The market for AI providers exploded overnight — from solo consultants to large agencies, suddenly everyone offers "AI". That makes choosing the right partner the decisive skill. How do you spot one who ships working software instead of slides? Here are seven criteria.
Why the choice decides success or failure
Most failed AI projects don't fail on technology — they fail on partner choice. A provider who writes concepts for months burns budget before anything runs. One who ships a prototype early makes the value visible and keeps the risk small. So look less at glossy decks and more at the points below.
The 7 criteria
1. Do they actually build — or only advise?
The most important question first: does the provider deliver running software, or does the project end in strategy papers? Ask to see concrete, shipped solutions, not just concepts.
2. Do you talk to the developer directly?
At big agencies, sales sells and a rotating junior team builds. Founder-led means you talk to the person who actually develops your solution — less friction, more speed.
3. How fast is the first prototype?
A serious partner can show your first use-case as a working prototype in days to a few weeks — not quarters. Speed here isn't a luxury, it's risk reduction.
4. EU hosting & GDPR from day one?
Data protection is often a deciding factor. Clarify early: where is data processed, which models are used, is EU hosting possible? A good partner has clear answers.
5. Technical depth: agents, RAG, integrations?
A searchable knowledge system (RAG), autonomous AI agents and integration into your existing tools (CRM, ERP) are craft. Check for real experience here — not just a ChatGPT wrapper.
6. Transparent, project-based quotes
You should understand what you pay for. A clear, individual quote with defined scope beats any open-ended hourly model. Be wary of providers who keep the scope deliberately vague.
7. Clear use-case focus, not buzzword bingo
The right partner asks about your bottleneck first — not your budget. Anyone throwing around "revolutionary" and "disruptive" without naming a concrete use-case rarely has one.
Common pitfalls
- Pure strategy consulting that ends in slides, not software
- Month-long concept phases with no visible result
- Sales sells, a rotating junior team delivers
- Vendor lock-in: you can't move forward without the provider
- Buzzwords without measurable value
The key questions before hiring
- Can you show me a solution you've shipped?
- Who actually builds — and do I talk to that person directly?
- When will a first working prototype be ready?
- How is data protection handled (EU hosting, models)?
- Who owns the code after the project ends?
Conclusion
You won't spot the right AI agency by the biggest pitch, but by shipped software, a direct line to the maker and a clear use-case. If you're unsure where your strongest lever is, our free AI use-case check helps in 60 seconds. For useful application areas, read AI in the Mittelstand: 7 use-cases with real ROI.