Why AI Adoption Fails: The Four Forces That Actually Decide

By Martin Bergmann | AI Project Lab | August 2026

Back in 2000, I worked on a financial analysis platform for the insurance market. It was a strong product. Our analytics were more advanced than anything else out there, we delivered updates quickly, and our technology was ahead of its time.

But hardly anyone used it.

The reason was simple. Many of our users still had dial-up internet. Those with better connections didn’t want to put financial data online at all. We designed for where we thought users should be, not where they really were.

Is adoption really just a communication problem?

At the time, I thought we’d just misread the market and that better research would have caught the dial-up issue earlier. That’s true, but it’s only a surface-level lesson. It took me years to see the deeper point.

I still hear this thinking in almost every rollout discussion. If what we’re building is truly better, then getting people to use it is just a matter of communication. If we explain the value clearly and train people well, they’ll use it. Building is seen as the hard part, and adoption as the easy follow-up.

After twenty-five years of watching rollouts, I haven’t seen much proof that this is true.

What actually decides whether people adopt

People don’t pick the best option. They pick the safest one.

This isn’t resistance to change. It’s a rational choice made by someone who has more at risk than your plan probably considers.

Around 2012, Bob Moesta and Chris Spiek created a model called the Forces of Progress, building on the Jobs-to-Be-Done approach. They found that four forces act on anyone deciding whether to switch to something new:

• Push: the frustration with how things work today

• Pull: the attraction of the new option

• Anxiety: the fear of the new option

• Habit: the hold of what already works

Push and pull drive the switch. Anxiety and habit resist it. Change only happens when push and pull are stronger than anxiety and habit. How good your product is doesn’t factor into this equation.

Our 2000 platform had a strong push and pull. The analysts really disliked their old spreadsheets, and our tool was clearly better. Still, it lost out to slow internet and a reasonable fear of putting financial data online. We never planned for anxiety or habit. We didn’t even have words for them.

Why AI rollouts fail the same way

Most parts of a typical enterprise AI program focus on the push-and-pull side of that equation.

Productivity stats are push. Demos, champions networks, training, license rollouts, and launch communications are all pull. I’ve sat in the planning for many of these, including ones I led, and the anxiety and habit sections were usually blank.

That might be okay if the anxieties were minor. But they aren’t, and they aren’t just generic worries:

• If I use this and it’s wrong, I own the output. PMI’s AI standard is explicit on that point, and it isn’t a reassuring thing to hear when you’re being asked to try something unfamiliar.

• If I become very good at this, have I demonstrated that my role was automatable?

• If I need it to do the work, does using it reveal that I couldn’t do the work?

None of these concerns are irrational. They’re real, and no amount of excitement in a kickoff presentation will address them.

Habit is even tougher. For a seasoned project manager, habit isn’t just inertia. It’s a method that’s worked for twenty years and has a track record to protect. That’s the same barrier I described in the Einstellung problem, just from a different angle. Einstellung explains why experts can’t see the new tool. The four forces show you what to do once you accept that.

How to plan an AI rollout that accounts for all four forces

Before I approve any adoption plan, I lay out the four forces in columns on a single page and make sure each one is filled in.

Push and pull are easy to fill out. Everyone brings those to the table. Anxiety and habit are almost always blank at first, and the discussion to fill them in is the most valuable part. Two questions help get you there:

What does a person specifically lose if this goes wrong for them?

What does their current method do well that we’re about to interrupt?

You can’t fix anxiety with a slide about benefits. You need a backup plan, a clear statement about who is responsible for AI mistakes, and open permission for people to struggle with it at first. You can’t change habits just by saying the new way is better. You have to fit in with what people already do, or be upfront about asking them to give something up.

Neither of these is just a communication issue. They’re design choices, and they should be part of the plan right alongside the training schedule.

The equation does not care how good your tool is

Push and pull have to outweigh anxiety and habit. That’s the key point.

We’ve spent decades getting good at building the push and pull side, but we rarely design for anxiety and habit. Most rollouts that people later blame on culture actually failed because no one wrote down the real math.

Better doesn’t win. Safer does, and you can design for that.

---

If you want to see where your own AI use actually stands, the AI Adoption Radar maps eight dimensions across integration and accountability in about 90 seconds.

Martin Bergmann is the Director of an IT PMO and founder of the AI Project Lab. He writes about AI, project management, and the future of intelligent work.