What "AI-Powered" Actually Means in Advisory Services
A non-technical leader's decoder for the claims you're hearing from every vendor.
The premise
Every vendor now claims "AI-powered."
For a non-technical buyer, the phrase has become useless. It could mean we built something structurally impossible without AI. It could mean we wrapped a language model around a workflow that already worked, and now it works 40% better. It could mean we added a chatbot to the homepage and put a sticker on the box. All three vendors use the same words.
The problem isn't academic. Budget gets allocated to claims that are real AND to claims that are theater, and there's no clean way to tell them apart from the surface. Non-technical leaders end up either over-trusting (everything is AI-powered, so AI-powered means nothing) or over-rejecting (none of it is real, so why bother evaluating).
Neither posture is right. The substantive AI work in 2026 is genuinely transformative — but it sits in a market where 70–80% of the claims are decoration or worse. This note is a decoder. The goal: give a non-technical leader the vocabulary, the questions, and the fast filter to tell which is which.
The pattern: four flavors of "AI-powered"
Every "AI-powered" claim falls into one of four tiers. From most to least substantive:
AI-native. The product is structurally impossible without AI. Remove the model and the product stops working. A research tool that synthesizes a hundred sources in sixty seconds is AI-native — without the model, you're back to manual reading. A code assistant that completes functions in your editor is AI-native — turn off the model and it has nothing to suggest.
AI-augmented. There's a real, load-bearing AI feature that meaningfully improves a specific use case. Remove the AI and the product degrades, but doesn't collapse. AI-assisted meeting summaries from real transcripts. AI-generated draft replies in email. AI-enriched lead data in a CRM. The product had value before AI; the AI feature makes specific parts of it noticeably better.
AI-decorated. The product has an AI feature bolted on for the marketing page. Core product value isn't AI. Users click the AI button twice during onboarding, then never again. The feature exists to put "AI-powered" on the sales deck — not because it changes what the product does. Most "AI-enhanced" SaaS in 2026 lives here.
AI-cosplay. The vendor changed nothing about the product. They added "AI-Powered" to the headline, refreshed the website, and raised prices. Sometimes there's a chatbot. Sometimes there isn't. Either way the product functions identically to its 2022 version.
The rough field distribution, based on what I see in audits — not a survey, just an honest estimate — looks something like: 10% AI-native, 25% AI-augmented, 40% AI-decorated, 25% AI-cosplay. Most of what you're being pitched is in the bottom two tiers.
The frame: five questions to ask any "AI-powered" vendor
This is the interview script. Used straight from a printed copy, it surfaces the tier within fifteen minutes.
- What specifically does the AI do in the product? Not the marketing pitch. The actual function, in concrete terms a junior engineer could implement against. If the answer is "it learns and adapts" or "it's powered by advanced machine learning," you're talking to marketing. AI-native and AI-augmented vendors can answer in one or two sentences. AI-decorated and AI-cosplay vendors hedge.
- What did the product do before "AI" was added — and what does it do now? This forces the vendor to describe a change in capability, not a change in branding. AI-native vendors say "it didn't exist before." AI-augmented vendors describe specific new features. AI-decorated vendors describe new buttons. AI-cosplay vendors describe new pricing.
- What happens to your product if I turn the AI off? The cleanest single question in the interview. AI-native: it doesn't work. AI-augmented: these specific features stop, the rest is fine. AI-decorated: nothing important changes. AI-cosplay: vendor doesn't understand the question.
- Show me a customer who measured the AI's specific impact, separate from the rest of the product. If the AI claim is real, somebody has measured the lift attributable to the AI piece in isolation. If the vendor can't produce that customer, the AI claim is unfalsifiable — which means it's either decoration or cosplay.
- What's your model? Is it your own, or are you calling an API? Not a trick question — both are legitimate. But the answer reveals whether the vendor has technical depth or is reselling someone else's work with a markup. Calling OpenAI's API to power a feature is fine. Calling it while charging enterprise prices for "proprietary AI" is not.
The 30-second sniff test
When you don't have time for the full interview, four shortcuts surface most of the cosplay:
- Open the product's pricing page. If "AI" shows up in tier names but not in the feature list, it's decoration.
- Search the product's documentation for "prompt" or "model." If neither word appears anywhere, the AI claim isn't load-bearing. Real AI products have prompt design and model behavior to document.
- Ask a technical advisor to use the product for ten minutes. If they come back with "it's just X with a wrapper," it's a wrapper. Their pattern-matching is faster than yours.
- Read the product's changelog from twelve months ago. If the product was already "AI-powered" then, with the same description, the AI claim isn't load-bearing — it's been the marketing line through any number of underlying model changes.
The four shortcuts aren't perfect, but together they catch most cosplay and most decoration in well under a minute.
Bottom line
Non-technical leaders don't need to evaluate AI architectures to spend AI budget well.
What they need is a decoder for the claims they're hearing. The four flavors give you the vocabulary. The five questions give you the tool. The 30-second sniff test gives you the fast filter. The real ones survive all three. The cosplay ones don't.
Buy the native and augmented work. Cancel the rest.
Where "AI-powered" lands in your operation is one of the questions the ISA diagnostic surfaces. Take the 90-second version at isaadvisory.com →