Services: how we work with you

We work in four steps: Discover, Pilot, Build and Run. You can start small, stop after any step, and see a working pilot before you commit to a full build.

Last updated 8 Nov 2026

We work in four steps: Discover, Pilot, Build and Run. You can start small and stop after any step. Each step has a fixed scope, a timeline range and a clear deliverable.

The four steps

Step 1: Discover
2–3 weeks
We agree the problem, the people who will use the tool and how we will measure success. We look at your data and the systems the tool must work with.
You receive: A short scoping note with the plan, the risks and a price range.
Step 2: Pilot
6–10 weeks
We build a small working version for one task and test it with a small group of real users. A pilot, not a platform.
You receive: A working pilot and a written evaluation of what worked and what did not.
Step 3: Build
3–6 months
We turn the pilot into a production tool, with the integrations, access control, privacy notices and tests it needs.
You receive: The production tool, its documentation and a handover session.
Step 4: Run
Ongoing, by agreement
We host, monitor and improve the tool. We use the same health checks and support process that we use for our own applications.
You receive: Monitoring, support and regular updates. Support hours: Monday–Friday, 09:30–18:30 IST, excluding Indian public holidays, with urgent issues raised on the support ticket channel, and a monthly service report.

Timelines are ranges. They depend on your data, the systems we connect to and how quickly decisions are made. We confirm them in the scoping note.

What buyers usually ask us

Have you run AI tools in production?

Yes. We build and run our own applications in production, mostly for our group companies. You can read about them on our work, where we list a live app only while its health check passes.

How do you check that the AI gives good answers?

We agree how to measure quality in Discover, test the pilot against real examples, and report what worked and what did not. We score a fixed test set agreed at kick-off, typically 100–300 real but anonymised items, before and after, with your reviewers and an agreed rubric, and report accuracy, time per item and reviewer agreement.

What happens after launch?

In the Run step we host, monitor and improve the tool. We use the same health checks and support process as for our own applications.

Do you have independent client reviews?

Most of our published work is for group companies, and we say so on every case study. We will publish independent reviews once client projects close and the clients agree.

Who owns the code and the data?

We agree this in writing before any work starts. Our standard terms: you own the custom code and configuration we build for you, and all of your data, once invoices are paid; we keep our pre-existing tools and libraries and give you a perpetual licence to use them with the delivered system.

Is this a fit?

A good fit if

You have a clear task that takes your team many hours, such as screening documents or answering repeat questions.
You can share sample data or documents for a pilot.
You want to see a working pilot before you commit to a full build.

Not a fit if

You need a general strategy report with no build.
You expect a fixed accuracy figure before any testing.
You want an off-the-shelf product with a self-serve checkout.

Start a project

Tell us what you want to improve. A person reads every request. You get an indicative scope and price range, confirmed after a call.

Start a project See pricing

The four steps

Step 1: Discover

Step 2: Pilot

Step 3: Build

Step 4: Run