Is AI just the latest bandwagon?

First posted on LinkedIn, September 2024.

Is AI just the latest technology bandwagon for organisations to jump on?

I don’t think so, and I’d be a hypocrite if I did. AI and automation can genuinely enhance a service offering and take real cost and drudgery out of a business. But plenty of organisations are rushing in, sometimes with the “help” of vendors who are less than scrupulous about it, without fully grasping either the potential benefits or the shortcomings.

Without clear goals and a proper understanding of the outcomes you’re chasing, AI and automation projects risk failure and wasted money from the very start. Everybody involved needs to be on the same page before anyone starts building.

The vendor dynamic

Most vendors are perfectly decent. The failure mode isn’t villainy, it’s a solution arriving in search of a problem, plus an unusually persuasive demo. AI demos are the most flattering demos ever built: clean data, a happy path, a curated question, and an audience who don’t yet know which of their own edge cases will break it.

The tell is the direction of the conversation. If a supplier opens by asking what you’re trying to achieve, what your data looks like and who currently does this work, that’s a good sign. If they open with capability (and the business case gets reverse-engineered afterwards to justify a tool somebody has already fallen in love with) you’re on the bandwagon, whatever the slide deck calls it.

Thirty years of the same joke

There’s a Dilbert strip doing the rounds where the boss announces the company should build an AI, Dilbert wonders whether he understands what he just said or simply saw it in a trade magazine advert, and then asks what colour the AI should be. The boss reckons mauve has the most intelligence.

Except that isn’t the strip. The original ran on 17 November 1995 and the boss wanted an SQL database; the punchline was that “mauve has the most RAM”. Somebody has swapped two words in a thirty-year-old cartoon and it lands just as cleanly today.

Which is the whole argument, really. The technology in the boardroom changes every few years (databases, data warehouses, offshoring, SOA, big data, blockchain, digital transformation, now AI) and the failure pattern doesn’t change at all: a decision taken on the strength of a name, and the details worked out afterwards by whoever has to build it. The trade magazine ad has become a LinkedIn post, but the mechanism is identical.

The evidence is fairly brutal

RAND’s 2024 study of why AI projects fail puts the failure rate above 80%, roughly twice that of comparable IT projects that don’t involve AI. Gartner expects a substantial share of generative AI projects to be abandoned after the proof of concept.

Look past the number to the cause, though. RAND’s leading root cause is that the organisation misunderstood or miscommunicated the problem to be solved. Not model quality. Not compute. Not talent. The projects fall over because nobody nailed down what was actually being fixed: the one thing that costs nothing at the start and a fortune to discover at the end.

What “clear goals” means in practice

It’s a phrase that gets nodded at in every kick-off and defined in almost none. In practice I’d want honest answers to these before spending anything:

• What number moves, and by how much? “Improve efficiency” is not a goal. “Cut the average time to first response from four hours to one” is a goal, and it tells you immediately whether you succeeded.
• Whose job changes? Every automation lands on somebody’s desk. If that person hasn’t been in the room, you haven’t got a project, you’ve got an announcement coming.
• Is our data anywhere near ready? Most AI disappointment is really data disappointment wearing a better suit.
• What happens if it works? Who owns it, who retrains it, who checks its output, and what’s the plan when the pilot ends and it needs to run for real? Pilots that succeed and then quietly die are more common than pilots that fail.
• What would make us stop? Agreeing the abandon criteria while everyone is still calm is one of the cheapest things you will ever do.

I’m thoroughly pro-AI, for the record; I use these tools daily and build with them. But the organisations that get value out of this are the ones treating it as a business change that happens to use AI, rather than an AI project that happens to involve the business. The technology is the easy part. It usually is.

Update, July 2026

Two years on, the numbers have got sharper and they point the same way. MIT’s Project NANDA study in 2025 found that around 95% of generative AI pilots were delivering no measurable return to the P&L, with only about 5% capturing value at scale. S&P Global’s 2025 survey found 42% of companies had abandoned most of their AI initiatives, up from 17% the year before. Gartner has pinned a large share of abandonment specifically on organisations not having data that was ready for the job.

So: not a bandwagon, but an awful lot of people fell off one. The gap isn’t between companies using AI and companies not using it; it’s between the few who operationalised it and the many whose pilots never became anything. Which is the same lesson I keep arriving at from different directions: the hard part was never the technology.