Your neighbour has just finished a stunning extension. Bi-fold doors, a kitchen island, the works. Perhaps you walk past every morning. Perhaps you see it over the fence or through a window. Either way, you start to feel your own house looks a little tired.

You could get a surveyor in, check the foundations and draw up proper plans. Or you could grab a hammer and start building, determined to catch up before summer.

Six months later, you’ve spent a fortune. The new room doesn’t quite join up with the rest of the house, the floor slopes, and life is no better than it was.

Right now, a lot of businesses are building AI the second way. I call it keynote envy: the urge to buy whatever a competitor has just shown off on stage.

I understand the pull. Nobody wants to be the business that fell behind. But in my experience, the businesses that rush to copy are rarely the ones that end up ahead.

“We need one of those”

I recently spoke to a business that had just come back from an industry event. One of their competitors had been on stage showing off a slick new AI capability. The room loved it. So did their board.

Within days, the message from the top was clear: “We need one of those, and soon.”

So we asked three simple questions.

1. What problem would it solve for you?
2. Who owns the data it would rely on?
3. Which system holds the truth?

Nobody could answer them with confidence. And when we looked under the bonnet, the core records any AI would depend on were duplicated, inconsistent and spread across systems that didn’t talk to each other. AI sitting on top of that wouldn’t have answered questions badly. It would have answered them confidently and wrongly, at scale.

Here’s something else worth remembering the next time a demo wows the room. What looks production-ready on stage is often a carefully prepared pilot, running on a hand-cleaned dataset in controlled conditions. That makes for a brilliant demo. It’s a long way from business as usual.

That story isn’t unusual. In my experience, it’s the norm.

The numbers behind the hype

If that sounds pessimistic, look at the research.

Gartner predicted that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025. The reasons it gave were poor data quality, inadequate risk controls, escalating costs and unclear business value. Not one of them is about the AI itself.

Gartner has also warned that through 2026, organisations will abandon 60% of AI projects that aren’t supported by AI-ready data.

And MIT’s 2025 GenAI Divide report found that around 95% of enterprise generative AI pilots delivered no measurable impact on profit and loss. The researchers traced the problem to how AI was fitted into the business. The models themselves were largely fine.

Put those together and a pattern emerges. Most AI projects fail because they were built on foundations nobody checked, and the AI itself is rarely the weak link.

There’s a hidden cost too. Every failed pilot makes the next business case harder to win. After two or three projects that overpromised, boards stop believing, even when the next opportunity is exactly the right one.

Your best first AI project is a data project

This is the part nobody puts on stage.

AI doesn’t fix bad data. It scales it.

Duplicate customers, incomplete product records and inconsistent supplier details all get worse once you put AI on top of them. The errors happen faster, they reach more people, and they sound convincing. An analyst might spot that a figure looks wrong. An AI agent will act on it.

So here’s my slightly unfashionable view: the most valuable AI investment many businesses can make this year might not look like AI at all. It’s getting master data clean, governed and trusted, so that when the agent arrives, it has something real to work with.

And unlike most pilots, that work pays for itself before any AI is switched on. Reports get more accurate. Invoices go out right first time. Fewer hours disappear into fixing errors by hand. If the AI project never happened, you’d still be glad you did it.

If you want to know where your own data stands, our data quality guide is a good place to start.

“But won’t we fall behind?”

It’s a fair question, and it deserves a straight answer: no.

Getting your foundations right doesn’t mean pausing everything for a two-year transformation programme. It means knowing where the cracks are before you build on top of them, and that takes a matter of weeks.

The fastest route to AI that actually works looks like this:

1. Start with the outcome. “Our competitor has one” is not a business case. What would change for your customers, your people or your margins?
2. Check what it will stand on. That means data quality, clear ownership, and whether your systems can share information with each other.
3. Test it with real users before you commit real money. As one of only 19 SAP AppHaus Network partners worldwide, this is how we work: we prove an idea against an organisation’s actual systems, data and people before anything is built.
4. Pick the tools last. In an SAP landscape, that might be SAP BTP, process automation, Joule, or simply better master data. Often, the right answer isn’t AI at all.

I’d back the business that spends six weeks on its foundations to overtake the one that spent six months on a pilot it later had to abandon. Every time.

Five questions to answer before your next AI board meeting

1. What problem are we solving, and what is it costing us today? If you can’t put a number on the pain, you won’t be able to put a number on the return.
2. Would we trust our data if a customer could see it? With AI, they effectively will.
3. Who owns the data this depends on? If nobody can say for certain, ownership is the gap to close first.
4. Can our systems talk to each other? An agent that can only see half the picture will give you half-right answers.
5. What will this cost a year after go-live? Licences, integration, support and model monitoring are where pilots quietly turn into budget black holes.

If any of these made you pause, congratulations: you’ve just found your first project.

Build for your own house

The businesses that win with AI won’t be the ones that move first. They’ll be the ones that move on solid ground.

So the next time a competitor’s AI announcement lands in your inbox, or gets a round of applause at a conference, take it as a cue to ask better questions about your own business. Their extension was built for their house.

Your neighbour’s extension may look great. Make sure yours is built to last.

If your board is asking the “what are we doing about AI?” question, I’m always happy to compare notes. Get in touch, or read what we heard from SAP users at the recent SAP NOW AI Tour.