Don't build what you haven't measured.
We map the work, the systems, the data and the real cost of one process. You leave with a recommendation and a plan you can start building.
There will be no demo.
A demo runs on hand-picked data, with no access rules and no exceptions. It proves a model can write, not that a workflow will hold up in your business. We measure your real process, then build the smallest version that can be judged on your own numbers.
How do we start?
Every stage ends in a decision: continue or stop. You can walk away after any of them.
-
Diagnostic
We tell you whether this is a case for AI, a simpler automation, or something that should stay human.
-
Process audit
We map the work, measure what it costs, and design the smallest version worth building.
-
Production pilot
One workflow built and integrated, with a measured baseline and a decision at the end.
-
Managed operations
We monitor the workflow in production, handle exceptions and extend it as your rules change.
The audit fee comes off the build. Setup starts at €2,900, then €590 a month.
What's in each stage?
The diagnostic decides whether there is anything here. The audit says what to build, at what scope and at what price.
The diagnostic
- One operational problem, explained in your own words
- Questions about volumes, systems and consequences
- An honest view on whether AI belongs here at all
- A shared decision: is the audit worth doing
It's not an audit, not an architecture, and not free consulting.
The process audit
- Interviews with the people who actually do the work
- A map of the current process, manual hand-offs included
- A measured baseline: volume, time, cost, errors
- An assessment of the data, the integrations and the risks
- The target workflow, the pilot scope and a fixed price
The audit may conclude that AI is not the answer. That's a useful result.
Why first projects fail
Companies start with the idea that impresses, not with the process that best balances value and risk. Six causes come up almost every time.
-
Starting from the technology
The tool was chosen before anyone agreed on the problem.
-
No baseline
Nobody measured the current process, so nobody can prove the new one does better.
-
No owner
Nobody inside the business owns the workflow once it's in production.
-
No exception handling
It works on the normal case and collapses on the remaining twenty per cent.
-
No approval path
Either a human checks everything forever, or nothing gets checked at all.
-
No plan after go-live
The system drifts, nobody maintains it, and by month three the team is back to the old way.
How do we rank an opportunity?
On six axes, never on how impressive a demo looked. The highest-value process is rarely the right one to start with: a big gain on poor data is exactly how a first project fails.
| Axis | What we look at |
|---|---|
| Business value | Time tied up, the delay the customer feels, the risk created. |
| Frequency | How often it happens, and whether there is enough repetition to justify building. |
| Data readiness | Whether the information exists, whether access is reliable, and whether it is accurate enough to trust. |
| Actionability | Whether the system can take a real action, or only produce text that a person then rekeys. |
| Measurability | Whether quality can be measured, and whether we can test before widening autonomy. |
| Risk | What a mistake costs, and whether it can be undone. |
What people ask us about delivery
- What happens when the system isn't sure?
- It asks for clarification, returns nothing, or hands the case to a person, depending on confidence and risk. Saying nothing is often the correct behaviour.
- How do you prove the value?
- Before building, we record the starting position and agree the metrics. The pilot is then judged against that baseline, not against how impressive a demo looked.
- How do you handle sensitive data?
- The architecture follows the use case: private instances, European hosting, self-hosting, access controls, audit logs. Nothing you hand over trains a third party's model.
Bring a process, not an AI idea.
Twenty minutes is enough to know whether an audit is worth doing. If the answer is no, you'll have heard it before you pay anything.