So, that means chatting with those who are doing this day-to-day. And that meant my first call was to Dan Boyles.

Dan runs Hello AI Collective, a year-old consultancy that works both sides of the table: vendors like Microsoft, Lenovo, Nvidia and Sage send him their clients to drive AI transformation, and he also works directly with brands like Monster Energy, Formula One teams and Premier League clubs. Before that, he spent eight years at Microsoft (two of them as one of only twelve Global Master Trainers). His entire job now is answering one question for businesses: is this actually making you money?

It might be a simple question, but it’s finally starting to get a lot more scrutiny.

56% of CEOs report no increase in revenue and no drop in costs from their AI spend in the last twelve months (Source: PwC’s 29th Global CEO Survey, 4,454 CEOs, January 2026).

We’ve bought the tools expecting them to magically deliver business outcomes. But for the majority of businesses, this is not happening.

For the first year that Dan’s team ran Copilot deployments, 80% of the projects failed. Whilst Copilot has its ‘quirks’, this wasn’t the tool that was at fault, it was the approach that was: Give everyone basic training and let them figure out the rest. This approach to training works for typical software products, where the same input always gives the same output. It doesn’t work for a tool where the result depends entirely on how well someone can think about the problem in front of them.

His team rebuilt their whole model around three words: Reason, Solution, Scale. Before any tool gets switched on, they sit the leadership team down and ask why they’re doing this at all. “We need to be more competitive” isn’t a reason. “We’re at 8% EBITDA and need to hit 13% before an acquisition” is. Once there’s a real business reason and a clearly defined objective, they trace it down through every department until they find the specific, boring bottleneck that’s actually costing money.

Where real organisation transformation happens

One client, a $1.2 billion turnover company, had a quoting process that took five hours per job because someone had to pull data from thirteen different places and shuttle a PDF between people. Dan’s team spent four and a half months rebuilding that one workflow. EBITDA moved from 8% to 9.3%. That’s $17 million in extra revenue, from a single process.

A second client, a distribution business processing 8,000 invoices every fortnight across a team of 40 accountants, was bleeding money to late payment fees because invoice dates didn’t match what was logged in their system. Dan’s team built an agent to flag every mismatch automatically. £332,000 saved a month.

Neither of those started with “let’s use AI.” Both started with someone sitting inside the business for weeks, mapping exactly where the time and money were actually going. The tool was the last five per cent of the job, not the first.

The instinct with any new tool is to ask, “how do I use this to save money fastest?” But efficiency is table stakes now. The businesses that are seeing returns from AI are the ones that are re-thinking their processes (as with Dan’s examples above) and those that are using AI to achieve something that wasn’t possible before.

Dan’s view is that the businesses that win are the ones using AI to make the customer experience itself unbeatable. Dan’s own accountant sends him a full call summary and quote within two minutes of hanging up. She’s more expensive than the next accountant he could find. He’s never once considered switching.

How ‘uneven’ AI capability is a hidden drag on a business

AI capability inside a business is nowhere near evenly spread. Dan told me about one private equity client where a graduate had built custom automations for $500 a month, while someone else on the same floor didn’t know how to turn on dark mode. And it’s often the people who’ve been in a role for fifteen years, not the recent graduates, who spot the biggest opportunities, because they’re the ones who’ve lived with a bad process long enough to actually hate it.

The traditional fix, classroom training and hackathons, gets you so far. What actually drives change is the opposite: Micro-learning tailored to a person’s role, along with freedom to experiment and a structured process of sharing workflows and best practices across an organisation. A training session is an event. Capability is a state, and most organisations are still only buying the first one.

You can check out the full chat here:

Hope you find the discussion useful,

Andy