What an AI scan does and does not deliver

An AI scan delivers a well-founded decision, not a working system. After an AI Readiness Scan you know where your organisation stands, which five to ten opportunities exist, what the best three return on your own figures and in which order to tackle them. At that point you have built nothing, your data is as messy as it was before, your people still work the same way and nobody has guaranteed your return. Anyone who tells you otherwise is selling you an expectation the scan cannot meet.

That second half of the answer rarely appears on a sales page. Yet it is the half that is most useful to you as an owner or director. You are not deciding about a tool, you are deciding about a sum of money and a share of your organisation's attention. Then you want to know in advance where the boundary lies of what you are buying.

This piece goes through both sides, with the boundary drawn as precisely as possible.

The short answer in a table

The left column comes straight from our own page about the AI Readiness Scan. The right column is not on that page, and that is why it is here.

Four people in a meeting room. The scan ends with a decision that this table still has to make.

Why we start with what it is not

There is a simple reason to start with the limits. A provider who says honestly what something does not do can be believed about what it does do. A provider who promises everything cannot be held to anything.

For a company of fifty to three hundred and fifty people, that weighs more heavily than for a large corporation. A large company can write off a failed exploration and start again. In your company there is a name attached to it, yours, and the next time you mention the word AI in a management meeting, everyone looks at what the last attempt produced.

A scan that promises more than it delivers therefore costs you twice. Once the amount, and once the organisation's confidence that the next step is worth taking. The second is harder to earn back.

Setting expectations correctly up front is therefore risk management, and it is exactly what you may demand from a small, reversible first step.

What an AI Readiness Scan does deliver

The scan page lists four deliverables. We go through them one by one, with what you concretely do with each.

One: your position. An honest picture of where your organisation stands today, relative to your peers and your ambition. The value is in the word honest. Most management teams misjudge their own position, in both directions. Some think they are far behind while they have had usable data lying around for years. Others think they are doing well because a few employees use a chatbot, while nothing has changed in the operation. A position assessment takes that estimate out of gut feeling.

Two: a priority matrix. Five to ten concrete opportunities, ranked by return against feasibility, so you know where to start. For most management teams this is the part that saves the most time, because it reduces an endless list of ideas to a short list with an order. How such a matrix works and why feasibility weighs as heavily as return is covered in the priority matrix for ten AI opportunities.

Three: return on your own figures. A calculation of the return on your top three opportunities, based on your real figures and not on benchmarks. That difference is bigger than it sounds. A benchmark tells you what an average company could save. You are not an average company. A calculation based on your own hours, your own rates and your own volumes is a figure you can work with yourself, and one you can defend to your accountant or your shareholders. How to set up that sum yourself is in calculating AI return on your own figures.

Four: a roadmap for the boardroom. A well-founded plan that you present to your board and can start executing next quarter. Not a thick report, but an order of steps with, for each step, what it requires and what it should deliver.

Around that, the page lists a few practical facts worth knowing before you start. The scan takes two weeks from start to delivery. It consists of a digital intake, conversations with a number of key people, an analysis and a delivery that we present live at your office. Your IT does not need to be changed for it: nothing is migrated and your existing systems stay intact. And the roadmap remains yours, even if you do nothing further with us. What those ten days look like day by day is in from AI-Foto to roadmap in ten days.

What a scan does not deliver

Now the right column, written out. These are the limits we set on the scan ourselves. They apply to nearly every scan in this category, not only ours.

It builds nothing. After two weeks no new application is running. There is no pilot, no prototype, no trial in production. That is a deliberate choice. A diagnosis and a build are two different activities with a different risk profile, and anyone who mixes them up starts building before it is clear what is worth building. The building happens afterwards, if at all, in a Capability Sprint on a project basis, with your own people.

It does not make the decision for you. The matrix shows which opportunity is at the top and why. Whether you start there is a choice only you can make, because only you know what else is on your plate. Perhaps an ERP replacement is underway, perhaps an acquisition is coming, perhaps the department at the top has just lost two people. A scan takes those things into account if you tell us about them. It cannot weigh them for you.

It does not fix your data. The scan shows where your data is a bottleneck for a given opportunity. That is useful information, because it often explains why a high-return opportunity still sits lower in the matrix. But after the scan the word customer in your ERP still means something different from what it means in your CRM. That work still has to be done, and it belongs to the build. What that involves is in getting your data in order for AI agents.

It does not guarantee a return. The calculation for your top three is a well-founded expectation, based on the figures you supply and the assumptions we make together. It is as good as those figures and those assumptions. Anyone who promises you a guaranteed return before anything has been built knows something nobody can know, or is telling you something that is not true.

It does not teach your people anything. In the conversations people are asked about their work, and that sometimes sets something in motion. But a one-hour conversation is not training. After the scan nobody in your organisation can do anything they could not do before. The ability to work with AI yourself is built during the build, not during the diagnosis. That is why an advisory report is not a capability, and a roadmap is not one either.

It does not look at everything. Five to ten opportunities is a selection. There are processes in your company we do not look at, because the conversations and the intake do not bring them up or because they have too little volume to be worth it. That is the price of a two-week scan. A full review of every process takes months, and usually produces a list too long to act on.

