How we work

Small steps, in the open

We don't disappear for months and come back with something. You see the work as it happens, and you can stop it at any point.

We learn how you actually work

Not how the org chart says you work. We sit with the people doing the job and follow the real path a task takes — where it waits, where it gets re-typed, where someone has to remember something that isn't written down.

Most of what we need comes from this stage. It's also the stage that needs the most of your time, and we try to keep it tight.

We find where AI actually belongs

Very little of a business needs AI. Some of it needs a better process, and some of it is fine as it is.

We come back to you with what we found: what's worth doing, what isn't, and what a simpler fix would solve faster. If nothing on the list is worth building, we'll tell you that too.

Often the honest answer is that something already exists. A tool on the market already does what you need — or one you're paying for every month does it, and nobody has switched it on. When that's the case, we'll point you at it instead of building our own version. And if you'd like help setting it up properly and getting your team using it, that's work we're glad to do.

We build the smallest useful piece

Not the full system. The one piece that proves whether this works for you.

Small enough that if it turns out to be wrong, you've lost very little — and real enough that you can actually use it, not just watch a demo.

You use it

This is the part that matters. Your team puts it into their actual day and finds the things nobody predicted — the edge cases, the awkward inputs, the step that made sense in a meeting and doesn't make sense at a desk.

We expect this. It's why we build small first.

We improve it while it's running

Based on what you found, not on what we assumed. Then, if it's earning its place, we widen it.

Each round is a decision point. You choose whether to keep going.

Why there are no timelines on this page

You'll notice we haven't put weeks or months against any of these stages.

Some of this work is genuinely quick. Some of it isn't, usually because of what's sitting inside your existing systems. We don't know which it is until we've looked properly — and we'd rather say that than invent a date to win the work.

What we will do is tell you honestly what a piece of work involves once we understand it. You decide from there.

Your data

Your data stays yours. We don't reuse it, we don't sell it, and we don't train anything on it. How and where it's stored is your decision, and we'll set it up that way.

You choose the model. Open-source models running on infrastructure you control, so nothing sensitive leaves your environment. Or a commercial API, which is usually simpler and cheaper to run. Both are legitimate. They trade off differently on cost, control and effort.

We explain the trade-off, you make the call. This isn't a decision we make quietly on your behalf and mention later.

No hidden dependency. We build on standard tools and document what we've done, so your team isn't left guessing. We'd rather you stay because the work is good.

What we won't do

Put AI somewhere it adds nothing, just to have AI in it

Build something ourselves when a ready-made tool already does the job

Quote a timeline before we understand the problem

Ask your team to abandon the tools they already use, unless there's a real reason

Keep going with something that clearly isn't working

Not sure where you'd even start?

Neither are most people at this stage. Tell us what's slow or frustrating, and we'll tell you honestly whether there's something here worth doing.

Start With a Conversation