Built into the systems you already have, live and in daily use, and run by your own people.
Tell us what you want AI to do, or the process you would point it at. We come back in writing within five working days with what we would build first. No charge, and no call unless you want one.
What AI inherits
An AI tool reaches across whatever your systems already let it reach. So the four questions that decide whether it works are not model questions. What data do we hold. Why do we hold it. Who can reach it. And when does it go.
Leave those unanswered and the gaps stop being a paperwork problem. The same unclear rule now runs a hundred times a day, across half a dozen systems, and nobody is reading it on the way past.
Answer them as part of the build, which is where they are cheap, and the same tool holds up when somebody asks how it reached that answer. It is why the work starts with your organisation rather than with a product.
The order of work
Services
Those four disciplines arrive as these five pieces of work.
Start here
Two to three weeks. Under half a day of your team’s time in total.
We find where the time and the risk go. We do not take the process document’s word for where that is. You get the processes costing you most, ranked, with the hours attached, and a shortlist of where AI would earn its place against each. You also get a first-ninety-days list with an owner on every item, and a straight answer on which of the four below you need.
The access map: who can open what today, and who no longer should. A retention position, including the parts of it that turn out to be silent. Every gap written up as something checkable, with a name against it.
Where AI would help, where it would not, and what has to be true before either. Then one use taken all the way live on your own systems. Whether that is your first or your fourth makes no difference to the method.
Repetitive work rebuilt, with AI and automation carrying the dull part. Role-based access, review points, a record of what happened, and an off switch that works. Where a job needs judgement across several systems, that can be an agent rather than a fixed flow.
Short handbooks and small-group sessions, built from your own controls and your own workflows. The people running it after we leave should be the people who work there.
Where it goes wrong
Switching Copilot on is an access review whether you planned one or not.
If everyone in the organisation can already open everything, an assistant has not created that exposure. It has just found it much faster, and put it in front of someone who was never going to go looking.
So the access model gets reviewed as part of switching the assistant on, not instead of it. The permissions were never right in the first place. What has changed is that there is finally something efficient enough to prove it. That is a fortnight of work rather than a reason to wait.
Reporting works the same way. If two teams hold different definitions of the same figure, automating the report will not settle the argument. It just publishes the argument faster, on better-looking paper.
Scope
The first step
Or send us one process and we will tell you. What the work is, what data it runs on, who touches it, what comes out. Within five working days you will have something in writing: what we would build first, and what has to be true before it runs. No charge, and no call unless you want one.
