From a clear problem to a live solution

We agree what success means before anyone writes code. Then we build straight to production and test the result against the way the job is done today.

01 / FIND

Choose the problem

We look at the decision being made, the data sitting behind it, and how your people handle it now. Out of that comes an agreed test, a fixed scope and a date.

Output
  • the chosen problem and its owner
  • data sources and access
  • the baseline
  • the acceptance test

Discovery is only charged when there's real uncertainty to clear up.

02 / BUILD

A live first version in six weeks

We build the data work, the model, the application and the connections into your systems. Scope and price are fixed before anyone starts.

Included
  • data and integrations
  • model and application
  • tests
  • monitoring
  • documentation and training
03 / PROVE

Compare it with the way the job is done today

We run it on real historical cases and put the result next to the baseline we agreed at the start, including the cases where it does badly.

What we report
  • result against the baseline
  • where the model is uncertain
  • error cases
  • go or no-go
04 / RUN

Support starts at go-live

We watch it, we handle the incidents, and we maintain everything it depends on. Support is a monthly service with a scope you have seen in writing.

Included
  • monitoring and alerting
  • incident investigation
  • data and model checks
  • compatible updates
  • usage and cost
  • quarterly review
05 / DECIDE

Choose the next step

Keep it, adjust it or stop. Anything beyond this first build gets its own scope.

What counts as new work

A new data source, workflow, interface or major change to business logic gets its own scope and price. See what's covered by support.

Agree the test before writing the code

The test records the current baseline, what we will measure, which data we will use and what result is good enough to launch.

CAPABILITY
Four-week demand estimate per product group
BASELINE
The planner's current forecast for the same periods
MEASURE
Mean absolute percentage error, weighted by volume
DATA
Held-out periods not used in training
THRESHOLD
Agreed with the owner before the build starts
DECISION
Launch, adjust or stop, based on the agreed result

What we need from you

  • an owner who can decide on the outcome
  • access to the relevant data and systems
  • a short weekly review during the build
  • a decision at the acceptance test

Have an owner and a decision to make?

If someone can own the outcome and the data is reachable, we can put together the acceptance test on the call.

Book a 30-minute call