It is not a legal assessment. The scan does not assess whether an application complies with the AI Act or the GDPR. If that matters for an opportunity, you need a separate assessment by someone who carries legal responsibility for it.

A hand writing in a notebook. Write down what a scan does not do before you sign, not afterwards.

Where exactly the boundary lies: an example

An example makes the boundary more concrete. It is made up and serves only to show the reasoning.

Suppose: a wholesaler with 120 employees. In the intake and the conversations it emerges that the purchasing department spends a lot of time retyping order confirmations from suppliers into the ERP, and chasing deviations in price and delivery time. Customer service mentions something similar for incoming customer orders by email. The planner mentions the weekly stock forecast, which he adjusts from memory.

In this example the scan delivers three things about these three opportunities. A ranking: the order confirmations come first, because the volume is high, the outcome is easy to check and the data is already there. The stock forecast ranks lower, because the return may be larger but the historical data is incomplete. A calculation: hours per week times the full hourly rate times the share that can reasonably be automated, worked out on the figures the wholesaler supplies itself. And an order: first the order confirmations, then the customer orders, and the forecast only once the data is in order.

What the scan does not deliver in this example: a system that reads the order confirmations. A decision about who in purchasing becomes responsible. A cleaned-up product master file. A guarantee that the saving will be achieved. And a buyer who knows how to check a model.

That last list is not a shortcoming of the scan. It is the list of things the wholesaler has to arrange afterwards, and it is better to know that list in advance than to discover it halfway through the build.

What is still asked of you after the scan

Because the scan delivers a decision and not a system, there is work for you after the delivery. That work is manageable, but it is there.

You have to choose. The roadmap gives an order, but you decide whether to follow it, wait, or take the first step yourself. There is no wrong answer, as long as it is a decision and not a postponement without a date.

You have to appoint an owner. Someone in your organisation who is responsible for the first opportunity, even if an external party gets involved. If nobody can take on that role, that is useful information: then the first step is freeing up time, and the build comes later.

You have to free up time. Every opportunity you take on needs hours from the people who know the work best, and those are almost always your busiest people. The scan shows who those people are. It does not clear their diaries.

And you have to decide how you will measure in six months whether it worked. The calculation from the scan is a good start for that, because it is based on your own figures. Agree in advance which number you want to see back in six months.

How to assess a scan, ours or anyone else's

More parties offer an AI scan, and the content varies widely. A few questions help to put them side by side. Ask them of us too.

1. What do I concretely get on paper, and can I use it without you?
2. Do you calculate on my own figures or on market benchmarks?
3. What do you build during the scan, and what not?
4. Who in my company do you talk to, and how much time does that cost them?
5. What do you say if the outcome is that I should not start yet?

The last question is the sharpest. A scan that can never conclude that you would be better off waiting is a sales conversation with a report attached. An honest outcome can also be that the most important step is something else first: finishing an ERP replacement, freeing someone up, aligning the definitions in your data. A party willing to name that as a possible outcome can be believed when the outcome is that you should start.

A second test: ask who the scan is really for. A scan that starts with IT produces an IT plan. Why it should start with the owner is in why an AI scan is not an IT project. This piece is about what comes out. That piece is about who sits at the table.

A long warehouse aisle with shelving. The opportunities are in the daily work, the scan only ranks them.

Why a diagnosis is still worth it

After all those limits the question is fair: why start at all? The answer lies in what happens when you skip the diagnosis.

Statistics Netherlands (CBS) measured that in 2025, 45 percent of companies with 50 to 250 employees used AI, against 66 percent of companies with 250 or more. Among mid-sized companies that figure was still 20 percent in 2023. So plenty of companies are starting. The only question is with what.

A widely discussed 2025 report from MIT's Project NANDA concluded that around 95 percent of the generative AI pilots studied had no measurable effect on profit and loss. That report is not peer-reviewed and is partly based on interviews and surveys of executives, so partly on self-reporting. Do not base policy on it. The direction is recognisable though: a lot is tried and little is calculated in advance.

That is where the value of a diagnosis lies. It does not prevent you from building something that disappoints. It reduces the chance that you build something in the wrong place, because you know in advance which opportunity is at the top and why. For a company that is careful with money and risk, that is the cheapest insurance available: a few thousand euros to prevent a much larger investment from resting on gut feeling.

Where to start

If you first want to see where you stand without any obligation, start with the AI-Foto. It is a free self-scan of twelve questions that takes ten minutes and gives you a first picture of your starting position. What the AI-Foto measures exactly is described on what is an AI-Foto.

If you then want a well-founded plan, the AI Readiness Scan is the next step. It starts from 3,900 euros, depending on size and modules, and in two weeks delivers the four outcomes described above. No more, and no less. A short definition is on what is an AI Readiness Scan, and if you first want to check whether the organisation is ready for it, you can start with how do you know whether your organisation is ready for AI.

About this page

The four deliverables, the two-week lead time, the structure of intake, conversations, analysis and delivery, and the price from 3,900 euros come from our own AI Readiness Scan page. The list of what a scan does not deliver is Radical's own position. The figures on AI use by company size come from CBS (provisional 2025 figures). The 95 percent figure comes from the MIT Project NANDA report, which is not peer-reviewed and is partly based on self-reporting. The wholesaler example is made up for illustration; no client data was used. This is the state of play on 5 October 2026.