How we work
Audit, build, run
Small scope first, working software early, and a person in control of anything that matters.
Week by week
What actually happens
Timings are typical. The variable is how quickly access to your systems can be arranged.
Day 0 · The call
One hour, no preparation
You describe how a job runs today. We ask questions and say what we think, including when the answer is simpler than AI. You leave with a view on your best first project.
Week 1 · Audit
We follow the work
Short sessions with the people doing the job, not just the managers describing it. We watch the actual steps, count the actual hours, and note every system the work passes through.
Week 2 · Roadmap
A plan in priority order
Candidates ordered by payoff against effort, each with an estimate to build and to run, and a note on what it needs from you. We walk your team through it live. You own the document either way.
Weeks 3–6 · First build
One job, working, in your tools
We build the top candidate against real data and show it running early, usually in a limited or read-only mode first, so you can watch what it would have done before it is allowed to do it.
Go-live
Switched on for one team
Training, written guidance, and approval gates set where you want them. We watch it closely for the first weeks and fix what surfaces.
Ongoing
Run and extend
Monitoring, maintenance and a monthly review of what it did, where it handed back to a person, and what to build next. This is where the second and third jobs get chosen, and the roadmap already says what they should be.
Principles
How we make decisions
One job at a time
A system with a single, clear responsibility can be tested and trusted. Platforms that promise to do everything are impossible to verify and the first to be abandoned.
Working software early
You should see something real within weeks, not a status deck. Early contact with real data is what exposes the awkward cases while they are still cheap to fix.
Watch before it acts
Where we can, a new system runs alongside the manual process first and shows what it would have done. Trust is earned on real cases before anything is switched on.
Boring where it counts
Reconciliation belongs in a database, arithmetic in a calculator, rules in code. We use AI for the language and judgement, not for jobs that already have a reliable answer.
Escalate, never guess
Every system we build has a defined behaviour for “not sure”, and it is always to involve a person, never to produce a confident answer anyway.
Yours to keep
The systems we build for you are yours, including if you take operation in-house later. No lock-in through obscurity.
Control
Access, approvals and audit
Here is exactly how access works, so you can put it in front of whoever needs to approve it.
- Access is granted by you, through your own admin tools
- Each connection is scoped to the job, read-only wherever reading is enough
- You can revoke any access at any time, without our involvement
- Every run is logged: what it read, what it decided, what it changed
- What a system may see and what it retains is written down before you grant anything
Where your data goes
Work is processed by the AI provider chosen for that task, under business terms that exclude training on your inputs. For sensitive material we can keep processing inside your own environment using open models.
We hold no third-party security certification today. What we do instead is the list beside this: access you grant and can take back, read-only wherever reading is enough, and a log of every action taken on your behalf.
If you have a security questionnaire, send it before the first call. We would rather answer it early.
Working together
What we need from you
Projects stall on these four things far more often than on the technology.
One decision maker
Someone who can approve scope and say yes to access without a committee.
Time with the doers
A few hours with the people who actually run the process, spread across the audit.
Access, when it is time
Whoever administers your systems, available to grant scoped access without a six-week wait.
An honest definition of done
Agreement on what a good result looks like, so success is a fact rather than an opinion.
And what we commit to in return. Success criteria are written down before the build starts, so whether it worked is a fact rather than an opinion. Every action a system takes on your behalf is logged and attributable to the job that caused it.
Start with the audit
Two weeks to a prioritised plan you own.