Blog

A hundred new AI tools, and not one of them knows your company

Every week a new company appears in your LinkedIn feed that will fix your planning, your quotes or your customer service "with AI". Each one is convinced of its own genius. And most of them do solve something: one small piece, and often only for a while.

Manuel Pattyn · October 2026

Short answer. Ask the same question of every new tool: what's stopping my own IT from building this themselves? If the answer is "nothing", the value isn't in the tool but in knowing how your company works. No tool can sell you that knowledge. Start there, ask hard questions about your data, and count on a human in every step that matters.

Why ten more appear every week

Building software has never been this cheap. With AI, a small team puts a working application together in a few weeks. That's good news, and it explains the flood: the threshold to start a company has almost disappeared.

Look under the bonnet and you often see the same pattern. A screen, a few integrations with the systems you already have, and an API call to a language model from one of the big AI vendors. That model isn't theirs. The heart of the solution is bought in. What they add is the glue.

There's nothing wrong with that. Some strong companies are among them, with real expertise. But many of those tools are built around one problem, at one customer, with no view of what happens before and after. They close a gap. They don't ask why the gap is there. That's how you end up with patches: each useful in its place, together a layer nobody oversees.

The question to ask every time

What's stopping a solid company with its own IT department from making this patch itself? Or better still: from tackling the bigger picture itself?

Often the honest answer is: nothing. The same language models are for sale to everyone. The same AI tools the start-up builds with are on your developers' screens too. If technology is all a vendor brings, you're buying something you could have had yourself within a few months. Without an extra subscription, and without one more party that sees your data.

So where is the value? In knowing how companies work. How a plan actually shifts when four technicians call in sick. Why an accountant writes a customer number differently from sales. What an acquisition does to your numbers. Whoever lacks that knowledge builds a tool that solves the problem in the quote, not the problem in your company.

A vendor whose added value doesn't come from understanding your organisation, and dozens of others, has little to offer that your own team can't.

Security and data: where it often gets vague

A tool with an API to a language model sends your data outside. Customer names, addresses, contracts, time registrations, sometimes payroll data. The website says "secure" and "GDPR-proof". Ask further and the answer often gets vague. These are the questions that deserve an answer on paper, before you sign:

Where does my data go? To which model, from which vendor, and in which country does it run?

Is my data used for training? And is that written down in the vendor's contract with the AI company, not just in a brochure?

Who are the sub-processors? Hosting, logging, the language model itself. Every party is a link in your processing chain.

Who has access, and is it logged? Including the start-up's own developers.

What happens if the company stops? Many of these companies won't exist in three years. Where is your data then, and how do you get it back?

Is there a data processing agreement? And does it match what your own IT and your customers require of you?

Whoever has no clear answer to those questions hasn't thought about them yet. That's no detail: it's your data, and your responsibility.

A human will always be needed

"Fully autonomous" is a nice phrase in a pitch. In practice a language model sometimes guesses, and it doesn't say when. A customer without an email address, a technician who calls in sick at the last minute, an invoice with an unusual amount: no model decides those without someone looking.

Good automation doesn't hide that. It builds it in: a place where a human sees the exception, approves the large amount, calls the customer who wants it differently. That's not a weakness of the technology. It's a choice that decides whether you still trust it a year from now. So ask every vendor not only what their tool automates, but also where the human steps in, and who that human is.

Where the value does lie

Not in another tool, but in someone who looks at the whole first. What you may expect from a partner:

Walk the floor first, then build. A few days on the floor, from the driver to the CEO. Only then do you know which gap is really a gap, and which disappears once the basics are right.

Knowledge of hundreds of companies, not of one product. Specialists who have done the job themselves, in purchasing, logistics, accounting or service, and who keep seeing at other companies what works.

In the central system, not on top of it. What can go in your ERP goes there. A separate tool only if it does something the core can't.

Build and maintain. With AI, building has become the easy part. Maintaining through every upgrade is the real work, and it belongs in the deal.

On your network, under your rules. Named access, documented changes, your IT checking. No data quietly leaving for a model.

Honest when you can do it yourself. Sometimes the best advice is: let your own team build this. A partner who dares to say so isn't selling you patches.

Five questions for every new tool

What's stopping my own IT from building this themselves?

Which gap does this tool close, and why does my system leave that gap?

Where does my data go, and is that in a contract?

Where does a human step in, and who is that on our side?

What happens to my data and my processes if this company no longer exists in three years?

What such patches look like in euros, with three real proposals: baling wire between the applications →

The best AI tool is rarely the smartest. It's the one built by someone who understood your company first.

Let's talk →

← All posts