Do we have to start with the audit?

No, and about a third of clients do not. If you already know which process to automate and have the data access sorted, we can go straight to a sprint. The audit earns its place when there are several candidates and no agreement on which matters most.

How much of our team's time does this need?

For the audit, roughly eight to twelve hours total across process owners plus read-only system access. During a build, a couple of hours a week from someone who knows the process, and more during the hypercare period. We work around your calendar rather than demanding a workshop week.

What happens if the accuracy target is not met?

We keep working on it — the target is what we committed to, and a fixed-scope engagement means fixed. If the target turns out to be unreachable for reasons visible in the data, we tell you early with the evidence and either re-scope or stop. What we will not do is ship something below target and describe it as done.

Who owns the code during the engagement?

You do, from the first commit, in your own repository and your own cloud account. There is no stage where the work lives somewhere you cannot reach it.

Can you work with our existing delivery process?

Yes, and we prefer to. Your Jira, your review standards, your release process. Imposing our own would make handover harder, which is the opposite of the goal.

How we work

Narrow scope, measured accuracy, real handover

The pattern that fails is a pilot that handles the happy path and impresses a steering committee. The pattern that works is one workflow automated completely, then repeated. This is what that looks like week by week.

Delivery

Ten weeks from conversation to operating system

Timelines are indicative — a bigger scope takes longer. The sequence does not change, because each phase produces the input the next one needs.

1
A conversation, not a qualification call
Week 0
Thirty minutes with an engineer. You describe the process; we say whether automation helps and where it would break. If we are not the right fit you hear it here rather than after three meetings.
2
Audit — find out what is actually worth doing
Weeks 1-2
We interview the people doing the work, trace the data, and model the economics. Output is a ranked backlog with effort, risk and projected savings against every item — including the rows that say do not automate this.
3
Scope one workflow, narrowly
Week 3
One workflow, one owner, one number that moves. We write down what correct looks like and what accuracy is acceptable, and get the risk function to agree it before anyone writes code.
4
Build with evaluation from the first commit
Weeks 3-5
The labelled test set exists before the agent does. Every prompt and retrieval change is scored in CI, so improvement is provable and a regression fails the build rather than reaching users.
5
Ship behind a guardrail
Week 6
Live traffic with confidence thresholds routing uncertain cases to a human queue. Autonomy increases only as the accuracy data earns it. Cost per successful task is instrumented from the first request.
6
Hypercare, then a decision
Weeks 6-10
Thirty days of close monitoring while real inputs find the gaps. Then you choose: your team takes it with documentation and training, or we operate it under a retainer. What we will not do is leave it unowned.

Commercial models

Three ways to engage

Any of our twelve services can be delivered under any of these. Which fits depends mostly on how well-defined the problem is.

See pricing

Fixed-scope project

A defined outcome, a fixed price and a date. Best for audits, pilots and single-workflow automations.

Best for: Clear problem, known scope

Managed retainer

A monthly subscription covering operations, on-call and continuous improvement of what we built.

Best for: Running systems in production

Dedicated pod

An embedded squad — AI engineer, platform engineer and a lead — working as part of your team.

Best for: Ongoing roadmap, multiple workstreams

Ground rules

How we behave during delivery

These are the commitments that most affect whether an engagement is pleasant, and they are the ones worth checking against our references.

The awkward 15% is priced in

Edge cases are usually 60% of the engineering. Any proposal that prices only the happy path has priced about a third of the job, and that gap is where fixed-price projects go wrong.

We stop rather than absorb scope

If discovery shows the requirement is materially different from what was described, we stop and re-quote. No quiet absorption, no surprise change order at week three.

Weekly, in writing

A short written update every week: what shipped, what the accuracy number is, what is blocked and what we need from you. Not a deck, and not a status meeting nobody prepares for.

Your engineers in the loop

We work in your repositories, your review process and your standups where you want that. Handover is much cheaper when it has been happening all along.

For managed engagements, the operating side is covered by our service levels, and the exit commitments are in our terms.

FAQ

Questions we get asked

Let's find out what is actually automatable

Bring a process that annoys you. In 30 minutes we will tell you whether AI helps, what it would cost, and where it would fail — even if the answer is don't bother.

Book a consultation Get a quote

Or email [email protected] · we reply within 1 business day