AI washing: what is actually safe for small businesses after KPMG

On 13 June 2026, KPMG published a report titled "Redefining excellence in the age of agentic AI". Within days, it was gone. According to reporting that followed the withdrawal, a large proportion of the citations reportedly did not check out, and some of the organisations named disputed the AI deployments attributed to them. KPMG did not publish a detailed explanation.

Around the same time, the Builder.ai story was doing the rounds again. This is AI washing in plain sight: a company marketed as an AI-powered software platform reached unicorn valuation, only for it to emerge post-collapse that it had been backed by a large workforce of human developers doing work the AI was supposed to be doing. The "AI" was, in significant part, a label.

These are not isolated incidents. They are the visible end of something that runs through the whole industry right now.

Why the overclaiming happens

AI is under pressure to perform before it is ready. Consultancies need to publish thought leadership. Startups need to justify valuations. Software vendors need to add "AI" to their feature list or risk being seen as behind. The pressure to claim capability is immense, and the gap between a plausible-sounding claim and a verified one is easy to paper over in a well-designed deck.

The result is that a lot of what gets called "AI" in the market right now is one of three things: a thin wrapper around a general-purpose language model, a human-in-the-loop process that would not scale without the humans, or a genuinely capable system being applied to a problem it cannot reliably solve yet. Sometimes it is all three at once.

None of this means AI does not work. It means the marketing language has run ahead of the evidence, and the people buying on the basis of that language are taking on risk they cannot quantify.

What to watch for when a vendor pitches you

The most useful question you can ask any AI vendor is simple: can you explain the mechanism? Not what the tool does, but how it does it. If the answer is vague - "proprietary AI", "intelligent automation", "our model learns your business" - that is not an explanation. It is a placeholder.

A few other tells worth knowing:

Outcomes with no denominator. "Our clients save 40% of their time" means nothing without knowing 40% of which tasks, across how many users, measured how. Ask for a specific example with a specific customer.

Named case studies that cannot be verified. If a vendor names an organisation that implemented their tool, you can ask that organisation. If the vendor is reluctant to connect you with a reference, that is useful information.

AI described as the lead benefit, not the mechanism. When "AI" is the headline and "what it actually does for you" is buried in slide three, the sales motion is built around the label, not the result. The label should not matter. The hours saved should.

No mention of what it cannot do. Any honest AI tool has limitations, failure modes, and tasks it should not be trusted with. If a vendor does not name any of these, they either do not know, or they are hoping you will not ask.

What is genuinely safe and measurable at small-business scale

The AI that works reliably for small businesses in 2026 is not the AI that makes the press. It is quieter and more limited, and that combination is exactly why it works.

Bounded workflow automation covers tasks with a clear input, a consistent output format, and a low cost of error. Drafting a first version of a follow-up email from your notes. Summarising a long meeting transcript to three action points. Pulling key figures from a supplier document so you can scan them in two minutes rather than fifteen. Generating a first draft of a job advert from a bullet-point brief.

These tasks have something in common: you are already checking the output anyway. The AI is saving you the time to produce a starting point, not replacing your judgement about whether the starting point is good enough. The loop stays short. The risk stays manageable.

This is measurably different from deploying an autonomous agent to handle customer queries, automate complex decisions, or manage external relationships on your behalf. Those applications exist and some of them work well. But they require more investment, more oversight, and more understanding of what can go wrong. They are not where most small businesses should start. The Birmingham haulage AI case study is a useful example of what a staged, bounded approach looks like when it works at scale - the firm automated 400 daily calls but only after the underlying workflow was understood and structured first.

The step that the hype cycle skips

The KPMG report was about agentic AI - systems that take sequences of actions to complete complex goals. The Builder.ai story is about a platform that marketed transformation and delivered something much more ordinary. Both cases, in different ways, collapsed because the claims were not anchored to reality.

For a small business owner being pitched AI right now, the equivalent risk is smaller but still real. It is buying a tool before you know which problem it is solving. A ChatGPT subscription that nobody uses because nobody mapped the workflow first. A transcription tool that produces summaries your team ignores because the real bottleneck was never the transcript. An "AI assistant" that automates a task that took twenty minutes a week. The post on the five AI mistakes UK small businesses keep making names these patterns directly, with a fix for each one.

The diagnostic step is not exciting. It does not make for a good press release. But it is what separates the businesses that see a return from AI this year from the ones that have a line item on the P&L and nothing to show for it.

Before you buy anything or sign up for any trial, spend time mapping the tasks that are actually eating your hours. Not the tasks you assume are the bottleneck, and not the ones the vendor wants to address. The ones that, if they took half as long, would make a real difference to your week.

That is the question the HoursBack Assessment is built around. A 60-minute conversation, a structured report within two working days, and a five-day plan written in plain English - with specific tools recommended only where there is a clear case for them.

If you want to start with something quicker, the free two-minute quiz gives you a rough read on where your business stands. Or if you are ready to go deeper, find out more about the Assessment.