Back to blogs

// What's on our mind

7 signs your AI rollout has a change management problem

We list 7 signs in the daily work that show an AI rollout has a change management problem and the first steps leaders can take to address each one.

7 signs your AI rollout has a change management problem

Resistance to AI shows up in the work long before it shows up in a survey.

Most AI rollouts at banks and insurers report strong launch numbers, then plateau. Only 23% of U.S. employees used AI at work a few times a week or more in late 2025, which means licences and training rarely translate into daily habits. Employee resistance to change rarely announces itself in a town hall. It hides in spreadsheets, side channels and quiet reversions to the old process.

The useful question is where to look. Each sign below is something you can observe in the work itself, and each points to a barrier to AI adoption that better software won't fix. Spot two or three, and you're looking at a change management problem.

7 signs adoption is stalling inside your AI rollout

1. Shadow spreadsheets keep running beside the new tool

When people keep a private spreadsheet next to the AI tool, they don't trust its output yet. A reconciliation analyst might run the AI-assisted match, then rebuild the same match by hand in a personal workbook before signing off. That double work feels safe to the analyst, and it quietly doubles the cost of the process.

Shadow files are useful evidence. They show exactly where confidence breaks down, often on edge cases the tool handles poorly or that nobody has tested with the team. Ask to see them instead of banning them. Each one marks a spot where a review standard, a worked example or a fix to the tool would let people retire the backup copy for good.

2. Workarounds spread faster than the approved workflow

A workaround that travels across teams is a stronger signal than one person's habit. If relationship managers start sharing a prompt template on a group chat that skips a required approval step, the approved workflow is too slow or too rigid for the job they're doing.

Workarounds carry risk because they move outside governance, so the instinct is to shut them down. Shutting them down without understanding them usually pushes the behaviour further underground. Treat the workaround as a design input. Find out which step it skips and why, then fold the useful part into the approved path with the controls your risk team needs.

‍

"Shadow files are useful evidence. They show exactly where confidence breaks down, often on edge cases the tool handles poorly or that nobody has tested with the team."

‍

3. Training attendance stays high while practice stays rare

Full training rooms can hide an adoption gap. In early 2025, 69% of U.S. workers said they don't use AI chatbots at work, even as employers expand courses and licences. Attendance measures exposure, and exposure isn't skill.

The fix is practice on live work. A claims team that spends one hour a week running real files through the tool, then comparing results with a peer, builds more ability than a full-day course ever will. Track practice hours alongside completions so the gap stays visible.

‍

Workers not using AI chatbotsIn early 2025, 69% of U.S. workers said they don't use AI chatbots at work, so full training rooms can hide an adoption gap.69%of U.S. workers said theydon't use AI chatbots at workIn early 2025, 69% of U.S. workers said they don't use AI chatbots at work, even as courses expand.

‍

4. Manual override rates keep climbing after launch

A rising override rate means people are rejecting AI output more often over time. Early overrides are healthy, since people are testing the tool. A rate that keeps climbing months after launch tells you trust is eroding, or that the tool has drifted from how the team actually works.

Look at which overrides repeat. If underwriters keep rewriting the same section of AI-drafted risk notes, the tool may be missing context they rely on, such as regional loss history. That's fixable with better inputs, clearer prompts or a narrower scope, and each fix should be shared back so the team sees its feedback acted on.

5. Managers still review work the old way

Teams follow the review standard over the launch memo. If a manager still expects a fully handwritten credit memo, nobody on that team will risk submitting an AI-assisted draft. The old standard wins by default, whatever leadership announced.

Managers need explicit permission and guidance to reset what good work looks like. That means agreeing which parts of a task AI can draft, which parts need a human check and how reviewers should judge the result. Once a manager accepts an AI-assisted draft in public, the rest of the team follows quickly.

6. Usage concentrates in a handful of enthusiasts

Healthy adoption spreads across a team. If usage data shows the same five people driving most activity, the tool has found its fans and missed everyone else. Enthusiasts are valuable, yet they often work around problems that would stop a typical colleague cold.

Pair each enthusiast with a reluctant colleague on a shared, recurring task for a few weeks. The enthusiast learns where the tool confuses people, and the colleague sees it handle work they recognize. That pairing spreads habits faster than another round of announcements, and it surfaces the friction points worth fixing first.

7. Nobody can say how much work runs through the new path

If leaders can't say what share of the target work now runs through the AI-assisted path, adoption is being assumed instead of measured. Logins and licence counts describe access. They don't tell you if the onboarding files, chargeback responses or research notes actually moved.

At Electric Mind, we treat that share as the signal leaders watch. Pick one workflow, define what counts as running through the new path and report the percentage monthly. A team at 15% with a rising trend is in better shape than one at 40% that has stalled.

The table below pulls the 7 signs into one view you can use as a quick diagnostic.

‍

Sign in the workWhat it tells you
1. Shadow spreadsheets keep running beside the new toolPeople still don't trust the output, so they rebuild it by hand before signing off.
2. Workarounds spread faster than the approved workflowThe approved path is too slow or rigid, and teams are designing their own version.
3. Training attendance stays high while practice stays rareCourses build exposure, and skill only grows through regular practice on live work.
4. Manual override rates keep climbing after launchTrust is eroding, or the tool is missing context the team depends on.
5. Managers still review work the old wayThe old review standard quietly overrules whatever leadership announced.
6. Usage concentrates in a handful of enthusiastsThe tool has found its fans and has yet to reach the typical colleague.
7. Nobody can say how much work runs through the new pathAdoption is being assumed when it should be measured as a share of real work.

‍

Adoption trend over levelTrack the share of work on the new path over time, because a team at 15% and rising is healthier than a team stalled at 40%.Share of work on the new pathTime40%Stalled15%RisingA team at 15% with a rising trend is in better shape than one at 40% that has stalled.

‍

What to do once the signs show up

Seeing two or more of these signs means adoption needs active management before any further tool rollout. Start small and close to the work, because the signs above all trace back to habits, standards and trust that only shift with practice and visible follow-through.

A practical starting sequence covers five moves.

  • Pick one workflow and measure the share of it running through the new path
  • Collect shadow files and workarounds as input for the next fix
  • Reset the review standard with each manager on that workflow
  • Schedule weekly practice on live work for the whole team
  • Report progress monthly and share which feedback changed the tool

‍

"A team at 15% with a rising trend is in better shape than one at 40% that has stalled."

‍

Leading change through AI is mostly patient, unglamorous follow-through. Our teams at Electric Mind see the same pattern each time, where the organizations that keep measuring and adjusting are the ones whose adoption holds a year later.

Got a complex challenge? Let's solve it together, and for real.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
Top