Skip to content

When It Stalls, and Keeping It Alive

Many adoption programs stall. Usage climbs at launch, plateaus, and then slips. This is common, not a verdict on your work. The difference between a program that recovers and one that dies is whether someone diagnoses the stall instead of guessing at fixes.

When to use this

Use this when the numbers have gone flat or started to fall. Use it when a rollout that started well has cooled, or when you inherit a program that is already stuck.

Diagnosing a stall

Do not reach for a fix before you name the cause. Applying the wrong remedy is how a stall becomes a death. Ask, in order:

  • Is the tool working and available, or is this an IT problem in an adoption costume?
  • Did people ever get past first use, or did they never really start?
  • Is it a skill gap, a trust gap, or a workflow that AI does not actually fit?
  • Are managers modelling the change, or quietly ignoring it?

The answer points to a different move. A trust gap and a tooling gap need opposite responses.

Common failure modes

Most stalls are one of a few shapes:

  • Shelfware. Licenses bought, never adopted. Access was mistaken for adoption.
  • Adoption zombies. People log in and go through the motions, with no change in the work.
  • Pilot purgatory. A promising pilot that never gets the support to reach production.
  • Champion collapse. The network was launched and then starved, and it decayed.

Recovery moves

Match the move to the diagnosis:

  • Tooling gap — fix access and reliability first. Nothing else works until the tool does.
  • Trust gap — shrink the scope to a workflow people can verify, and rebuild confidence there.
  • Skill gap — return to enablement, anchored in the real task, with reinforcement.
  • Manager gap — go back to the accountability map. Adoption does not survive managers who opt out.

Recover one workflow visibly before trying to relaunch everything. A single restored win rebuilds credibility for the rest.

Converting shadow AI

When people use personal AI tools outside sanctioned channels, that is not only a risk. It is demand, already proven. Banning it without a sanctioned path drives it further underground.

  • Find where shadow use is happening. It marks the workflows with real pull.
  • Offer a sanctioned tool at least as good, and make the safe path the easy path.
  • Recruit the heaviest shadow users as champions. They have already done your discovery for you.

Reaching steady state

A program that only knows how to launch has no answer for year two. Adoption is not a project that ends. Move from push to pull:

  • Fold AI use into onboarding, so new joiners start fluent.
  • Hand ownership of embedded workflows to the teams that run them.
  • Keep a light central function for standards, new tools, and the next wave.

The goal is a state where adoption continues without a campaign behind it.

Watch out for

  • Fixing before diagnosing. The fastest way to waste a recovery is to guess.
  • Relaunching everything. A broad relaunch of a stalled program repeats the original failure at scale.
  • Punishing shadow AI. Suppression without a sanctioned path destroys your best signal of demand.
  • Declaring victory. A program left unattended at steady state slides back to where it started.