You have an AI product idea. You've validated it with potential customers. Now you need someone to build it. The problem: everyone claims to "do AI" now — from freelancers who completed a prompt engineering course to enterprise consultancies charging €2,000/day for slide decks.

I run diepen.io, a product studio that builds AI products. Here's how to evaluate partners like us — honestly, including when you should NOT hire an external team.

When you don't need an external partner

Let's start with the uncomfortable truth. You might not need us (or anyone like us).

Handle it yourself if:

Your product just needs a basic ChatGPT/Claude integration (API wrapper + UI) — a decent developer can do this in a week
You have a technical co-founder with AI/ML experience
You're pre-revenue and haven't validated the idea — spend that money on user research first
Your "AI feature" is really just a rules engine wearing an AI hat

Hire a freelancer if:

You need a single, well-defined integration (chatbot, document classifier, email automation)
Budget is under €15K
Timeline is flexible (2–4 months)

Hire a studio/agency if:

AI is the core of your product, not a feature
You need design + engineering + AI working together
You want a working product in 4–8 weeks
Budget is €25K–€100K+

The 7-point evaluation framework

1. Have they shipped AI products to production?

Not proofs of concept. Not demos. Products that real people use and pay for.

Ask for:

Live product URLs you can try
Case studies with specific metrics (latency, accuracy, user adoption)
The failures — any honest team has stories about AI approaches that didn't work

Red flag: "We've done extensive AI research and consulting." Translation: they've made presentations about AI, not products.

At diepen.io, our own product Rezeptiona is an AI phone assistant handling real calls for real businesses. We didn't just build it for a client — we built it for ourselves, which means we've dealt with every edge case, scaling problem, and 3am bug that comes with production AI.

2. Can they build the full product, not just the AI part?

An AI model without a product around it is a Jupyter notebook, not a business.

Your partner needs to deliver:

Frontend/UX — How users interact with the AI (this is where most AI products fail)
Backend infrastructure — APIs, databases, authentication, deployment
AI/ML layer — Model selection, prompt engineering, fine-tuning, RAG
Design — Making complex AI outputs understandable to normal humans
DevOps — CI/CD, monitoring, observability

A pure "AI consultancy" will hand you a model and wish you luck. A product studio delivers the whole thing.

3. How do they handle AI uncertainty?

Here's what makes AI projects different from regular software: you don't know if it will work until you try it.

A good partner:

Builds validation checkpoints into the project plan
Starts with a 2–3 week proof of concept before committing to full build
Has honest conversations about accuracy thresholds ("this will work 85% of the time — is that good enough?")
Has backup plans when the primary approach fails

A bad partner:

Gives you a fixed price and timeline for an exploratory AI project
Promises 99% accuracy before seeing your data
Has never pivoted mid-project because an approach didn't work

4. What's their stance on build vs. buy?

The best AI partner is the one who tells you NOT to build custom AI when an off-the-shelf solution works.

Good answers:

"For this use case, Intercom's AI or Zendesk's AI agent might be enough"
"We'd use GPT-4's API with good prompt engineering rather than training a custom model"
"This specific part needs custom work, but these three parts we can solve with existing tools"

Bad answers:

"We need to train a custom model for everything" (expensive, slow, usually unnecessary)
"We only work with [specific framework/model]" (vendor lock-in disguised as expertise)

5. How do they handle data and privacy?

If your product touches user data (it almost certainly does), your AI partner needs to be GDPR-literate — especially if you operate in the EU.

Non-negotiable questions:

Where is data processed? (EU servers vs. US)
Does the AI provider train on your data? (Most business APIs don't, but check)
Is there a Data Processing Agreement (DPA)?
What's the data retention and deletion policy?
How do you handle personally identifiable information (PII) in prompts?

If your partner can't answer these confidently, they'll create compliance liability for your business.

6. What does their team actually look like?

The bait-and-switch is real: a senior team pitches you, then juniors build the product.

Ask:

Who specifically will work on my project? Can I meet them?
What's the team's actual AI experience? (Years, projects, specific models/frameworks)
Will the people I meet in the sales process be the ones doing the work?

Ideal team for an AI product build:

1 senior full-stack developer with AI experience
1 product designer who understands AI UX patterns
1 technical lead / architect (can be the same as the developer in a small team)

You don't need 10 people. You need 2–3 who are excellent.

7. How do they price the work?

| Model | Good for | Watch out for | |-------|----------|---------------| | Fixed project price | Well-defined scope, clear deliverables | Scope changes get expensive, partner cuts corners | | Weekly/monthly retainer | Ongoing work, evolving scope | Can drag on without clear milestones | | Time & materials | Exploratory work, uncertain scope | No cost ceiling, requires trust | | Phased (PoC → Build → Launch) | Best for AI projects — validates before committing | Slightly higher total cost, but much lower risk |

Our recommendation: Phased pricing. Pay for a 2–3 week proof of concept first. If it works, commit to the full build. If it doesn't, you've spent €8K–€15K to learn that, not €80K.

The evaluation process: step by step

Step 1: Prepare your brief (before reaching out)

Write down:

The problem you're solving (not the solution — the problem)
Who has this problem and how they solve it today
What "success" looks like in 3 months
Your budget range (being honest saves everyone time)
Your timeline constraints

Step 2: Shortlist 3–5 partners

Look for:

Portfolio with shipped AI products
Team size appropriate for your project (2–5 people for most products)
Clear pricing model
Location/timezone compatible with your needs
Reviews or references from real clients

Step 3: First calls (30 min each)

A good partner will:

Ask more questions than they answer
Push back on assumptions ("Are you sure you need custom AI for this?")
Give you a rough estimate range, not a hard quote
Be transparent about what they don't know

Step 4: Technical deep-dive (1 hour)

With your top 1–2 candidates:

Walk through the technical approach
Discuss architecture decisions and trade-offs
Ask about their experience with your specific AI use case
Discuss data requirements and privacy approach

Step 5: Proposal review

Compare proposals on:

Clarity of deliverables (what exactly will you get?)
Risk management (what happens if approach X doesn't work?)
Team composition (who works on your project?)
Milestones and checkpoints (when can you evaluate progress?)
Total cost of ownership (not just build cost — maintenance, API costs, hosting)

What it costs: real ranges

Based on projects we've built and what we see in the market:

| Project type | Timeline | Cost range | What you get | |-------------|----------|------------|-------------| | AI feature in existing product | 3–6 weeks | €15K–€40K | Integrated AI capability, tested with real users | | AI-native MVP | 6–10 weeks | €30K–€80K | Working product, first users, validated core loop | | Complex AI platform | 10–20 weeks | €60K–€150K+ | Multi-feature product, integrations, scale-ready | | Proof of concept only | 2–3 weeks | €8K–€15K | Validated (or invalidated) approach, clear next steps |

These are European market rates for senior teams. You'll find cheaper options (offshore teams, junior developers) and more expensive ones (enterprise consultancies, US agencies). The sweet spot for most startups and SMBs is a senior 2–3 person team.

Your next step

If you're evaluating AI development partners, here's a simple way to start: book a 30-minute call with us. No pitch, no proposal, no commitment.

We'll tell you honestly whether your project needs a team like ours — or if there's a simpler path.

Book an intro call →

We build AI products from our studio in Düsseldorf, Germany. Clients across DACH and internationally.