Choosing an AI company is not a tooling choice
Anyone looking for an AI company right now is staring at a market that has grown fivefold in two years. Hundreds of Dutch agencies, from solo operators to Big Four arms, claim to do "AI implementation for SMEs". The gap between a party that delivers a PowerPoint and one that puts something into production is worth a factor of ten. This piece is my honest checklist, written from my experience as a builder and tested against the peer agencies I run into during pitches. One thing up front: the agencies that pitch hardest are usually the least suitable partner. My site has no "book a call" button high on the page. That is not an oversight, that is a deliberate choice.
The gap between a party that delivers a PowerPoint and one that puts something into production is worth a factor of ten. And the agencies that pitch hardest are usually the least suitable partner.
The seven criteria in short
| # | Criterion | The core question |
|---|---|---|
| 1 | Production or pilots | Is something running live at a client right now, and can I call them? |
| 2 | Builder or advisor | Do they write the code themselves, or do they deliver a report? |
| 3 | Who sits across from you | Are you talking to the builder, or to an account manager? |
| 4 | Tool-independent | Do they pick per project, or defend a partner bonus? |
| 5 | What happens to your data | EU-only, no model training, a clear data processing agreement? |
| 6 | AI literacy (article 4) | Do they include training and a compliance file per use case? |
| 7 | Price and settlement | How are scope changes, exit and code ownership handled? |
The seven selection criteria
1. Can they show you what is in production?
The question. An AI company that only shows pilots, demos or POCs typically delivers nothing that is still running a year later. Ask concretely: which client has an AI system live right now, how many hours a week does it save, and can I call that client for ten minutes? An honest agency has three to five examples ready. A marketing agency that does AI as a side product does not. Anyone who hesitates or falls back on "we can demo it" has nothing in production.
2. Do they build or only advise?
Two camps: builders and strategists. Strategists deliver reports, builders deliver working systems. Both have their place, but SMEs usually get more out of a builder who also thinks strategically than out of a strategist who then hands you off to a vendor. Ask who actually writes the code. If the answer is "a subcontractor in Poland", you know where you stand. AI strategy without pilots is not a strategy, it is a report.
3. Who sits on the other side of the table?
At large agencies you get account managers and project managers wedged between you and the tech. At small agencies you talk directly to the builder or the founder. For an SME, direct contact with the founder is faster and cheaper, because middle layers cost money and let information leak. Ask: who is my fixed point of contact and how often do I speak with them? If the answer is "a delivery lead" and the tech sits two layers deeper, you know you are sticking your own letter onto a mail sorter.
4. Tool-independent or tied to a partner?
A certified UiPath partner recommends UiPath. A Microsoft Gold Partner recommends Copilot Studio. That is not necessarily wrong, but it is biased. Ask how they arrive at a tool choice and what their advice is if you do not want Microsoft or UiPath. An honest agency evaluates per project and has no partner bonus to defend. For the RPA trade-off see /rpa, for agentic AI see /ai-agents.
5. What do they do with your data?
Three things to check. Where is your data processed, EU-only or US cloud? Is it used to train models, which is the default in consumer subscriptions? And what does the data processing agreement say about retention and sub-processors? For healthcare, legal and financial this is not a detail, this is a dealbreaker. A serious agency shows you how they can build on-premise or EU-only when needed. For the broader AI Act context see /ai-act.
6. AI literacy (article 4) in order?
As of 2 February 2025 the EU AI Act requires that employees deploying AI have sufficient knowledge. An agency that does not explain this, does not include training and does not deliver a compliance file per use case exposes you to fines and reputational damage during an audit. And the signal runs deeper: does the agency itself use AI well? If their own people only open ChatGPT for a birthday email, you are not going to have an AI system built by people who barely hold the instrument themselves. AI-first at your supplier is a hard requirement. See /ai-training for how I fill this in.
7. What does it cost and how do they settle up?
Three models you will encounter: fixed price per project, time and materials, and retainer. For scoped pilots, fixed price works best. For open-ended work, hourly is a good option, provided there is a ceiling budget. For ongoing monitoring plus small changes, a retainer makes sense. The real conversation is not about the hourly rate, but about how scope changes are settled, what the exit clause looks like and who owns the code, prompts and documentation. Anyone who answers vaguely is parking the problem with you.
