The AI Vendor Red Flag Checklist: 8 Warning Signs Before You Sign

An e-commerce company interviewed three AI vendors last month. One promised "40% productivity gains" with a six-month implementation (£150K). Another offered an "end-to-end AI transformation" with vague deliverables and a proprietary platform they wouldn't let the client own. The third said upfront: "Here's what we'll build, here's the timeline, here's what success looks like, and here's the cost." They chose the third and saved themselves a year of regret.

The AI vendor market exploded in 2024-2025. Many consultancies rebranded overnight. "We're an AI company now." Most aren't. Most are selling hype and lock-in. Some are selling nothing at all, just expensive theory and proprietary software you'll be stuck with.

Why Vendor Selection Matters (And Most Companies Get It Wrong)

Bad vendor choice compounds over time. Month one: you're optimistic. Month three: you realize the deliverables were vague. Month six: you're locked in to their platform and you can't leave. Month nine: the project is dead but you're contractually obligated to pay through the end of the year.

I've seen it happen across 120+ implementations. The companies that got burned didn't hire bad vendors. They hired vendors that looked good in the sales call but had no accountability once the contract was signed.

The 8 Red Flags

Run these checks in every vendor conversation. If you spot three or more, walk away.

Red Flag 1: "Productivity Gains" Without Specifics

Vendor: "We'll improve your productivity by 35%."

You: "For which processes? How are you measuring it? What happens if we don't hit it?"

Vendor: "Well, it depends on your baseline..."

They don't have a specific answer. They're quoting industry averages, not your situation. Productivity is unmeasurable without definition. 35% of what? Time? Output? Errors prevented?

Good vendors say: "We'll automate your invoice processing from two hours per day to 15 minutes per day. That's 7.75 hours of labor freed up per week. At your salary rate, that's £X per week in recovered time. Here's how we'll measure it: we'll track time logs before and after."

Red flag: generic percentages, vague metrics, hand-wavy claims.

Red Flag 2: Vague Deliverables (The Contract Says "AI Solutions")

Vendor: "We'll implement an AI solution for your customer service."

You: "What does that mean? Will you build a chatbot? Integrate with our existing system? Who owns it afterwards?"

Vendor: "That's something we'll figure out during the engagement."

Translation: "We haven't thought about what you actually need. We'll charge you to discover it." Vague scopes end in scope creep. Scope creep ends in blown budgets and missed deadlines.

Good vendors show you: a signed specification before work starts, including what they'll build, what it will connect to, what it will do, and what happens when they're done (handoff, training, documentation).

Red flag: contract uses words like "solution," "transformation," "implementation" without defining what's actually being built.

Red Flag 3: No Measurable Success Criteria

Vendor: "We'll succeed if you feel like your workflow is more efficient."

That's not a success criterion. That's subjective. You feel efficient when coffee is good.

Good vendors say: "Success is: (a) the system processes 100+ invoices per day with less than 2% manual review, (b) error rate drops from 8% to under 1%, (c) your team spends under 30 minutes per day on this process instead of 3 hours. We'll measure these weekly."

Get criteria in writing. Specific numbers. Measurable. Tied to business outcomes. If the vendor can't name three specific success metrics, they don't have a plan.

Red flag: success criteria that can't be measured, or no criteria at all.

Red Flag 4: "AI-Powered" Everything (Even Things That Don't Need AI)

Vendor pitch: "We'll use AI to optimize your scheduling."

You: "What does that mean? Are you using machine learning? Algorithms? What's different from a rule-based scheduler?"

Vendor: "We use... AI. Advanced algorithms."

They're slapping "AI-powered" on standard software to charge premium prices. Sometimes the right answer isn't AI. Sometimes it's a spreadsheet. Sometimes it's a workflow rule. Sometimes it's hiring the right person.

Good vendors say: "We tested three approaches. Rule-based scheduling solves 80% of the problem and costs £5K. Machine learning improves it to 85% and costs £25K. For your use case, rule-based is the play. You'll get ROI in month one. ML costs too much for the delta."

They prioritize your outcome over their feature set.

Red flag: every proposal is AI, every problem requires their AI platform, no options, no alternatives.

Red Flag 5: Proprietary Platform (You Don't Own the Output)

Vendor: "The automation runs on our proprietary platform. You'll pay per transaction."

Translation: You're renting. Forever. If you outgrow them or prices change, you're stuck moving the automation to a new platform (expensive) or staying and paying forever (expensive).

Good vendors: "We'll build this using standard tools (Zapier, Make, your existing system). You own the automation completely. If you leave us, you take it with you. No lock-in."

Proprietary platforms make sense in rare cases (highly specialized, complex logic, they're genuinely the only ones who can do it). Most AI projects don't need proprietary platforms. If they're pushing one, ask why. Push back if the answer is weak.

Red flag: proprietary platform, per-transaction pricing, lock-in clauses, you don't own the output.

Red Flag 6: No Track Record for Your Industry

Vendor: "We've done AI projects before."

You: "Any in finance?"

Vendor: "Uh... we can learn your industry quickly."

They've never done this before. They'll charge you to learn. You'll be their training project.

Good vendors show: three past projects in your industry, specific outcomes (not anonymized vagueness), references you can call, case studies with numbers.

You don't need them to be experts in your exact company. But they should have shipped similar projects in your sector. Ask directly: "Have you automated customer service for a SaaS company? Have you built compliance automations for financial services?" If they equivocate, they haven't.

