted.
Week 6 reviewWeighted error against ≤16%
If the target is met
Go: scale the pilot
The forecast earns its next phase on evidence your owner has already read.
- Extend the forecast to more sites and SKUs
- Scope the next phase from the six weekly runs
- Keep the held-out set as a standing check
If it is not
Stop: keep what you paid for
A stop is a planned outcome. Nothing built for the pilot is left to write off.
- The repaired feeds and ownership register stay
- The repair plan holds for the next use case
- No sunk build beyond the six weeks
Signed by your operations owner. Datril records the decision and the evidence behind it.
Pricing
One fixed price, then a quoted pilot
The assessment is the only fixed price. Repair work and the pilot are quoted from what it finds, never before.
Readiness assessment
Start hereFor your operations owner, before any pilot.
€6,000fixed fee
two weeks
two weeks
- Readiness report
- Feed ownership register
- Prioritized repair plan
Repair
When the assessment finds feeds to fix first.
Quotedfrom your
repair plan
repair plan
- Feed reconciliation
- A named owner per feed
- Re-sample until the rate is ≤1%
Pilot
When the gate holds and the use case is viable.
Scopedand quoted once
the gate holds
the gate holds
- Six-week evaluation period
- Baseline and target signed first
- Your owner decides go or stop
No pilot price or outcome is promised before discovery.
Start with a two-week assessment
€6,000, fixed before it starts. A readiness report, a feed ownership register and a repair plan your owner can act on.