Two weeks. A working team of two. It ends with a document, a ranked shortlist and a number, and you can take all three to another agency if you want to.
What we look at
- The workflows that cost the most human time today
- What data exists, where it lives and how clean it really is
- Your constraints: regulation, contracts, procurement, appetite for risk
- What is already automated, and why the last attempt stopped
- What a first release could realistically touch
What comes back
A shortlist of candidate projects, each with an estimated build cost, a running cost, a risk note and an honest confidence level. Ranked. Plus a second list — the ideas that came up and should not be built, with the reason written down, so they stop being raised every quarter.
The unglamorous finding
More often than not the top recommendation is not a model. It is fixing the data that a model would need, or replacing a manual step with an ordinary integration. We say so plainly. A review that always recommends the expensive thing is a sales call with a document attached.
Who needs to be in the room
The person who owns the process, the person who owns the data, and someone who can say yes to a budget. Two hours a week from each. Without the process owner the review produces a plausible plan that nobody can execute.
After the assessment
If you build with us, the work goes to the relevant page — usually LLM development services or agentic AI development. If you build elsewhere, the document still stands on its own.
The full AI map shows where each recommendation would land, and how an engagement runs is the same for this as for anything else we do.