Understand the problem. Follow the process. Check the information underneath it. Automate what rules can handle. Add AI where it genuinely earns its place. Then hand the whole thing over.
Skip a step and it holds together for about six months. Automating a process nobody has agreed just produces the wrong answer faster and more consistently.
Plenty of organisations only ever need one of the three, and that is completely fine.
| Stage | What it means | What you get |
|---|---|---|
| Consult | We understand the work before recommending the technology, and prioritise the opportunities by value, effort and risk. | A map of how the work really happens, a systems and information review, an opportunity register, and a prioritised roadmap you own outright. |
| Build | A prioritised opportunity becomes a working improvement — automation, AI, or both, depending on what the job needs. | The smallest useful version, tested with real examples and real users, measured against an agreed baseline, documented and handed over. |
| Sustain | The improvement becomes part of the organisation rather than a tool that quietly stops working. | Documentation, handover, adoption support and clear ownership — plus development routes for your people where there is a genuine skills need. |
Not a values page. These are the rules that decide what we recommend, what we refuse and how a build is designed.
We do not begin by asking where AI can be inserted. We begin by asking where work is expensive, slow, repetitive, inconsistent, frustrating or difficult to scale.
The best solution is the least complex system that reliably produces the outcome. Sometimes that is AI. Sometimes it is automation, a dashboard, a process change or an agreed definition.
A dataset can be good enough to show a broad trend and nowhere near good enough to calculate pay. The question is always: good enough for this use, given what happens if it is wrong?
Accurate data still misleads when definitions and business rules have never been agreed. AI needs organisational context as much as it needs access.
For every use case we decide whether technology should inform, assist, recommend, decide or act. Capability is not consent.
Consequential, uncertain or externally visible outputs get proportionate human review, an escalation route and an audit trail.
The baseline, the outcome and the owner are agreed before the build. Time released, error rate, cycle time and adoption beat a vague transformation claim.
Documentation, ownership, training and support are designed with the build, not added at the end when the budget is gone.
Minimum access required, credentials and personal information protected, testing done safely, data paths documented, and explicit approval before anything touches production.
Technology can take on more or less of any piece of work. Deciding how far it should go is a business decision, not a technical one, and it belongs to you.
Surface information for a person to use.
Gather, check, calculate, summarise or prepare.
Form a view and suggest an action.
Make a defined decision within agreed boundaries.
Take an action on someone's behalf.
What a system can do and what your people would willingly hand over are two different questions. The second one is usually the real constraint.
This is what separates a demonstration from something that still works next year.
| Risk | Control |
|---|---|
| Someone oversells you on what AI can do | We fix the process first, and only reach for AI when it demonstrably beats the simpler option. If you do not need it, we will tell you. |
| The work grows, and so does the bill | Scope, assumptions and success criteria are written down and agreed before anyone starts. Changes are quoted, not slipped in. |
| AI running costs get away from you | We monitor usage, pick the cheapest model that does the job properly, and review it regularly so the bill does not surprise anyone. |
| You need something bigger than we can deliver | We say so early and point you at someone who can, rather than taking the work and struggling through it. |
| Sensitive information ends up somewhere it should not | We agree the data, access, purpose, controls and responsibilities before delivery. High-risk work may need a DPIA or a more restricted approach. |
| The apprenticeship route feels like an upsell | We raise it only where the work reveals a genuine skills gap. If your team can already run it, we will say nothing. |
| We take on more than we can support | We start small, are honest about our limits, and bring in partners rather than stretching ourselves thin. |
| The build works, but nobody uses it | Adoption is designed in from the start: real users in testing, documentation, training and a named owner on your side. |
You do not need a finished brief or a preferred technology. Tell us what is slow, repeated, difficult to trust or harder than it should be. We will arrange a short conversation and tell you whether we think there is a useful next step.