1. What we actually solve
Five situations we are asked about over and over, phrased the way the people living them phrase it:
- “We have tried several things with AI and none of them reached production.” Pilots that worked in the demo and died on contact with the real process, usually because nobody had looked at how that process actually runs.
- “The board is asking for an AI strategy and we do not know where to start.” The pressure is there, so is the budget, and what is missing is the criterion to put things in order.
- “Every department wants its own tool and nobody is looking at the whole.” Three separate subscriptions, no shared data and a monthly bill that keeps growing.
- “We have three proposals in front of us and no way to tell which one makes sense.” They all promise the same thing and none of them says what you need to have in place first.
- “We know we are losing hours, but not how many or where.” The intuition is there, the number is not, and without the number there is no way to decide and no way to prove anything afterwards.
2. How we work
Order matters more than technology, which is why the diagnosis comes first and not last. It takes two to four weeks:
- Map of the current process. How it actually runs today, not how the procedure says it runs. This is where the invisible steps nobody had counted show up, and they are usually half the problem.
- Measurement of the starting point. Current hours per month, number of cases, error rate and lead time. This is the phase most often skipped and the one that causes the most trouble: without this number, the later argument about whether the project worked comes down to opinions.
- Scoring of candidates. Every candidate process is scored by impact and by effort. The first one should be the best ratio of the two, not the most eye-catching or the most exciting.
- Roadmap with dependencies. What goes first, what needs something else done before it, and what stays out of the next twelve months.
- A fixed price for the first project and a written success criterion before it starts.
The diagnosis does not oblige you to hire us for the implementation. It comes out detailed enough for your own team, another supplier or us to execute. A diagnosis that only the person who wrote it can execute is not a diagnosis, it is a quote in disguise.
3. What you get in writing
This is what is in the folder when we are done:
- The map of the current process, including the steps that were not documented.
- The table of scored candidates, by impact and effort, with the reasoning behind each score.
- A sheet per candidate: what gets automated, what stays in human hands, which systems have to be touched and what risk level it carries under the European AI Regulation.
- The six to twelve month roadmap, with order and dependencies.
- The measured starting metrics, not estimated ones.
- The list of what was discarded and why. It is usually the most useful part of the report and the part nobody else hands over.
4. Typical starting points
Three situations people usually arrive from. Our clients are currently in manufacturing — discrete and process: automotive, food and drink — which is where our track record is; the other two are patterns that repeat in any sector.
Manufacturing: automotive, food and drink
Everyone knows hours are being lost in plant administration, but each manager points to a different process as the worst one. With no common number, the decision goes to whoever pushes hardest rather than to where the money is.
Candidate processes end up scored on the same four metrics, and the first one picks itself.
A company that has already tried AI without results
There are one or two experiments done, a live subscription somewhere and no process actually changed. The organisation has learned that AI does not work, when what failed was the fit with the process.
The diagnosis separates what failed because of the technology from what failed because of the approach, and recovers whatever is salvageable.
A board that has to justify the investment
The budget is approved and someone has to explain what it is being spent on and what is expected in return. With no measured starting point, any return figure presented is an opinion.
The roadmap arrives with the first project priced and the success criterion written down before it starts.
5. What you need before we start
If any of these is missing, we say so in the first conversation and not in the invoice:
- Access to the people who run the process, not only to the people who describe it. Those are the best-spent hours of the whole engagement.
- Willingness to measure before touching anything. If the starting point cannot be measured, it is not worth starting.
- Someone with the authority to decide. A roadmap that has to pass through four committees arrives late to its own execution.
- Two to four weeks, with the people involved genuinely available.
- Honesty about what has already been tried. Previous attempts are the most valuable information there is, and the most frequently hidden.
6. When this is NOT the answer
Saying no in time saves more money than most projects make:
- You already know what to automate and the case is clear. Then the diagnosis is redundant: go straight to automating that process and save yourself the weeks.
- You want a report to back a decision already taken. We are not going to write it, and if we did it would be no use to you.
- There is no budget to execute anything afterwards. A roadmap with no execution is a document that ages in a drawer.
- The company is in the middle of an ERP migration. Better to wait: whatever is decided now changes as soon as the management system does.
7. How the return is measured
Consulting is measured oddly, because its first value is negative: what it stops you spending. A six-month project discarded in week two has already paid for the diagnosis. Beyond that, it is measured by what it enables:
- Hours per month identified as recoverable, with their equivalent cost.
- Time to the first project in production. With the prioritisation done it is usually another two to four weeks; without it, months.
- Number of candidates discarded on evidence, which is money not spent.
- Starting metrics available so the effect of each project can be demonstrated afterwards.
The full method, phase by phase, is in the guide on artificial intelligence applied to business processes. And if the process handles personal data, the framework of obligations is in AI compliance.
Note: this page is for information only; it is not legal advice nor a guarantee of results. The figures quoted are usual ranges in real projects, not contractual commitments, and they depend on the scope and the starting point of each company.
8. Frequently asked questions
What does AI strategy consulting include?
A map of how your processes actually run today, measurement of the starting point in hours, cases, errors and lead time, a score for every candidate by impact and effort, a sheet per candidate with the systems involved and its risk level under the regulation, a six to twelve month roadmap with dependencies, a fixed price for the first project, and the list of what was discarded with the reason. That last part is usually the most useful and the one almost nobody hands over.
How long does a diagnosis take?
Two to four weeks, depending on how many processes are in scope and on the availability of the people who run them. What sets the pace is not our work but your team’s hours: we need to talk to the people who do the process, not only to the people who describe it.
Do we have to hire you for the implementation afterwards?
No. The diagnosis comes out detailed enough for your own team, another supplier or us to execute. A diagnosis that only the person who wrote it can execute is not a diagnosis, it is a quote in disguise.
How is this different from a report by a large consultancy?
It comes out with numbers measured in your own building rather than market benchmarks, it says explicitly what should not be done, and the person writing the roadmap is the one who would know how to implement it. We do not deliver slides: we deliver something that can be executed the following week.
And if the conclusion is that AI is not worth it?
We say so and explain why. It happens, and it is a valid outcome: there are processes where classic automation is cheaper and more predictable, others where the volume will never pay anything back, and others that need redesigning first. Discarding a six-month project in week two already pays for the diagnosis.
Is it useful if we have already tried AI and it did not work?
It is one of the best starting points, because there is real information to analyse. Most pilots that never reach production fail on the fit with the process rather than on the model: nobody measured the starting point, nobody defined what counted as correct, or the process was disorganised before anyone started. The diagnosis separates what failed because of the technology from what failed because of the approach.
Do you work with companies outside manufacturing?
Yes. Our clients are currently mostly in manufacturing, discrete and process, but administrative processes look far more alike across sectors than they do from inside any one of them: a supplier invoice reads the same in a factory as in a law firm. We are not limited to one sector.
Services that usually go with this one: process automation · AI agents.