42% of Companies Walked Away From Most of Their AI Projects This Year. Last Year It Was 17%.

In plain terms: S&P Global asked more than 1,000 companies how their AI projects were going. The share that had walked away from most of their AI initiatives jumped to 42% this year, up from 17% the year before. The average company scrapped 46% of its trial projects before they ever went live. The demos worked. The trip from demo to something that actually runs is where they died.

Yesterday I wrote about the MIT finding that about 95% of AI pilots show no measurable impact on profit. A fair pushback was that the exact number is fuzzy. So here is a completely different study, from a completely different group, measuring a completely different thing, that points the same direction.

And this one has a twist that should stop you: it is getting worse, not better.

What the research says

S&P Global Market Intelligence ran its enterprise AI survey across more than 1,000 companies in North America and Europe. Two numbers stand out:

• The share of companies that abandoned most of their AI initiatives climbed to 42%. One year earlier it was 17%. That is not a plateau. That is companies more than doubling their odds of quitting in twelve months.
• The average organization scrapped 46% of its AI proof-of-concepts before they reached production. Nearly half of what got started never shipped.

The reasons companies gave were not "the AI cannot do it." They were cost, data privacy, and the plain organizational grind of getting a thing from a demo into daily use.

Why more companies are quitting

Here is the part that connects yesterday and today. A year ago, everyone was in the honeymoon. You spin up a slick demo, it wows the room, everyone claps, and the pilot goes on a slide deck as a win. Nobody had hit the hard part yet.

Now the bill has come due. All those demos had to actually become something people use every day, and that is the part almost nobody budgeted for. So the quit rate doubled. Not because the tools got worse, but because a year in, more companies reached the exact spot where AI projects die and found they had no plan for it.

That spot has a name. It is the road from proof-of-concept to production. The demo proves the tool CAN do the thing. Production means it does the thing, every day, wired into how the work really happens, with someone responsible for it. Forty-six percent of projects never make that trip.

The fix is not a better demo

If projects die on the road to production, then a flashier demo does not save you. You could have the best demo on earth and still be in the 46% that never ships.

What survives that trip is almost always the same shape: one narrow, specific job wired all the way through, instead of a big ambitious AI transformation that tries to boil the ocean. The companies that quit tend to be the ones who aimed huge. The ones that ship aimed small and finished.

A good example is the simplest one in any business: what happens the moment a new lead comes in. That is narrow enough to actually finish. The tool watches for the lead, responds in seconds, and hands a real person a real conversation. No committee, no year-long rollout. It ships because it is small enough to ship.

What this means for you

If you have an AI project that has been "almost ready" for a while, that is not bad luck. That is the road to production, and it is where most projects stall out for good. The question is not whether your idea is good. It is whether it is small enough to finish.

So before you expand the scope, shrink it. Pick the single most valuable moment in your business where a fast response makes or breaks the sale, and wire up just that. Speed-to-Lead is the narrowest, most finishable version of this: catch every new lead in seconds, so the AI project that pays you back is the one that actually ships.

FAQ

Why would the quit rate double in a single year? Because a year ago most companies were still in the demo stage, where everything looks great. This year they hit the hard part, the trip from demo to daily use, and many had no plan for it. The failure did not get worse. More people just reached the point where it shows up.

Is 42% quitting a reason for me to wait on AI? The opposite. The 42% quit because they aimed too big and could not finish. If you aim small and finish one thing, you are not in that group. Waiting just means the competitor who shipped one small system gets a year head start.

How do I know if my project is too big to finish? Ask whether one person could own it, and whether you could point to a single number it moves. If it needs a committee and touches ten departments, it is a transformation, and transformations are what the 42% were quitting. Narrow it until it fits in one person's hands.