Red flags
Three patterns in the market I read as warnings.
"Six months of AI strategy, then we start building." For 90 percent of SME cases that is overkill. A solid AI roadmap takes two to four weeks, then you build. Agencies selling six months of strategy are often selling strategy to you and not building themselves. In that case you are their research budget.
No fixed prices, no scope document. If an agency starts with "we can only quote a price after an extensive intake process" and that intake costs 5,000 to 15,000 euro, the scope is not clear and you are paying to be a prospect. For scoped use cases, a serious builder gives you a ballpark after a 30 minute conversation. Not exact, but honest.
A tool pitch instead of problem analysis. "We use UiPath and GPT-4 and n8n and LangChain" is a tool stack, not an approach. A good conversation starts with "what is this costing you now, in hours or euros, and what would halving it be worth?". Tools come after. Anyone who leads with their stack is telling you what they want to sell, not what you need.
Realistic rates
DataDream's pricing: no fixed rates on the site. A fixed price per project after a 30 minute conversation. Scope varies too much to slap a number on it up front, and a price without context poisons the conversation. What you can expect: scope first, price after. At every agency you compare, ask for a fixed price for scoped work, a scope change clause for open-ended work, and who owns the code, prompts and documentation after delivery. Anyone who cannot give a clear answer in a first conversation will not do so in a quote either. Book a call via /contact.
How DataDream scores on this checklist
This checklist is not something I invented, clients asked exactly these questions in selection conversations. My answers in the same order.
Production: five plus clients with live AI systems, voice agents, document extraction, content pipelines. References on request. Builder or strategist: primarily a builder, strategy only where it sharpens the build. Direct contact: you work directly with me (founder and lead engineer), no account manager layer in between. Tool-independent: no partner lock-ins, per project I evaluate UiPath, Power Automate, n8n, Workato, custom Python and the right LLM provider. Data: EU-only hosting is available, on-premise possible for sensitive sectors, GDPR compliant by default. AI literacy: article 4 compliant training via /ai-training, compliance file per use case. Rates: fixed prices for scoped projects and an hourly rate for open-ended work. Ownership stays with the client, always.
For the full services overview see /ai-oplossingen. For agentic AI and voice AI see /ai-agents. For RPA and workflow automation see /rpa.
Conclusion: three questions to ask up front
In every conversation with a potential AI company, these three questions filter out the time wasters.
One: which client currently has an AI system in production that you built, and can I call them for ten minutes? Hesitation or a fallback to "we can demo it" is a red flag. Two: say I want a working first version in six weeks, which use case would you recommend, and with which tools? A good agency gives a concrete answer within five minutes. A bad agency says "that depends on the intake". Three: what does your standard data processing agreement say about AI data? If they have one and can walk you through it, fine. If not, keep looking.
Want to run this checklist against DataDream? Book a free 30 minute call. You get an honest read from me on whether I am the right choice for your situation. Or start with the free AI Readiness Scan for a first direction.
Curious what AI can do for your business?
Take the free AI Scan and find out in 1 minute.
Frequently asked questions
- How do you choose an AI company in the Netherlands?
- First question to ask: which client currently has an AI system from you running in production, and may I call them for ten minutes? An honest agency has three to five production examples ready. Also check whether they build or only advise, whether you speak directly with the builder, whether they are tool-independent, and what they do with your data.
- What are red flags with an AI agency?
- Three. 'A six-month AI strategy, then we start building' (unnecessary for 90 percent of SMBs; you become their research budget). No fixed price or scope document, while the intake process itself costs thousands of euros. And a tools pitch ('we use UiPath, GPT-4, n8n') instead of a problem analysis.
- What does an AI agency cost?
- Three models: fixed price per project (best for scoped pilots), hourly rate with a max budget (for open-ended work), and a retainer (for ongoing monitoring plus small changes). The real conversation is not about the hourly rate, but about how extra work is billed, the exit clause, and who remains the owner of code, prompts and documentation.
- Do I need a builder or a strategist?
- SMBs usually benefit more from a builder who also thinks strategically than from a strategist who then hands you off to a supplier. Strategists deliver reports, builders deliver working systems. Ask who actually writes the code. An AI strategy without pilots is not a strategy, it is a report.