Red flag: no specific examples, no industry references, "we'll learn quickly," overconfidence about a new sector.

Red Flag 7: No Clarity on Timeline ("It Depends")

You: "How long will this take?"

Vendor: "It depends on complexity. Could be three months, could be nine months."

That's not a timeline. That's a blank check. Projects with no fixed endpoint blow budgets and timelines.

Good vendors give you: a fixed timeline with milestones (week 2: requirements and scope sign-off; week 6: first automation live; week 10: full implementation and training), deliverables by date, and a mechanism for scope management ("if requirements change, timeline adjusts").

Fixed timeline doesn't mean no flexibility. It means they've committed to delivering something by a date, and they'll manage scope to make it happen.

Red flag: vague timelines, "it depends," no milestones, no commitment to a ship date.

Red Flag 8: Expensive Discovery Phase (They Want to Charge You to Figure Out What You Need)

Vendor: "Let's start with a £30K discovery phase. Then we'll know if this is even possible."

Maybe. But also, maybe they're charging you to do their job. A good vendor can scope most AI projects in two hours of conversation. If they need a £30K engagement to tell you what's possible, they don't have experience.

Good vendors: offer a free scoping call (30 minutes), write up a rough proposal (free), then either move into implementation with a clear contract, or advise you it's not the right time.

Some discovery is normal for complex projects. But it should be short (one week, not one month) and it should feed directly into implementation (not be a separate, expendable phase).

Red flag: expensive discovery, extended timeline, unclear path from discovery to actual work.

Questions to Ask in the Sales Call

Use these verbatim. Listen for specifics, not smooth answers.

On scope: "Walk me through exactly what you'll deliver. I need to see it in writing before we sign anything."

On success: "What would it look like if this project failed? What metrics would tell us that? And if we hit those bad metrics, what happens?"

On ownership: "After you're done, who owns this automation? Can we modify it? Can we move it to another platform?"

On timeline: "What's the fixed end date? What happens if we go over?"

On track record: "Show me three similar projects you've done. I want to talk to the clients."

On cost changes: "If the scope stays the same, does the price stay the same?"

On accountability: "If we miss the success metrics, what's your responsibility?"

Vendors who answer these directly and specifically are usually legit. Vendors who hedge, redirect, or push back are usually hiding something.

The One-Page Vendor Evaluation Scorecard

Use this to compare vendors fairly:

Vendor A | Vendor B | Vendor C

Has delivered similar projects in our industry? Yes/No

Specific deliverables in writing? Yes/No

Measurable success criteria? Yes/No

Fixed timeline with milestones? Yes/No

Clear ownership (we own the output)? Yes/No

No proprietary lock-in? Yes/No

References I can call? Yes/No

Reasonable discovery phase (under one week)? Yes/No

Score (out of 8):

Vendor A: 6/8

Vendor B: 4/8

Vendor C: 8/8

Choose Vendor C. Not because they're the cheapest, but because they've committed to something measurable.

What Good Vendors Do Differently

They talk about your problem, not their platform. They ask questions. They say "no" sometimes ("We don't have experience in that sector, so I'd recommend talking to X instead"). They show examples of their work. They're willing to give you a fixed price and timeline. They talk about what you'll own when they leave. They act like they're building something with you, not selling something to you.

That's the pattern. It's not complicated. Most vendors just don't do it.

Frequently Asked Questions

Is it bad to use their proprietary platform if they own it?

Not always. If it's a very specialized tool and they're genuinely the only ones who can maintain it, maybe it makes sense. But get a buyout clause in the contract: if you leave or they go out of business, they hand you the code and the data. And negotiate transaction fees downward if you commit to volume. Don't just accept their terms.

How do we know if a vendor's past projects are real?

Call the references. Ask: "Did the project ship on time? Did it meet the metrics you were promised? Would you hire them again?" If they're vague, the project probably wasn't successful. Good vendors have happy clients who'll tell you the truth.

What if the vendor is cheap but sketchy?

Cheap plus sketchy is the most expensive option. You'll end up rebuilding their work or paying someone else to fix it. Spend more upfront for clarity and track record. You'll save money in the end.

Can a small vendor be good, or do we need a big consultancy?

Small vendors can be excellent. Big consultancies can be terrible. Size doesn't matter. Track record matters. References matter. The vendor scorecard applies to anyone. A one-person shop with a clear proposal and five happy reference clients is better than a 50-person firm with vague deliverables and no references.

What if we're already locked in with a bad vendor?

Look for an exit clause in your contract. Renegotiate the deal: either they improve deliverables and timeline, or you terminate with minimal penalty. If the contract is ironclad, you're stuck, but the next vendor you hire won't make the same mistakes.

Richard Batt has delivered 120+ AI and automation projects across 15+ industries. He helps businesses deploy AI that actually works, with battle-tested tools, templates, and implementation roadmaps. Featured in InfoWorld and WSJ.

What Should You Do Next?

If you are not sure where AI fits in your business, start with a roadmap. I will assess your operations, identify the highest-ROI automation opportunities, and give you a step-by-step plan you can act on immediately. No jargon. No fluff. Just a clear path forward built from 120+ real implementations.

Book Your AI Roadmap, 60 minutes that will save you months of guessing.

Already know what you need to build? The AI Ops Vault has the templates, prompts, and workflows to get it done this week.