Scroll through the release notes for any software your firm uses and the same word turns up again and again. CRM platforms, financial planning tools, back-office systems, practice management software: in the last eighteen months, almost all of them have shipped something they call AI. Some of it is useful. A lot of it is a search box with a new name.

That’s the predictable result of a market where “AI-powered” has become something buyers expect to see, whatever the product does. Vendors know firms are under pressure to show they’re adopting AI, so they label existing features as AI, put them in a premium tier and wait. Plenty has been announced. Far less has changed how anyone works.

Almost everything uses AI in some sense now, so asking whether a tool uses AI doesn’t tell you much. Ask whether the feature changes a real piece of work in a way you can measure.

What AI-washing actually looks like

AI-washing usually takes one of three forms, and the most common is rebranded search. A feature that used to say “search your client records” now says “ask your data anything”. Underneath, it’s the same keyword lookup it always was. The screen has changed and the capability hasn’t.

The second is the added summary. A dashboard that used to show a table now shows a paragraph above the table. The paragraph says something like “your top five clients by revenue are…” and lists what was already on screen. That’s a language model reading data you could already see and writing text you didn’t ask for.

The third is the assistant that only works in demos. You’ll have seen it: a vendor walkthrough where the AI produces a clean, accurate answer from a carefully written prompt. Then you try it on your own data and it invents a client’s pension value, misreads a date range, or gives up because the document was too long. It works in controlled conditions. It doesn’t work in yours.

The test that matters is whether anyone in your firm has changed how they spend their time because of the feature.

A plain-English checklist for evaluating AI features

Run through these five questions before you upgrade a licence tier for an AI feature, and before you write one off as worthless.

  • Does it replace a step I do by hand, or add a new one? Useful AI takes work away. AI-washing adds a new screen to manage on top of the old one. If using the feature means exporting data, pasting it somewhere, checking the output and then typing it back into the original system, that’s more work, not automation.
  • Can I describe the before and after in concrete terms? “Before: forty minutes every Monday pulling the weekly client review list together from three systems. After: it happens automatically and takes four minutes to check.” That’s a real change. “Before: I searched for things. After: I still search for things, in full sentences” isn’t.
  • Is it working on my data, or on a general model that doesn’t know my firm? This matters a lot in a regulated firm. A general assistant that can answer questions about pensions legislation is different from a system that can answer questions about your client’s pension, using the file your firm holds. The second needs grounded AI, where the answers are restricted to a specific knowledge base. Many vendor “AI features” are the first dressed up as the second.
  • Who carries the liability if the output is wrong? Most platform vendors would rather you didn’t ask. If the AI suggests something that turns out to be wrong and a client is harmed, it almost certainly won’t be the vendor. The FCA and Bank of England’s 2026 joint statement made clear that AI governance for UK-regulated firms is moving from voluntary best practice towards formal expectation, and “the platform said so” won’t be an acceptable answer.
  • Is it solving a problem I have? Vendors build AI features around what they can feasibly build, not around what your operation needs. A planning platform might ship AI that drafts client emails. If your bottleneck is suitability report sign-off, that feature doesn’t help however well it works.

How to apply the L1/L2/L3 lens here

A vendor AI feature that passes the checklist is effectively a Level 1 (education) or Level 2 (integration) fix that someone else has built for you. That’s a good thing. If a tool you already pay for takes manual work out of a real process, use it. It costs nothing extra and you can start today.

One of the top hotels in the UK came to me thinking it needed an outside partner and probably new tools. It already had Microsoft Copilot across the business, and staff were using it, just not fully. Most of what it wanted could be done in Copilot, and the parts Copilot couldn’t do became a short, clearly scoped list instead of a big new project. Finding out what your current tools can already do is usually the cheapest step there is.

The trouble starts when a vendor feature fails the checklist and you’re still paying for it, or when it nearly solves the problem and you find yourself considering a separate Level 2 (integration) project to patch the gaps in a tool that was meant to make life simpler.

The more expensive mistake is assuming that because your software vendor has shipped AI, your firm’s AI needs are covered. What your platform can do and what your workflows need are separate questions. Mixing them up is how firms end up paying for three AI features across different tools, none of which connect, and none of which have changed how anyone spends a Tuesday afternoon.

What good looks like

A useful AI feature has an owner. Someone in your firm can name it, say what it does and tell you when it last saved them time. If nobody can, it isn’t being used, whatever the renewal pitch says.

It has a measurable before and after: four hours saved a week, twelve fewer manual exports a month, one step removed from a six-step process. If you can’t put a number on it, the feature is either too new to judge or not being used.

And it fits the way your work is set up rather than a generic version of it. It runs on your data, in your systems, in the way your firm handles information. The closer a vendor’s feature gets to that, the more seriously it’s worth taking. If you’re weighing a vendor’s agent features specifically, the first-agent playbook covers when to buy and when to build.