What we do

One path. Five steps. You can stop after any of them.

Most AI projects fail because the first commitment is the largest one. Each step here is priced on its own and produces something you can act on. You go further only when the previous step has proved itself.

A napkin sketch of the five steps: talk, answer in 2 to 3 weeks, proof in 8 to 12 weeks, roll-out in 3 to 9 months, keep it sharp.

The seven things we do are spread across five steps you buy one at a time.

Step	What happens	How long	What you get	What it costs	What you bring
Talk	A 45-minute conversation about the job, the machine, the data and what is at stake.	One call.	A clear yes, no or not yet.	Nothing.	The person who owns the problem.
The answer Architecture and design Model selection	We look at your job, your data and your hardware and work out which model fits, what it will cost to run, and how long it will take.	2 to 3 weeks.	A written recommendation your board can read, with the technical detail your engineers need in the appendix.	A fixed price we tell you before we start.	Sample data, two hours of one engineer's time per week, one decision maker for two meetings.
Proof Model selection Routing Training and tuning Deployment	We build the model, train it on your examples and install it on one line, one vehicle or one device. We measure it against a target we agreed in writing.	8 to 12 weeks.	A working model in your environment and a results report.	A fixed price or milestones, agreed after The answer.	Access to the line or device, the same engineer, the same decision maker.
Roll-out Deployment Hosting and operation	We take what worked and put it on every line, every vehicle or every device, with your team learning as we go.	3 to 9 months, depending on scale.	The model everywhere it is needed, your team trained, everything documented.	A project, scoped after Proof.	Your operations people and your IT.
Keep it sharp Hosting and operation Continuous optimisation	We watch the model, retrain it when the world changes, patch it, and report to you monthly.	Ongoing.	A model that stays as good as the day it launched, and a monthly one-page report.	A monthly fee.	Ten minutes a month to read the report.

Talk

What happens: A 45-minute conversation about the job, the machine, the data and what is at stake.
How long: One call.
What you get: A clear yes, no or not yet.
What it costs: Nothing.
What you bring: The person who owns the problem.

The answer

Architecture and design
Model selection

What happens: We look at your job, your data and your hardware and work out which model fits, what it will cost to run, and how long it will take.
How long: 2 to 3 weeks.
What you get: A written recommendation your board can read, with the technical detail your engineers need in the appendix.
What it costs: A fixed price we tell you before we start.
What you bring: Sample data, two hours of one engineer's time per week, one decision maker for two meetings.

Proof

Model selection
Routing
Training and tuning
Deployment

What happens: We build the model, train it on your examples and install it on one line, one vehicle or one device. We measure it against a target we agreed in writing.
How long: 8 to 12 weeks.
What you get: A working model in your environment and a results report.
What it costs: A fixed price or milestones, agreed after The answer.
What you bring: Access to the line or device, the same engineer, the same decision maker.

Roll-out

Deployment
Hosting and operation

What happens: We take what worked and put it on every line, every vehicle or every device, with your team learning as we go.
How long: 3 to 9 months, depending on scale.
What you get: The model everywhere it is needed, your team trained, everything documented.
What it costs: A project, scoped after Proof.
What you bring: Your operations people and your IT.

Keep it sharp

Hosting and operation
Continuous optimisation

What happens: We watch the model, retrain it when the world changes, patch it, and report to you monthly.
How long: Ongoing.
What you get: A model that stays as good as the day it launched, and a monthly one-page report.
What it costs: A monthly fee.
What you bring: Ten minutes a month to read the report.

What we need from your machine.

Before anyone can give you a straight answer, five things have to be on the table. If you have them, the first call is short and useful. If you do not, the first call is about getting them, and that is fine too.

What the job actually looks like. Pictures or recordings taken by the camera or sensor that will really be used, in the light it really has, at the speed the line really runs. Not marketing photos.
Examples of good and not good. Labelled by someone who knows the process, because the model learns your judgement, not ours. How many you need depends on the job, and we will tell you the number on the first call rather than guessing here.
What the machine already has. Camera model and where it is mounted, which sensors are on it, which controller runs the line or the vehicle, and how we would be allowed to talk to it.
Where a small computer can sit. A place near the machine, the power and network that reaches it, and what your IT and security people will and will not permit.
The decision and the clock. What the model has to decide, how fast it has to decide it, and what it costs you today when a person gets that decision wrong.

You do not need to have all five ready. You do need to know who in your building can answer each one.

Where it can stop.

After The answer, the recommendation is yours. You can build it with us, with your own team or with someone else. We would rather earn the next step than lock you in.

What we will not do.

We will not sell you a platform.

We will not start with a roll-out.

We will not put your data anywhere you have not approved in writing.

The full deliverables for each step are on Under the hood.

Start with a conversation.

45 minutes, no slides, an honest answer.

Get an answer