10 Questions to Ask Before Hiring an AI Vendor (From Someone Who Is One)

We build AI systems for small businesses. Here are the questions we wish every client would ask us — and every other vendor. If a vendor can't answer these clearly, keep looking.

The AI vendor market right now is... chaotic. Everyone from Fortune 500 consulting firms to two-person startups is selling "AI solutions." Some are excellent. Many are not. And if you're a small business owner trying to figure out who to trust with your data and your money, it's overwhelming.

So here's an insider's guide. We're an AI vendor. We have obvious bias. But we also believe that informed buyers make better clients — and that the industry desperately needs more honesty. These are the questions that separate the good vendors from the ones who'll waste your money.

1. "Do I own the system when you're done?"

Why it matters: Some vendors build on proprietary platforms that only they can maintain. If you ever want to switch vendors, modify the system, or bring capabilities in-house, you're stuck. You don't own an AI system — you're renting it.

What a good answer sounds like: "Yes. We build on open-source tools and standard infrastructure. You get the code, the documentation, and the ability to hand it to any competent developer if you want to part ways."

Red flag: "The system runs on our proprietary platform." Translation: you're locked in. If they raise prices, you can't leave. If they go under, your system dies with them.

2. "What happens if you go out of business?"

Why it matters: Let's be real — a lot of AI companies are startups, and most startups fail. If your AI vendor disappears, can your system keep running?

What a good answer sounds like: "Your system runs on your infrastructure. If we disappeared tomorrow, everything keeps working. You'd need someone else for updates and improvements, but the lights stay on."

Red flag: Any hesitation or a response about "we're well-funded" (that's not what you asked). Also a red flag: systems that phone home to the vendor's servers to function. If they go dark, so do you.

3. "How long until my team can use it without you?"

Why it matters: A good vendor builds you something your team can use independently. A bad vendor builds something that requires them to maintain forever — because recurring revenue.

What a good answer sounds like: "Your team will be using it independently from day one of launch. We provide training, documentation, and a support period. After that, we're available for improvements and updates, but you don't need us for daily operations."

Red flag: "You'll need our team to manage the system on an ongoing basis." Sometimes this is legitimate (complex systems do need expert care), but if it's a basic search tool or chatbot, you shouldn't need a vendor on retainer just to keep it running.

4. "What's the ongoing cost — all in?"

Why it matters: The build cost is just the beginning. AI systems have ongoing costs: hosting, API fees (every AI query costs money), updates, and maintenance. A $20K build with $3K/month in ongoing costs is really a $56K first-year investment.

What a good answer sounds like: A clear breakdown of:

• Hosting/infrastructure costs (typically $100–$500/month)
• AI API costs (depends on usage — they should estimate based on your expected volume)
• Maintenance/support retainer (if applicable)
• What you can handle yourself vs. what requires their involvement

Red flag: "We'll discuss ongoing costs after the build." No. You need to know the total cost of ownership before you commit. Also watch for vendors who quote the build cheap and make their money on mandatory ongoing fees.

5. "Can you show me something working in 2 weeks?"

Why it matters: The best AI vendors build iteratively. They can show you a basic working version quickly, then refine it based on your feedback. Vendors who need 3 months before you see anything are either building something unnecessarily complex or don't know what they're doing.

What a good answer sounds like: "Absolutely. In the first two weeks, we'll have a prototype with a subset of your data that you can actually try. It won't be perfect, but you'll see the concept working with real information."

Red flag: "We need to complete the full discovery phase, design the architecture, and finalize the data model before we can show anything." This is waterfall development circa 2005. In AI, you build a small version, test it, and iterate. If someone needs 12 weeks of planning before writing any code, they haven't built enough of these.

6. "What data do you need access to, and where does it go?"

Why it matters: Your data is your business. You need to know exactly what data the vendor accesses, where it's stored, and who can see it. This is especially critical if you handle customer PII, financial data, or health information.

