01
Most AI projects don't fail on the technology
They fail because nobody worked out in advance what it will cost to run each month, how it connects to the systems already in use, or what happens to company data. An enthusiastic experiment ends up as something nobody uses six months later — or something that gets switched off because its running cost turned out unpredictable.
My job is to make sure that doesn't happen: to find out, before development even starts, what's genuinely worth building and what isn't.
Predictable cost
Before development begins, we calculate the monthly running cost and look at where it can be brought down.
Fits your existing systems
ERP, webshop, customer records, legacy in-house tools — you don't need to replace everything.
Data security
Customer and contract data handled in line with GDPR and the EU AI Act.
Built for production
Not a demo built to impress in a meeting, but a system that holds up under daily load.
The plan is not a sales tool
In the market, planning is usually done by whoever also wants to land the development work afterwards — often for free, because it pays off for them: the plan ends up shaped around whatever solution they're able to deliver.
With me, the plan is independent, paid work, and its fee is never credited against a later development contract. It contains at least two implementation paths, and it stays yours even if another team ends up building it. That independence is also why I can tell you if an off-the-shelf product solves the problem more cheaply — or if it's simply not worth doing right now.
I'm happy to take on the build if asked — that's agreed separately, under its own contract, after the plan is delivered. But my advice never depends on whether I get the development work.
04
Five ways to work together
00
Introductory call
45 min · free
No obligation on either side. This is a getting-to-know-you conversation, not consulting: in 45 minutes I can point you in a direction, not hand you a finished solution. Beforehand, I ask for answers to three questions by email: what eats up the most time at your company today, who makes the decisions, and by when you'd want to see improvement. If you don't have a fully formed idea yet, that's fine — a sentence or two is enough for the first question.
- We talk through how things work today and the biggest risks
- If you don't have an idea yet, I'll sketch out where AI would pay off most at your company
- I'll suggest a next step — including, if that's the honest answer, that you don't need a consultant right now
- Afterwards I send a short written summary of the direction
Does not include: a detailed assessment, a plan document, or a technology recommendation. After the call I decide, case by case, whether I can genuinely help — if not, I'll say so and point you elsewhere.
A
Situation assessment
Recommended start
The right starting point for most companies. Over two sessions, it becomes clear where AI pays off at your company, how big a task it is, and whether it's worth pursuing at all — before you spend any serious money on it.
- how it runs
- Two half-day sessions, on two separate days, at your office or online, with the relevant colleagues: we walk through today's workflows. That way you don't need to pull the team out for a full day.
- what you get
- A written summary, typically 3–5 pages. First, the most promising use cases — where the most work-hours could be freed up, with a rough estimate. Second, the three biggest pitfalls, in order of importance. Finally, the concrete next steps.
- when
- Typically within 5 business days of the second session
B
Implementation plan
2–4 weeks
A complete, build-ready plan: what to build, with what technology, and on what schedule. Any development team can implement the plan — your own in-house team or an external partner. Available in two sizes, depending on how many processes it covers.
- single process
- 2-week turnaround, 5–6 hours of your team's time
- multiple processes
- 3–4 week turnaround, 8–10 hours of your team's time
- which one you need
- This becomes clear during the introductory call or after the situation assessment. The turnaround time doesn't mean continuous presence — it's scheduled work.
What it includes
- A review of your existing systems, data, and security settings
- Data-quality assessment — whether your existing documents and records are actually fit for the task. Most projects live or die here, so if there's bad news, this is where I deliver it.
- Whether it's even worth building — whether an existing, off-the-shelf product could solve it more cheaply, and what the real difference is between the two
- Technology selection with reasoning — what's worth running on your own servers versus with an external provider
- Success criteria — what accuracy, response time, and error rate we call "working." Without this there's nothing to measure the finished build against when it's handed over.
- Human oversight built into the process — who reviews the AI's output, what happens on error, where approval is required
- A review of data-protection and regulatory risk (GDPR, the EU AI Act)
- A scheduled implementation plan for the developers, with checkpoints
What you get at the end
- A written system design with diagrams that developers can work from directly
- A cost model: what the monthly running cost will be at low, medium, and high load — calculated against your own expected user numbers and document volume
- At least two implementation paths, with different cost and risk profiles
- A closing presentation for leadership and the developers
Reviewing development quotes — included in the plan's fee. If you put the build out to tender, I'll review up to three of the quotes you receive with a professional eye and write up, in plain terms, what each one actually promises, where the gaps are, and where the hidden costs are. I won't recommend a specific vendor. If I'm bidding on the build myself, I obviously won't be the one evaluating my competitors' quotes.
Add-on — a working prototype on your own data. The planned solution, built small-scale, on your real documents and data — so instead of assumptions on paper, you can see what it actually does. Not a finished product, but a basis for a decision: it shows whether your data is even fit for the task, how usable the output really is, and roughly what scale of running cost to expect. The prototype's results are indicative — it runs on a smaller dataset than the final system. +2 weeks.
C
Monthly expert oversight
from 2 months
For when the system is already being built and you need someone tracking it with a professional eye — whether it's your own developers or an outside firm doing the work.
- If an external firm is building it: I check whether you're getting what you're paying for. Whether the delivered work matches the plan, whether the deadlines and extra costs are realistic, and whether a dependency is being built in that would be expensive to replace later.
- If you have your own developers: regular review of the plans and the code, and professional support for the team.
- Decision support on technology questions as they come up, explained in plain terms for leadership
- Keeping an eye on running costs before they get out of hand
What happens each month
- One longer sync (about 2 hours) — with the developers, leadership, or both, scheduled in advance
- One professional review of the delivered work, with a written summary — done asynchronously, no shared time slot needed
- Written questions answered within 2 business days — questions get resolved as they come up, not held over to the next sync
- One ad-hoc check-in, if something urgent comes up
This works out to roughly 20 hours of expert time per month. If a given month needs significantly more, I'll flag it in advance and we'll agree on it together.
- duration
- Starts with a 2-month base period — the first month is spent getting to know the system, real value kicks in from the second. After that, it continues month to month.
- ending it
- Either side gives notice by the 15th of the month to stop, and the engagement closes at the end of that month. So the earliest possible end is the end of month 2.
- pausing
- One month can be paused — for a summer shutdown or a project delay, for example. The paused month doesn't count towards the base period.
- scheduling
- Shared syncs are booked in advance, typically a week ahead.
- not included
- Standing availability, on-call duty, delivery against fixed deadlines, or development capacity.
- conflict of interest
- If I'm doing the development myself, this package can't run alongside it: I can't independently review my own work.
D
One-off troubleshooting
fixed-fee intervention
A specific blocker: something slow or broken, a cost that's suddenly spiked, a security question before launch. This also applies if you already have a half-finished or shaky AI solution someone else built, and you don't know what's wrong with it. Not billed hourly — you know the cost upfront.
- half-day
- Focused root-cause investigation of a specific problem, with a written summary of the proposed fix
- full-day
- A more complex case: reviewing multiple systems or processes, with recommendations and an implementation order
- beyond that
- A custom fixed-fee quote after assessment
- when it starts
- Scheduled after you get in touch, typically within 5–10 business days
An unscheduled, no-commitment intervention — which is why its fee runs higher than an ongoing engagement.
The next step is a 45-minute conversation.
If it turns out there's no real task here, or that it can be solved more cheaply, I'll tell you that too.
Pricing is sent as a custom quote after the assessment or the introductory call.
