How we work

Five stages. Five exits. One system that works.

Most AI projects fail quietly, months in, after the money is spent. Ours is built so that cannot happen: every stage ends with something you keep and a decision you make. Stop at any gate and you still walk away ahead.

The path

Metric before model

We agree in writing what 'it works' means before a line of code. A model nobody can measure is a model nobody can trust.

An exit at every gate

Each stage ends with a go or no-go that is yours to call. No stage assumes the next one, and no retainer is hiding in the fine print.

You own everything

Code, trained models, data pipelines and documentation are yours from day one. If we stop, nothing stays with us.

Frame

Week 0 · free

What happens

A 30-minute call with the engineers who would do the work, not a sales team. We talk about the problem, the data you have and what would have to be true for this to be worth it. Then we write it down.

What you get

A one-page brief: the problem, the users, the success metric
An honest first read on whether ML is even the right tool
An NDA signed before any detail changes hands

The gate

Is this worth two weeks of audit?

If you stop here

The brief. Use it with us, with another team, or on your own.

Prove the data

Weeks 1–2

What happens

Before anyone promises accuracy, we check what your data can actually support: volume, quality, labels, leakage and gaps. We build a simple baseline on it, so the next number we quote is measured, not guessed.

What you get

A data audit report in plain language
A baseline result against the agreed metric
A fixed scope, price and timeline for the prototype

The gate

Does the data support a prototype?

If you stop here

The audit and the baseline. If the answer is 'not yet', you also get the list of what to collect first.

Prototype

Weeks 3–8

What happens

A working model on your real data, measured against the metric from stage one. You see a demo every week, inside the tool you will actually use, so the first time you meet the result is not the last week.

What you get

A working prototype your team can try
A measured result against the success metric
A production plan with integration, cost and risks

The gate

Does the result justify production?

If you stop here

The prototype, its code and the trained model, ready for any team to carry on.

Productionise

Months 2–4

What happens

The prototype becomes a system: an API or batch job that puts predictions where decisions are made, with monitoring, security and tests around it. We hand it over with documentation your engineers can run without us.

What you get

Deployed in your cloud or ours, your choice
Monitoring for accuracy and data drift
Documentation, handover sessions and full ownership

The gate

Is the system accepted and in use?

If you stop here

Everything: the code, the models, the infrastructure definitions. There is no lock-in to leave.

Keep it useful

Ongoing · optional

What happens

Models age as the world changes. If you want us around, we watch for drift, retrain on a schedule and build the next feature on the same foundation. Or your team does it with the runbook we left.

What you get

Retraining and a response plan for drift
New features on the existing system
Support hours you actually use, not a fixed retainer

The gate

Is it still earning its keep?

If you stop here

A system that keeps running. Support is month to month.

Inside a week

What working with us feels like

One demo every week

Real progress in the real tool, never a slide about progress.

One shared board

You see what is planned, in progress and blocked, at any hour.

One business day

The longest you wait for an answer from us, from first email to final handover.