What a good answer sounds like:

• "Here's exactly what data we need access to during the build: [specific list]"
• "Your data stays on your infrastructure / in your cloud account"
• "We don't send your data to third-party AI providers for training"
• "Here's how we handle access controls and audit logging"
• "When the project is done, our access is revoked"

Red flag: Vague answers about data handling. "We take security seriously" without specifics. Any vendor that needs to send your data to their own servers (instead of deploying on yours) should have a very good reason.

7. "How does the AI handle mistakes? What happens when it's wrong?"

Why it matters: AI will be wrong sometimes. The question isn't whether it makes mistakes — it's how the system handles them. Does it hallucinate confidently? Or does it tell you when it's not sure?

What a good answer sounds like: "Every answer includes confidence indicators and source citations. If the AI doesn't have enough information to answer reliably, it says so instead of guessing. We also build in human review steps for any high-stakes outputs — the AI suggests, your team approves."

Red flag: "Our AI is 99% accurate." Nobody can guarantee that. If they claim near-perfect accuracy, they either haven't tested it rigorously or they're measuring accuracy in a misleading way. The honest answer is: "It's very good, but it will make mistakes, and here's how we handle that."

8. "Will this work with my existing tools?"

Why it matters: You use QuickBooks, Outlook, Salesforce, a specific CRM, a project management tool, shared drives — your tools are your workflow. If the AI system requires you to change your tools, adoption will fail.

What a good answer sounds like: "We'll integrate with what you already use. Most of our projects connect to email, CRM, document storage, and communication tools. We assess your stack in the discovery phase and tell you exactly what integrates natively, what needs a connector, and what might not be feasible."

Red flag: "You'll need to migrate to our platform." Unless there's a very compelling reason to switch (like your current tool is end-of-life), you shouldn't have to change your entire workflow to use AI. The AI should fit into your world, not the other way around.

9. "What does 'AI' actually mean in your product?"

Why it matters: "AI" has become the most abused word in technology. Some "AI-powered" products are genuinely sophisticated systems with custom models. Others are a $20/month ChatGPT wrapper with a branded interface. You're paying very different amounts for these — you should know which one you're getting.

What a good answer sounds like: A clear, jargon-free explanation of what the technology actually does. Something like: "We use a large language model (like the technology behind ChatGPT) to understand your questions, then search through your company's documents to find relevant answers. We show you the source documents so you can verify. The AI doesn't make up information — it finds information that already exists in your files."

Red flag: Buzzword soup. "We use proprietary machine learning algorithms with advanced neural architectures and state-of-the-art natural language processing." If they can't explain it in plain English, either they don't understand it themselves, or they're trying to impress you into not asking follow-up questions.

10. "What's the simplest version we could start with?"

Why it matters: The best AI projects start small and expand. If a vendor's minimum engagement is a $75K, 6-month project, they might be great — but they're not right for a small business that should be starting with a focused pilot.

What a good answer sounds like: "Here's what we could build in 4 weeks for a fraction of the full cost: [specific, limited scope]. It won't do everything, but it'll prove the concept with your real data and give your team a chance to see if it works. If it does, we expand. If it doesn't, you haven't bet the farm."

Red flag: An inability or unwillingness to scope something small. "We really need to do the full implementation to see value." No, you don't. If a document search tool can't prove value with 500 documents in 4 weeks, loading 50,000 documents isn't going to fix it.

Bonus: The Meta-Question

After asking all ten of these, ask yourself one more: "Did this vendor answer honestly, or did they tell me what I wanted to hear?"

The best vendors will push back on some of your assumptions. They'll say "actually, AI isn't great for that use case" or "you could solve that problem without AI for a tenth of the cost." They'll talk you out of things that don't make sense.

If a vendor agrees with everything you say and promises the moon, that's your biggest red flag. They're selling, not solving.

A Note About Our Own Bias

We wrote this guide, and we're an AI vendor. Obviously, we think our answers to these questions are strong — that's why we put them on our website. But we'd rather you ask these questions to every vendor (including us) and pick the best fit than hire us because you didn't know what to ask.

The AI industry needs more informed buyers. That's how the good vendors win and the bad ones get weeded out. Ask the hard questions. If someone gets defensive, they're not your vendor.