Nobody Trusts AI Because the CEO Said So
Why do employees abandon AI tools even after successful training? A customer care team's experience reveals that workplace AI adoption depends less on executive endorsement and more on trust, peer learning and employees' confidence in their own judgement.
What a stalled rollout in one customer care team taught me about how adoption actually spreads
I've been working with an e-commerce company whose customer care team used to start every morning the same way. Someone would spend the best part of every morning copying emails into a spreadsheet by hand. Replies went out in the order emails arrived, so an anxious customer could be left waiting behind a routine shipping question.
We fixed the triage and brought in AI to draft replies, and the team was trained on it. The first draft we reviewed together covered roughly 90% of what the customer had asked. It was a textbook Enhance use case: AI does the groundwork, and a person checks it and adds the human touch.
Weeks later, nothing had changed.
Turns out the team had tried the drafts a few times. One draft left out a line promising to remake the product at no additional cost, and only a reviewer caught it before it went out. A couple of others failed to generate at all. That was all it took. Everyone went back to the way they had worked for years.
If you've rolled out AI and watched usage flatline, this will sound familiar. The usual fix is more training and another reminder email, and it misses what is really going on.
The team wasn't resisting. They were protecting themselves.
Leadership looked at those drafts and saw a tool that was right nine times out of ten. The team saw the tenth email, the one with their name at the bottom, going to a customer who had just made an order.
From where they sat, going back to the old way made complete sense. The old way had never embarrassed them in front of a customer.
This gap in perspective shows up almost everywhere. A 2026 Great Place to Work survey of 4,000 employees found that 83% of executives think their organisation communicates clearly about AI, yet only 37% of frontline workers agree. Leaders tend to assume that once the tool works and the training has happened, adoption will follow. Your team is asking something else entirely: is it safe for me to rely on this?
No slide deck can answer that…. But a colleague can.
What turned it around
One team member had started getting the hang of the drafts and was saving real time, so we asked her to show her colleagues exactly how she worked, on real customer emails: how she checked each draft against what the customer had asked, what she kept and what she changed.
Seeing her work did more than the formal training ever had. Her colleagues could see what they had been doing differently, and they saw it from someone with the same customers, the same pressures and the same name at the bottom of the email. Trust spread from there, and today the whole team is comfortable using AI for this task.
The pattern isn't unique to one company. In BCG research on AI adoption, 69% of employees ranked learning from peers among their top three ways to build AI skills. That sample was software developers, so treat it as a strong signal rather than a universal rule. People rarely trust a tool because their leader endorses it. They trust it because someone like them is already using it and hasn't been caught out.
Three things you can do this week
Find the person who is already making it work. Ask each manager a single question: who on your team is saving real time with AI? Look past the loudest enthusiast. You want the person colleagues already go to when they're stuck, because it's their credibility that transfers.
Swap the demo for a screen-share. Give that person 30 minutes with their peers and their own real work, not a polished example. Ask them to show where they don't trust the AI as well as where they do, because hearing about the misses is what makes the wins believable. If you can, stay out of the room, so people ask the questions they would never ask in front of you.
Let the team decide what the AI touches. Sit down with them and list the tasks that make up their week. Sort them together into Automate (the machine handles it), Enhance (AI drafts, a person decides) and Elevate (the human relationship is the point), then ask them to mark where a human check is non-negotiable. They know exactly where a wrong answer costs you a customer, and asking tells them their judgement still matters.
The best AI training your company will ever run is already sitting at someone's desk.
Comments