Choosing an AI agency: twelve questions that show the difference

Take these questions with you. Pay less attention to the answer itself than to how quickly and concretely it comes.

1. 01Where do you start when you come in to work with us? A good answer is about your work and processes, not about a tool or a model.
2. 02How do you calculate what it delivers? You want to hear a method: time per step, number of people, frequency, and a scenario that's cautious enough. Not a percentage from a report.
3. 03What do you do if it doesn't pay off? An honest agency says it won't start then, or recommends something smaller.
4. 04Do you build yourselves, or do you outsource that? Advice without builders often leads to a fine plan that nobody carries out. Builders without advice sometimes build the wrong thing.
5. 05Can I call a client who has been working with you for a year? A reference from after go-live says more than a case on the website.
6. 06Who approves the AI's work? The answer should be about people: where is the check, and who decides on exceptions?
7. 07Where is our data stored, and which model do you use? You want to hear: in the EU, with a business data processing agreement, and a model that doesn't train on your data.
8. 08Who owns what you build? Code, data and documentation should be yours. Also ask how long it takes to leave.
9. 09Who keeps it running after delivery? Models change, and so do systems. Without management it won't work anymore in six months.
10. 10How do you bring our team along? Most resistance is in people, not in software. Training and guidance belong in the plan.
11. 11What does it cost, and where does the usage come in? You want a fixed price or a clear hourly rate, plus an honest picture of licenses, AI usage and hosting.
12. 12What's a small first step? A good provider can start without you signing a big project straight away.

• The solution is fixed before the process has been discussed. "You need a chatbot", in the first ten minutes.
• Savings without a calculation. "This saves 70 percent" without knowing how long the work takes now.
• Fully autonomous as the default. Most tasks need a human who checks. Anyone who waves that away hasn't put much live yet.
• No answer to the data question. Or: "the tool takes care of that".
• Only pilots in the portfolio. A pilot is easy. Something that's still running a year later is the real work.
• Lock-in. Code that only runs on their platform, or no documentation when you leave.
• Everyone in the meeting is from sales. Ask whether a developer or consultant can join the next meeting.

That depends on what you buy: knowledge, advice or building. So don't just compare the hourly rate, but also what happens after delivery. A cheap build without management often turns out expensive.

For comparison, our own prices, excluding VAT: a talk costs € 2,450, a workshop € 4,500 per half day, the AI Quick Scan € 6,000 one-off and the AI Partnership € 4,500, € 7,500 or € 12,000 a month, for a minimum of twelve months. A single rate of € 150 an hour for consulting and development. Everything at a glance, including the costs of licenses and usage, is on what does AI cost.

Fair is fair. We're not a fit for everyone.

• You want one app built and then carry on yourself. We only build within a partnership, because we want it to keep working. For a one-off assignment, a software agency is a better fit.
• You're only looking for a tool to buy. Then you'll get more out of a good vendor and AI training for your team.
• The business case doesn't clear the bar. Our guarantee is at least 2x back in the first year. If the expected scenario doesn't reach that, we don't start a partnership. A workshop is then often the better first step.

We are a fit if you want to go beyond experimenting, want to bring your team along and are looking for a permanent partner who advises, builds and then keeps it running. How we work is explained in the APO method.