How to Spot AI Hype and Vendor Snake Oil: A Buyer's Checklist
Karani Geoffrey6 min readHere is my thesis, and I will not soften it: most AI pitches that land in a Kenyan or Nigerian founder's inbox are engineered to exploit fear, not to solve a problem. The fear is that you are falling behind. The vendor's job is to convert that fear into a signature before you ask the questions that would kill the deal. This article gives you those questions.
I am not anti-AI. I run a technology business and I use AI every day. I am anti-getting-fleeced. And the data says the fleecing is real: the MIT State of AI in Business 2025 report found that 95% of enterprise generative-AI pilots deliver no measurable impact on the bottom line. Somebody sold all 95% of those projects. Do not be the next buyer in that column.
Why African buyers are a soft target
Global vendors know something about our market: compute is expensive here, budgets are tight, and the pressure to look modern is intense. Cloud GPU costs rose more than 300% globally between 2023 and 2025, and in Nigeria, Kenya, and Ghana cloud compute can run 25 to 40 percent higher than the equivalent service in Europe or North America. That means the exact same AI product costs you more to run and returns less, yet it is often sold to you at a premium because you are seen as under-served and eager.
The snake-oil seller counts on three things: that you will not ask for evidence, that you will not run a bake-off, and that you will confuse a slick demo with a working system. Let us remove all three advantages.
The 12-question vendor checklist
Print this. Put it in front of every AI vendor before money changes hands. If a rep dodges more than two of these, that is your answer.
| Question to ask | What a good answer sounds like | Red flag |
|---|---|---|
| What specific problem does this solve, in my words? | They restate your workflow accurately | They pivot to "the power of AI" |
| Is there real machine learning here, or rules and a chatbot wrapper? | Honest about what is ML vs. scripted | Vague "proprietary AI" with no detail |
| What model powers this and who hosts it? | Names the model and provider | "We can't disclose that" |
| Can I see it run on MY data, not your demo data? | Yes, in a scoped trial | Only canned demos |
| What is your accuracy on a case like mine, and how measured? | Numbers with a method | "99% accurate" with no context |
| Where does my data go and who trains on it? | Clear data residency and opt-out | Silence or legalese |
| What does this cost at full production volume? | A usage-based estimate | Only the pilot price |
| Who are three customers my size I can call? | Real references, ideally regional | Only enterprise logos on a slide |
| What happens when the model is wrong? | Human review, fallbacks, logging | "It's rarely wrong" |
| How do I export my data and leave? | A clear exit path | Lock-in by design |
| What is the total cost of ownership over 24 months? | Licences plus compute plus staff | Only the sticker price |
| What can this NOT do? | An honest list of limits | "It can do anything" |
The tells of snake oil
Beyond the checklist, train your ear for these patterns. They show up in almost every inflated pitch.
1. The unfalsifiable claim
"Our AI increases revenue by up to 40%." The phrase "up to" means the floor is zero. A real number sounds like "a logistics client in Nairobi cut invoice processing time from 6 minutes to 90 seconds; here is the case study." If a claim cannot be proven false, it is marketing, not evidence.
2. The demo that never touches your reality
Demos run on clean, hand-picked data in a sandbox. Your business runs on messy Swahili-English customer messages, inconsistent spreadsheets, and patchy connectivity. The gap between demo and production is where most projects die. Insist on a trial with your data before you believe anything.
3. AI as a noun, never a verb
Watch how often the vendor says "AI" as a magic object versus explaining what it actually does to your workflow. Snake oil talks about AI. A real solution talks about your accounts-receivable process.
If a vendor cannot explain, in one plain sentence, which task their product removes from a specific person's day, they are selling you a feeling, not a tool.
4. Regulatory theatre
In September 2024 the US FTC launched "Operation AI Comply," a crackdown on exactly this behaviour. It has already targeted companies for a fake "AI Lawyer" service, tools that mass-produce fake reviews, and a firm called Air AI for overstated claims about autonomous agents replacing staff. The lesson for you: regulators in mature markets are now suing vendors for the very claims you are being pitched. If it would get a company sued in Washington, do not sign for it in Westlands.
Run a bake-off, always
The single most powerful thing you can do is refuse to buy from one vendor in isolation. Shortlist two or three, hand each the same real task, define your success metric in advance, and let them compete on your data. A vendor who is confident will accept. A vendor selling snake oil will suddenly have scheduling problems.
- Define the metric first. "Correctly categorise 200 of our past support tickets." Not "impress us."
- Use your own messy data. Include the edge cases, the code-switching, the bad scans.
- Set a walk-away number. Below X accuracy or above Y cost, it is a no. Decide before you fall in love.
- Score blind if you can. Remove branding so the flashy pitch does not bias the result.
The uncomfortable question: do you even need this yet?
Sometimes the honest answer is that your business is not ready, your data is not clean, or the problem does not need AI at all. That is not a failure. It is a saved budget. Before you buy anything, read my longer argument on this in why you should not get pressured into buying AI before your business is ready. Timing is a strategy.
The bottom line
You do not need to be a data scientist to avoid snake oil. You need to be a disciplined buyer who asks for evidence, tests on real data, and is willing to walk away. The vendors selling genuine value will respect that process, because it makes them look good. The ones selling hype will resent it, because it makes them disappear. Let it. Your budget is too hard-earned to fund somebody else's marketing experiment.
Ask the twelve questions. Run the bake-off. Trust the numbers, not the theatre. That is how you stay on the right side of the 5%.