For the last couple of years, AI has enjoyed something most technology investments could only dream of: a licence to experiment, explore, and occasionally get things spectacularly wrong. All in the name of figuring out what this stuff could actually do for business, of course...
The business case didn't always need to be particularly sophisticated because, at the beginning, simply understanding the possibilities had value. Businesses bought Copilot licences, launched AI working groups, experimented with ChatGPT, built internal assistants, tested agents, and squeezed AI functionality into almost every piece of software they could. Because nobody wanted to be the organisation that got left behind.
(FOMO is a surprisingly effective technology investment strategy.)
For a while that was enough. But, as we are starting to plan for 2027, the mood has changed. We've seen the demos, we've played with the tools, and we've sat through enough presentations about how AI is going to revolutionise everything from customer service to making a cup of tea. Now the novelty is wearing off, and a much more commercially minded question is on the tip of every CFO’s tongue:
“Cool, but does it actually make us money?”
Welcome to the next phase of AI, where experimentation has to start earning its keep.
Let's clear something up before this starts sounding like another blog predicting the death of AI, we know businesses aren't abandoning it. Certainly not!
Quite the opposite, in fact, with Gartner reporting that 85% of functional leaders expect their AI spending to increase in 2026. That’s despite only 22% of organisations saying they have successfully scaled AI across multiple business units or adopted an AI-first approach.
Businesses clearly believe there is value to be found in AI. The problem is working out exactly where that value appears – and proving it. The question is shifting from whether organisations should invest in AI to whether individual AI investments deserve to survive.
During the experimentation phase, asking “Can AI do this?” made perfect sense because organisations needed to understand what was possible before they could make informed decisions about where to invest.
Now the questions need to get harder:
AI investment isn't slowing down. Patience with investments that can't answer those questions probably is.
An AI tool saves an employee five hours every week, an automated workflow removes 200 hours of manual work every month, or an AI assistant makes a team 30% faster at completing a particular task.
If those five hours allowed someone to serve more customers, generate more opportunities, produce better work, make decisions faster, or remove the need for additional resource, there is a clear route from productivity to business value.
If everyone simply gets through their inbox a little faster, though, good luck explaining the financial impact to the CFO.
Because time saved isn't automatically ROI.
It's a challenge that Gartner's 2026 CFO research brings into focus. 45% of finance AI investment lean towards productivity compared with only 20% focused on improving decision quality. Even though boards have been placing greater value on outcomes including growth, competitive advantage, and better decision-making.
There’s another uncomfortable possibility behind disappointing AI returns… Maybe the AI isn’t the problem.
Plenty of businesses have approached AI by looking at their existing processes and asking: where can we stick some AI?
We understand… it’s quicker, easier, and involves considerably fewer awkward conversations about why the process has been broken since 2017. But adding AI to an inefficient workflow doesn’t magically make it a good workflow. Sometimes, you’ve just given the inefficiency a turbocharger.
Take sales proposals. AI could help a salesperson write one 30 minutes faster. Great.
But if that proposal still spends three days bouncing between inboxes waiting for approval... we’re probably celebrating the wrong 30 minutes.
The bigger opportunity is redesigning the process itself. AI could research the account, pull relevant CRM data and case studies, create a first draft, check it against brand and compliance requirements and route it to the right person for approval.
Now we’re getting somewhere. You’re no longer asking AI to help someone type faster. You’re asking whether they needed to be doing all that typing in the first place.
Old process + AI = same process, shinier hat.
Redesigned process + AI = potentially different economics.
And that distinction is going to matter a lot more as the novelty wears off.
Saving time sounds brilliant on a PowerPoint slide but turning it into an actual financial return is slightly trickier.
Say AI saves someone 30 minutes on a task. You can multiply those 30 minutes by their salary and produce a lovely-looking number for the ROI spreadsheet. But unless the business does something productive with that extra capacity, you haven’t necessarily saved any money. You’ve created 30 minutes.
Useful? Absolutely. The same as £25 appearing in the company bank account? Not quite.
And the further AI moves into complex business processes, the messier attribution gets. If AI helps a salesperson spot a better opportunity, how much of the eventual deal belongs to AI?
If an AI security system prevents an outage, what’s the ROI of something that didn’t happen? This doesn’t mean AI isn’t creating value. It means we need to get better at proving where that value actually shows up.
So how do you know whether AI is actually creating value? One way to think about it is through five levels:
The problem? Plenty of organisations are measuring Level 1 and presenting it with the confidence of Level 4.
“We have 5,000 employees using Copilot” sounds impressive. But so does “we bought 5,000 standing desks.”
Neither tells you whether the business got any better.
The first phase of AI was fun! Big announcements. Shiny demos. Pilots. Agents doing things while someone stood nearby looking suitably impressed.
The next phase might involve rather more spreadsheets. And considerably fewer standing ovations.
Because the conversations that really matter are increasingly about
Not exactly keynote material, but these are the things that determine whether AI becomes a genuinely valuable part of the business or another expensive piece of software everyone was very excited about for six months.
Data is a particularly good example. Dun & Bradstreet's 2026 research found that more than three-quarters of enterprises reported some measurable return from AI, yet only 6% said their enterprise data was fully ready to support AI at scale.
In other words, the problem is increasingly less “can the AI do it?” and more “have we actually got our house in order?”
The businesses that win the next stage of AI might not be the ones buying the most tools, launching the most pilots, or adding “AI-first” to every second slide in the strategy deck.
They might simply be the ones doing the boring stuff properly.
It’s over because AI has already proved enough. We know it can write, analyse, code, search, summarise, automate, and, increasingly, act on our behalf.
The magic trick has been demonstrated but now finance would quite like to see the receipt.
Businesses don't need another two years of people explaining that AI is impressive. They need evidence that all that impressive technology is actually changing something that matters.
The first few years earned AI a seat at the table but that doesn’t mean it had an unlimited tab. And the next phase won’t be defined by which businesses use AI, because let’s be honest here… we all use it!
It’ll be defined by which businesses can answer one very simple question:
What actually changed because of it?
Because “everyone's using it” isn't going to cut it with the CFO for much longer.