HowtopickanAIconsultantinKurdistan—abuyer'schecklist

TL;DR: Most AI consultants right now are repackaging the same APIs at a markup. Here are eleven questions to separate the builders from the resellers — plus the red flags I see most often when business owners in Kurdistan get burned.

The eleven questions

1. Show me a system you've shipped that's still in production. Not a demo, not a pilot — something used by paying customers today.
2. Walk me through one bug you fixed last week. If they can't, they don't operate the systems they sell.
3. Who owns the code at project end? If they own it, you're renting.
4. Where does my data go? Which provider, which region? Vague answers = they don't know.
5. What happens when the LLM provider changes pricing? A real consultant has a contingency. A reseller doesn't.
6. Show me your evaluation harness. Anyone serious has a way to test their AI's quality across cases. If they don't, the AI's quality will drift silently.
7. How many other clients are using the exact same prompt? Some "AI consultants" sell one prompt to a hundred clients.
8. What's the failure mode? When the AI is wrong, what happens? Escalation to humans, fallbacks, audit logs — these are non-negotiable in production.
9. Will you teach my team to maintain it? If the answer is "no, you have to keep paying us," that's the business model talking, not your needs.
10. What won't you build? A consultant who says yes to everything is a consultant who'll waste your money.
11. Local references — can I call them? In Kurdistan, word-of-mouth is still the best filter.

Red flags I see often

• "AI revolution" / "AI is the future" sales talk with no specifics.
• Refuses to give a fixed price without a paid discovery.
• Can't name which LLM provider they use, or the answer changes between calls.
• No portfolio, just slides.
• Asks for full payment up front.
• Promises "we'll train an AI on your business" — usually means nothing technical.
• Disappears between proposal and contract.

Green flags

• Live system you can poke at.
• Fixed price after a written scope.
• Will tell you "don't build this" when it's the right answer.
• Talks about evaluation, monitoring, and what happens when things break.
• Speaks the language(s) your customers actually use.

The local context

In Kurdistan specifically, the AI consultant market is roughly six months behind hype but eighteen months ahead of capability — which is a recipe for being sold something that won't work. Slow down. Ask the eleven questions. Insist on production references. Pay for a written scope before any build.