Finding
The bottleneck is never the building
An agent that drafts quotations is built in two days. The question of whether it can reach the data it needs inside your business costs weeks. Almost every failed AI rollout fails on the second question, having started with the first.
Last updated: July 2026
01
Where the time actually goes
In a mid-sized business that has grown over time, the data is rarely where the org chart suggests it is. Part sits in the ERP, part in a spreadsheet on the network drive, part in a mailbox, and a not inconsiderable part in the head of a colleague who has known for fourteen years which supplier will move on which line item.
The model can process all of that. What it cannot do is ask for it. The work therefore does not consist of teaching the system to think, but of getting it access - and, before that, of finding out whether that access can be had at all.
The distribution surprises people regularly, and in both directions: businesses that are wary of the technology underestimate how far the models have come. Businesses that have already experimented with a chatbot underestimate what lies between a working prompt and a process that runs without supervision.
02
The most expensive mistake
The most expensive mistake is not a process that turns out not to pay. You notice that while doing the sums, and then you leave it. The most expensive mistake is a process that would have paid - but whose data nobody could reach.
That almost always only surfaces once the building has started. The interface the vendor lists in the data sheet does not exist yet in your version. The export runs, but delivers prices without the tiers. The drawings exist as scans, and specifically as scans of a fax.
A fixed price before the audit contains either a risk premium or an arithmetic error. The client pays for both in the end.
03
What an audit does instead
- Record the processes
- Not as they appear in the manual but as they run. The gap between the two is often half the finding already.
- Test access before it is needed
- We pull a real export, open a real file, call a real interface. A promise in a data sheet is not access.
- Prioritise the candidates
- Three to five processes, ranked by the ratio of effort to effect - with the figures from your business, not from a study.
- Say no where no is the answer
- If none of it is worth doing, that is what the result says. An audit that always leads to a project is not an audit but a sales appointment.
The audit costs €2,500 and is credited against the work as soon as we implement. That credit is not a discount but the removal of a conflict of interest: whoever earns from the audit has an interest in many audits. Whoever credits it has an interest in usable results.
Frequently asked
Why can I not simply get a quotation?
You can - a fixed price follows the audit. Before that we would have to guess which data we can reach, and that guess would go into the price as a premium. The audit costs €2,500 and is credited against the implementation.
How long does an audit take?
Two to three weeks. The effort on your side is about two half-days of conversation plus the access we need in order to test.
What happens if the audit finds that nothing is worth doing?
Then that is what the result document says, with reasons. For a mid four-figure sum you then have the certainty that you did not need a five-figure project.
Where it gets concrete
Use cases on this topic
Tell us where your time goes.
In the audit we record the process, work out what is worth automating, and you get a plan with fixed prices - credited against the build if you go ahead.