Why Most AI Projects Fail (And It Isn't the Technology)
Most AI projects don't die because the tool was bad. They die in the rollout, where the people work gets skipped. Here's the 30/70 split that fixes it.
Most companies that buy AI end up with nothing to show for it.
(And it's almost never the tool's fault.)
It's easy to blame the model, or the vendor, or the budget. But after sixteen years of watching rollouts up close, the pattern I keep seeing at winship labs is the same: the tech is the easy part now. The hard part is everything around it. Who decides what to hand over, who checks the output, and who actually changes how they work on Monday.
So here's the split I keep coming back to, the one we call the 30/70 rule. On any AI project, about 30% of the real work is the technology and the process. The other 70% is people. When a project dies, it's almost always because someone poured everything into the 30 and assumed the 70 would sort itself out.
This isn't just my read. RAND studied why AI projects fail and found that more than 80% of them do, about twice the rate of non-AI IT projects, and the most common root cause is people and miscommunication, not the algorithms (RAND, 2024). S&P Global clocked 42% of companies walking away from most of their AI initiatives in 2025, up from 17% a year earlier. The tech keeps working. The rollout keeps not.
The 30 is the part everyone overspends on
Picking a tool feels like progress. You can see it, demo it, put it on a slide. So that's where the attention goes, and that's the trap.
The tool was never the bottleneck. The bottleneck is the context and the access around it, and that's the unglamorous stuff: the messy real workflow, the one person who has to trust the output before anyone else will, and the ten small calls about what "good" even looks like, none of which show up in a demo but all of which decide whether the thing gets used or quietly abandoned.
A principle I lean on: give AI more of the context around a task, not more tools to do it. Most failed projects have that ratio backwards.
What it looks like when the 70% gets skipped
Picture this. You're a sales manager, and someone hands your team a shiny AI tool for writing follow-ups.
Tuesday morning, a rep opens it, types "write a follow-up email," and pastes whatever comes back without reading it. It's generic, because they fed the model a one-line prompt and hoped. The prospect can smell it. So the rep decides AI "doesn't really work for us" and goes back to doing it by hand.
Nobody showed them which deals to use it on, what a good prompt looks like, or where the real customer context lives. The tool was fine. The 70% never happened.
You bought the treadmill, not the habit
Buying an AI tool and expecting results is like buying a treadmill and expecting to get fit. The treadmill isn't the hard part. The hard part is the habit, the schedule, and the small changes that make you actually use it.
The teams that win with AI aren't the ones with the best tools. They're the ones who did the people work: they picked one real workflow, built a review step a human actually trusts (usually right inside Slack or Notion), and trained two or three people until it stuck. That's what we build as a Digital Associate, one narrow AI use case at a time, and it's the kind of thing a fractional head of AI owns with you instead of handing you a deck and leaving. The boring stuff is what actually sticks.
Where to actually start
Here's the part nobody likes: the 70% is slower and less fun than buying a tool. There's no demo for it, and I can't do it for you. It's your people, your workflow, your call on what "good" looks like. That's the catch, and it's also the whole reason it works once it's done.
So start small and honest. Pick the one task your team already does every week that drains the most time, and put the people work first. Who hands it over, who checks it, who owns it after. That's the whole game.
If you want a quick read on where your team actually stands, the free AI Audit scores it in five minutes, graded on this exact 30/70 split. Happy to talk through whatever it surfaces.
No rush on any of it. When you're ready, start with the work your team already hates doing by hand. Get that off their plate first. That's the win that actually feels like one.
