This list isn't made up. It comes from the Q&A round of a webinar run by a fellow provider, attended by people who were about to buy. That's more useful material than a survey, because these are the questions someone asks with their hand already on their wallet.

We've grouped them into eight clusters and written down, for each cluster, what a sound answer contains. Not what our answer is, because then it would be a brochure. Where we've written the subject out in full elsewhere, the link is included.

At the bottom is the checklist in seven questions. If you only read that, you have enough.

01 / ownershipWhat do I keep if we part ways?

Don't ask 'will I own it', because you'll get a yes to that everywhere. Ask what you're left with upon cancellation. A sound answer is: your accounts and all your data remain yours, and the built layer either stops or you take it over. Expect a mixed form, and the details determine everything.

Be wary of constructions where cancelling means you lose your customer data, email history or content. That isn't delivery, that's hostage-taking with a friendly bow. Also press on the format: 'you can export' without saying what and in which format is not an answer.

Four questions make that concrete, and all four are about the day you want to leave:

1. In whose name are the accounts things are built in? If in doubt, ask who receives the invoice for that subscription.
2. What can I export tomorrow, and in which format? An answer without a format is not an answer.
3. What happens to my data if I cancel, and is that in writing? A good answer refers to an article in the agreement, plus a retention period.
4. What keeps working without you, and what doesn't? A provider who can name their own indispensable part knows what they deliver.

Don't ask whether you're the owner. Ask what you can take with you tomorrow.

Our own answer to this question, including where the dependency does lie with us, is in who owns it when it works.

02 / the modelWhich AI model is underneath, and what if that changes?

A fair question that calls for a precise answer. You want to know which model is running, who supplies that model, what happens if it becomes more expensive, and whether you'll have to start over if something better comes along in a year's time.

What a good answer contains: the name of the supplier, whether the costs of that lie with you or with the builder, and whether the system is built such that a model switch is an adjustment or a rebuild. A provider who answers this with 'we'll sort that out' has dodged the question.

We don't yet have a publicly worked-out answer to this. It's a fair question and it deserves a better piece than a paragraph in an overview; that's still to come.

03 / the catchWhat's the downside that isn't in the brochure?

The sharpest question from the entire webinar came from the audience: it seems like a miracle machine, what's the downside? No substantive answer came to that. That question does deserve one.

Three downsides you should hear about with any honest system. The first few weeks are correction work: drafts feel off, summaries are too long, you invest attention before you get time back. AI makes mistakes, including convincingly worded ones. And a system doesn't fix a weak offering: if your proposition doesn't add up, the rejection now just arrives faster.

Anyone who responds to 'what's the downside' by only naming advantages hasn't understood the question.

A second question hangs off this one, and it's concretely testable: does the system send or publish anything without my approval? Our answer to that is in why we don't publish your posts automatically.

04 / too earlyI'm still at an early stage. Is this already something for me?

Probably not, and that's the most honest answer there is. If your offering isn't yet fixed and you don't yet know exactly who your customer is, a system locks in something that's still moving. You get acceleration of unclarity.

The order is: first know what you sell to whom, only then lock it in. A provider who tells you this while missing out on a sale by doing so is telling you something about how they do business.

How to determine whether your situation is actually ready for it is in you don't have a business that's running.

05 / expandingDo I have to pay again for every expansion?

Demand that all prices, including those for expansions, are set out in writing beforehand. What's included in the basics, what does an extra component cost, and what exactly is the monthly fee for. Prices that are decided on the spot or only apply by email belong to a sales trick, not to a supplier you want to work with for years.

Also ask what the monthly fee is for. There's a difference between rent and management: with rent, your access stops if you stop paying, with management, the maintenance stops. You only notice that difference at the moment you want to leave, and that's exactly the moment you'd have wanted to know it.

And ask what happens if the build overruns: does the provider carry that risk, or do you? That question is almost never asked. A fixed price with an open end is not a fixed price.

Our own figures are fixed in advance per step at datadriftdigital.nl/prijzen. Whatever is possible after On the Map carries no public price and is by invitation only.

06 / customer dataWhere is my customer data stored, and who can access it?

This question deserves a more precise answer than 'everything is encrypted'. Three things you want in black and white: where the data is stored and who can access it (including at the AI supplier), whether there's a data processing agreement, and what's arranged around the AI Act.

That last point is no longer a detail. Since February 2025, there's been an obligation around AI literacy for anyone using AI, and from August 2026 transparency about AI-generated content is added to that. A provider who raises these subjects themselves is further along than one who has to be asked about them.

We're still working out our own architecture on this point and won't publish about it until it's complete. What we can already say now: ask for the list of parties that process your data, and ask whether that list exists in writing. A supplier who can't provide it doesn't have one.

07 / where to startThere's so much possible, where do I start?

With one workflow, not with a plan. Choose the part you repeat most often and that requires the least judgement: your inbox, your meeting notes, your quotes. Expanding is always possible; scaling back after too big a start feels like failure, and there's no need for that.

The question behind this question is usually a different one: what's actually the bottleneck in my case? That's where it starts, not with the tool. A provider who starts with a package choice instead of with your week is selling tools.

Three examples of such a first workflow, each written out in full: your inbox as your first AI employee, customer records that keep themselves up to date and let AI write your quotes.

08 / voiceWill it still sound like me, and what if I don't understand something?

You don't need to worry about the language; that's solved. About your voice, you do. Ask where the system gets its picture of your way of writing from. If that comes from a form with four fields, it becomes the average of everyone. If it comes from your own material and from a conversation, there's something to work from.

On guidance: ask what happens if something breaks and who you'll speak to then. And pay attention to what's not promised. We, for instance, don't have a community, and we don't promise one either.

The full piece on this: language is easy, voice is the work.

09 / the checklistWhich questions do you ask every provider?

Seven questions, summarised from all eight clusters. Print them out, take them to every conversation, and pay particular attention to who answers calmly and who talks around it.

1. What exactly do I keep if we stop: accounts, data, built work? And in which format?
2. Where is my customer data stored and is there a data processing agreement?
3. What's arranged around the AI Act: literacy and transparency?
4. Does the system send or publish anything without my approval?
5. What do expansions cost, in black and white, before I sign?
6. What will disappoint in the first few weeks, and how is that supported?
7. What can the system explicitly not do?

A provider who answers all these questions calmly and concretely deserves the conversation. One who talks around them has already answered your question.

The best test isn't whether someone can do everything. It's whether they can name what they can't do.

This text was produced with AI assistance and checked and approved for publication by a human.