# Why AI adoption fails at step three

> Firms start with the most valuable problem. They should start with the most visible one.

- Category: AI in practice
- Published: 2026-09-29
- Author: Asaf Eyzenkot (Suf Zen), founder of Realization
- Page: https://realization.world/insights/why-ai-adoption-fails-at-step-three

## The short answer

Most AI adoption programmes stall at the moment they choose what to do first. They pick the most valuable domain, which is usually also the hardest, and lose the organisation's patience before anything ships. Start instead with the domain whose effect will be visible fastest. Visibility buys permission for the heavier work behind it.

## The five steps

A workable programme has five steps, each ending in a decision the client can act on or stop at: listen before mapping, map every work domain, sequence by visibility, prove it on one problem, then scale and withdraw.

Steps one and two are rarely the problem. Firms are good at describing their pain, and a scored map of domains is straightforward to build. Step three is where the programme is won or lost.

## Score two things, not one

Score every domain on visible impact and on setup effort. Plot them. The quadrant with high impact and low effort is where the programme opens, whether or not it holds the biggest prize.

In one engagement, two of the three items in that quadrant were not software problems at all. One was programme ownership: nobody had hours allocated to AI, so nothing moved. The other was three management decisions: one archive or two, what the firm already paid for and did not use, and who may access what. Both were cheap, and both gated everything else.

## List what already works

A credible map includes the domains that need nothing. In that same firm, visualisation was already about ten times faster thanks to AI. We left it alone and used it as internal proof that the technology worked.

We also recorded the domain the client ruled out, financial insight, on information-security grounds, instead of arguing about it. Saying "you are fine here, do not pay for it" is what makes the rest of the map believable.

## Visibility first is not a compromise

Sequencing by visibility can look like avoiding the hard problem. It is the opposite. The valuable, difficult domain, in that case making design knowledge itself searchable, still sits on the roadmap for months six to twelve. It simply arrives after the firm has seen a measured result in its own work and trusts the method enough to invest in it.

## Write your own exit

The last step, withdrawal, belongs in the first proposal. External support should taper on a published schedule, full for six months, then half, then a quarter, and end at month twelve. It removes the client's most reasonable objection: that a consultant has no incentive to become unnecessary.