Every company you deal with is now an AI company — at least according to its homepage. The label has spread so fast that it has stopped carrying any information: when the accountancy firm, the logistics broker and the project management tool you use all describe themselves in the same words, those words tell you nothing about any of them. What the label’s inflation hasn’t changed is the reality underneath it. Some companies genuinely are rebuilding how they operate around this technology, while most are adding a chatbot to the homepage and the word “AI” to the investor deck, and the gap between those two groups is enormous. It matters whether you’re choosing a vendor, weighing up a job offer, sizing up a competitor, or looking honestly at your own organisation.

This cycle is familiar to anyone who ran a business through the social media gold rush or the big data era (as I did): a technology arrives, the label becomes fashionable long before the substance does, and for a few years the claim and the capability separate almost completely. The useful skill in that period is knowing how to tell them apart — and you can, because genuine transformation leaves evidence all over a business. You can’t rebuild your operations without it showing up in your prices, your hiring, your documentation and your customers’ results. A rebrand leaves no such trail, and the absence is just as visible once you know where to look.

Two name changes, one transformation

This year has offered a neat controlled experiment in the form of two name changes from the same corner of the software world. SaaStr, the SaaS industry’s biggest community and conference, rebranded its annual gathering as “SaaStr AI Annual”: same event, same audience, same format, with the fashionable word inserted where it will be seen. I don’t particularly blame a conference for following its audience’s attention, but it’s a clean example of the cosmetic end of the spectrum: the name changed because the market’s vocabulary changed, and nothing else needed to.

Compare that with Intercom, which in May renamed the entire company after Fin, its AI support agent. The name followed the substance rather than substituting for it. Fin resolves a majority of customer conversations without a human — around 56% for the average customer, up from roughly 25% when it launched — and the business model inverted to match: instead of charging per seat for software that helps human agents, the company charges $0.99 per resolved conversation, which means it only earns when the AI does the work. The operational consequences followed: the queries Fin resolves are answered in seconds rather than sitting in a queue, and the human team’s job has shifted to the minority of conversations the AI can’t finish. In June Salesforce agreed to buy the company for $3.6 billion. Whatever you think of the price, nobody pays that for a chatbot on a homepage. One company changed its name to keep up with the times; the other changed its name because the technology had already changed the company underneath it.

What real transformation touches

What separates the two examples is where the change lives, and it gives us the general principle: a genuine AI pivot changes a company’s economics, not just its messaging. When AI is really doing work that people used to do, something structural has to move. Pricing shifts, because the cost of delivering the service has shifted — per-resolution and usage-based models appear where per-seat licences used to be. Roles change, because someone has to feed, supervise and improve the systems: you start seeing job titles that didn’t exist two years ago, data and machine learning roles in companies that never employed them, and postings for people to run AI operations rather than build slide decks about them. The way work flows through the business changes too, and with it the outcomes customers can measure: faster response times, shorter delivery cycles, prices that would have been impossible under the old cost structure.

The washing, by contrast, lives entirely in the messaging layer. The tell-tale pattern is AI that features heavily in the earnings call and the press release but is nowhere to be found in the product documentation, where plenty is claimed and nothing specific is shipped — often accompanied by “AI-powered” features that behave suspiciously like the rule-based automation the company was selling five years ago under a different name, and a bold transformation narrative from a business that hasn’t made a single technical hire to deliver it. We’ve already seen the same gap between claim and evidence in the way CEOs blame AI for job cuts they were going to make anyway — the label takes the credit, or the blame, while the organisation carries on unchanged.

The five-point test

None of this requires inside information. Almost all of the evidence is public (which means the test works just as well on a competitor as on a vendor or a prospective employer), and an hour with it will tell you more than any keynote. When I want to know whether a company’s AI story is real, these are the five places I look:

  1. Check the product changelog. The changelog rather than the announcements page: a company genuinely rebuilding around AI ships AI features continuously, documents them, and iterates on them in public. If the last substantive AI entry is months old and coincides with a funding round, the story is running well ahead of the product.
  2. Check the job postings. Hiring is the most honest signal a company broadcasts, because it costs money. Look for data engineers, ML roles, AI operations leads — and for AI expectations appearing inside ordinary roles. If nobody has been hired or retrained to deliver the transformation, it’s unlikely to be happening.
  3. Check the pricing model. Structural AI adoption changes what things cost to deliver, and sooner or later that reaches the price list: usage-based tiers, outcome-based pricing, or simply prices the old cost base couldn’t have supported. If the pricing page looks exactly as it did in 2023, ask what the AI is actually doing.
  4. Check the workflow documentation. Implementation guides, process maps, help-centre articles — anywhere the company shows where AI enters the work. Then ask the vendor, or the interviewer, a simple question: what does your team do differently on a Tuesday because of AI? Real adopters answer instantly and specifically, because the change is their daily life. Washers answer with strategy.
  5. Check the customer outcomes. Resolution rates, turnaround times, error rates, cost per transaction. Transformation you can’t measure from the outside is indistinguishable from no transformation at all. Intercom publishes its resolution rates; that willingness to be measured is itself a green flag.

The test has one limitation: it’s better at exposing washing than at crediting early-stage effort, because a company six months into a genuine rebuild can still look cosmetic while the evidence is yet to reach the changelog or the price list. I don’t have a reliable way to tell those two apart at a single glance, so don’t try — run the same check a couple of quarters later. A real rebuild will have left new footprints in most of these places; a rebrand will have left another press release.

Pointing the test at yourself

The least comfortable use of this checklist is on your own organisation, and it’s the most valuable one. Plenty of businesses that would never dream of AI-washing their customers are cheerfully AI-washing themselves: a strategy document, a pilot programme, a few licences, and a leadership team that talks about transformation while the actual working practices carry on exactly as before. The five questions work just as well internally. Has our own changelog — the things we ship, or the way we deliver — changed? Have we hired or retrained anyone? Has anything about our cost structure or pricing moved? Would our own staff give a specific answer about Tuesday, or a strategic one? Can our customers measure any difference?

If the answer to most of those is no, then whatever the board pack says, the transformation hasn’t started — and that’s worth knowing while it’s still cheap to fix, because the checklist doubles as a starting list: each question that gets a no is a place to begin. The companies that have genuinely rebuilt around this technology tend to lead with what changed: the resolution rate, the turnaround time, the price. As a rough rule for the next few years of this cycle, the more work the label “AI” is doing in the pitch, the less work the AI is doing in the business.