The AI That Matters Most Is the BORING Kind

Walk into any boardroom in 2026 and someone will be talking about AI. The conversation almost always lands in the same place: the customer-facing showpiece. The chatbot that will delight shoppers, the AI concierge, the marketing engine that writes itself. Big, visible, exciting. The kind of thing you can demo to the board and put in a press release.

And most of it will fail. Not because the technology is bad, but because it’s pointed at the wrong part of the business.

The numbers on this are now hard to ignore. MIT’s much-cited 2025 study of enterprise AI found that 95% of generative AI pilots delivered no measurable impact on the bottom line. Ninety-five. Billions spent, and the overwhelming majority of it produced nothing you could point to on a P&L. The instinct is to read that as proof the hype was empty. It isn’t. It’s proof the effort went to the wrong place.

Because buried in the same research is the part nobody puts on a slide: the biggest returns weren’t in the shiny front-office tools at all. They were in back-office automation. The unglamorous middle of the business, where invoices get matched, transactions get reconciled, orders get processed and reports get built. More than half of AI budgets were going into sales and marketing, and that’s precisely where the returns were thinnest. The money and the value were pointing in opposite directions.

The dull work is where the money is

There’s a reason the back office gets overlooked. It’s boring. Nobody launches a startup to automate bank reconciliation. Nobody’s going to feature your improved invoice-matching in a keynote. But that dull, repetitive, high-volume work is exactly what machines are good at, and exactly where the hours and the errors and the cost pile up unnoticed.

Look at what’s already working and it’s strikingly mundane. Accounting platforms now use AI to auto-match the overwhelming majority of bank transactions, work a person used to do line by line. They send a nudge before a VAT deadline, catch the error that would have triggered a penalty, flag the number that doesn’t look right. One of these tools reckons it saves a small finance team around five hours a week. Five hours isn’t a headline. Multiply it across a year and a team, though, and it’s a person’s worth of time handed back, spent on nothing anyone will ever tweet about.

“The AI that pays for itself is almost never the AI anyone gets excited about,” according to Tecvia, a UK Microsoft Dynamics 365 Business Central implementation partner. “It’s the invoice that matches itself, the reconciliation that just happens, the report that builds overnight. It’s invisible, which is exactly why it works. But it only works if the data underneath it is clean and the systems genuinely connect. Point clever AI at a mess and all you get is a faster mess.”

This is the paradox of useful AI. The more boring the task, the better the case for automating it, and the less anyone wants to talk about it.

Why the exciting pilots keep dying

That last point is the one most businesses learn the expensive way. The reason so many AI projects stall isn’t the model. It’s everything underneath it.

Aditya Challapally, who led the MIT research, was clear about what separates the handful of companies getting real value from the rest. The winners, he said, are the ones who “pick one pain point, execute well, and partner smartly” rather than trying to boil the ocean with a grand AI strategy. The failures tend to share a pattern: a clever tool bolted onto fragmented systems and messy data, expected to work magic on a foundation that was never built for it.

This is where the hype does real damage. It sells AI as something you sprinkle on top, a layer of intelligence over whatever you already have. But AI doesn’t rise above bad data, it inherits it. If your stock figures are wrong, your AI forecast is wrong. If your customer records live in four systems that don’t talk to each other, no model can give you a single reliable answer, because there isn’t one to give. The research bears this out: the bulk of the work in any successful AI project isn’t the AI at all, it’s the unglamorous groundwork of cleaning data and connecting systems so the tool has something solid to stand on.

Which is why the businesses that win with AI often look, from the outside, like they’re doing something boring. They’re not chasing the showpiece. They’re getting their data in order, connecting the systems that run the business, and then pointing AI at the specific, repetitive, well-defined tasks where it can deliver. The excitement comes later, if at all. The results come first.

Start where it’s dull

So the practical advice for any business wondering where to begin with AI is almost the opposite of the conference-stage version. Don’t start with the customer-facing moonshot. Start with the dullest, most repetitive, highest-volume task in your back office, the one everyone complains about and nobody wants to do. That’s where the clearest return is hiding.

Then be honest about the foundation. Before you spend on any AI tool, ask whether the data it will run on is clean and whether the systems it needs to read from truly connect. If the answer is no, that’s the project, not the AI. Fix the plumbing first and the clever tools have something to work with. Skip it and you’ll join the 95%.

The companies that will look back on this era as the one where AI changed their business won’t be the ones with the most impressive demo. They’ll be the ones who did the boring work: got the data right, connected the systems, and let the machine take over the tasks that were never worth a human’s time in the first place. It won’t make for a good press release. It’ll just make the business work better, which was always the point.