AI adoption fails on three psychological forces, identity threat ('this replaces what makes me valuable'), asymmetric trust (one bad output erases ten good ones), and workflow friction (the new way costs effort before it saves any). Change management for AI means addressing all three explicitly, with the same budget seriousness as the technology.
The three forces that stall rollouts
- Identity threat: people resist tools aimed at the work they're proud of. Aim AI at the work they complain about, drudgery first, craft later, if ever.
- Trust asymmetry: users forgive human errors but not machine ones. One hallucinated answer in week one can end adoption; expectations must be set before the first error, not after.
- Friction asymmetry: the old way is fast because it's practiced. The new way must be genuinely easier within days, or the pilot enthusiasm decays into quiet reversion.
A change plan that respects the psychology
- 1Pre-frame errors: 'it will be wrong sometimes; here's how you catch it' said before launch buys tolerance no apology can buy after.
- 2Start with volunteers plus one loud skeptic, the skeptic's conversion (or their feedback) is worth more than five fans.
- 3Make the first week effortless: templates, side-by-side sessions, someone to ask. Adoption dies alone at a keyboard.
- 4Show the loop working: when users report a bad output and see it fixed, trust compounds. Silence after feedback kills it.
- 5Let teams keep a visible win: if AI saves four hours, let the team feel some of that time, adoption that only feeds a metrics slide breeds resentment.
Reading the adoption signals
- Usage after week three, not week one; novelty inflates every launch curve.
- Voluntary usage vs mandated usage, only one predicts durability.
- Who asks for more access, spreading pull beats pushed rollout.
- Quiet reversion: watch for the old workflow reappearing; it's feedback, not sabotage.
