The Slack channel for the new AI tool has been live for six weeks. The consultant who set it up said adoption was on track. The delegate gave a positive update in the monthly review. And yet, when you look at what’s actually happening on the floor, nobody seems to be using the thing for anything that matters. The questions have stopped. That could mean things are running smoothly. It could also mean the team has found their way around the problem.
This is the gap the delegate’s update doesn’t close. Not because they’re concealing things, but because filtered progress reports are how that role works. The team has no such filter. They show you what they actually think through how they behave, and if you know what to look at, that behaviour is more informative than any rollout slide.
What does the team’s body language on an AI rollout actually look like?
Passive resistance to an AI rollout rarely arrives as a formal objection. It shows up as the workaround spreadsheet, vague answers when you ask who’s using what, the way people go quiet when AI comes up in all-hands. HR Executive research found that employees uncertain about a tool’s implications tend to engage with it in ways that confirm it inferior, feeding poor data in, using it on low-stakes tasks, or routing around it entirely [11].
The behaviour is predictable. People who were not convinced by the case for the tool, or who were not brought meaningfully into the rollout decision, will find the path of least resistance through the process. The signals are consistent: low-quality inputs to the system, workarounds that sit alongside the official process rather than replacing the old one, a gradual narrowing of who actually uses the tool day to day. All of it is visible on the floor, and almost none of it gets into the delegate’s monthly report.
Why does this matter if you have a delegate running it?
Your delegate is invested in showing the rollout is working. That is how the role is structured. What it means is that the summary reaching you has been filtered through someone whose credibility is tied to forward progress. BCG research on AI adoption found that senior leaders report substantially higher personal AI use than the teams below them, a gap that widens as you move further down the hierarchy [12].
At board level, that gap produces optimism about adoption. For a founder, it should produce questions. The distance between what leadership uses and what the floor does is one of the clearest leading indicators of a rollout in trouble, and it is a distance your delegate may not be able to see clearly from inside the project. The team’s behaviour is the unprocessed version of the story your delegate is telling, and they have no incentive to manage what you observe.
When a team is genuinely adopting a tool, you see it in small things: the way problems get reframed, the questions they ask about the system’s limitations, the organic sharing of what works. When adoption is shallow, the absence of those things is its own signal.
Where do these signals show up in practice?
Three patterns tend to surface the adoption gap most clearly. Shadow workflows are the first, where parallel processes appear alongside the official tool rather than replacing it. Adoption distribution is the second, showing how far down the team genuine use extends. Data quality inside the tools is the third, which degrades visibly when people use the system to satisfy a process requirement rather than to get work done.
The workaround spreadsheet is the clearest of the three. When someone builds a parallel process to do what the new tool was supposed to do, they are telling you something specific about whether they trust it, whether they were trained well enough to rely on it, and whether the rollout framing gave them reason to commit. PwC survey data on technology adoption shows a consistent pattern of selective use: people adopt tools where they see clear personal advantage and route around them elsewhere [27].
Adoption distribution tells the harder story. A rollout showing strong engagement at management level and thin engagement two rungs down is showing you exactly how persuasive the case was when it reached the floor.
Korn Ferry’s research on AI readiness found that leaders who frame rollouts around efficiency metrics (headcount reduction, speed gains) see lower adoption rates than those who frame it around capability, giving people better tools and freeing them for higher-value work [43]. The framing set at the start shapes how the team reads its own relationship to the investment.
When should the founder step in rather than let the delegate manage it?
Friction in the first 60 days of a rollout is normal and often a good sign. A team asking hard questions, challenging how the tool is being used, or pushing back on the framing is genuinely in the conversation. What you are watching for is the quality of the friction, or its absence. A team that stopped asking questions after week two may have settled in. It may also have given up.
The distinction worth holding is between active friction and passive withdrawal. Active friction means people are still engaged with the process, even if they are unhappy about parts of it. Passive withdrawal, seen in declining optional rollout sessions, low task completion in the new system, and minimum-effort submissions, means the conversation ended without anyone formally calling it.
Spencer Stuart’s research on CEO-level AI engagement found that visible personal use of AI tools by the person at the top is among the strongest predictors of organisation-wide adoption [2]. The floor follows the founder’s lead. A founder who handed the mandate to a delegate and visibly stepped back from it signals, without intending to, that the business does not think the investment warrants the principal’s own attention. The team reads that signal accurately.
What’s actually driving what you’re seeing?
Two things tend to underlie passive resistance and the adoption gap in AI rollouts, and they usually run together. The first is how the rollout was framed at the start, whether AI was introduced as something that would give people better tools, or as something that would reduce the need for people. The second is what the founder’s own visible behaviour communicates about how seriously the business takes the investment.
OECD research on AI adoption in owner-managed businesses found that trust (in the tool, in management’s intent about jobs, and in the business’s genuine commitment to making it work) is a consistent adoption barrier when not explicitly addressed from the outset [5]. The delegate’s townhall slide does not build that trust. The team builds it through what they observe over time: whether early adopters seem to benefit, whether the tool appears in decisions that actually matter, and whether the person at the top seems to think any of it is worth their own attention.
Modelling is the highest-return action available here. Not a product demonstration in the all-hands, but actual visible use in meetings, in how decisions get made, in ways the team can observe day to day. McKinsey’s research on workplace AI adoption found that visible leadership engagement with tools is one of the strongest adoption accelerators available, ahead of formal training programmes and incentive structures [9]. If you want the floor to read the rollout as real, your behaviour needs to tell the same story the delegate’s slides are telling.
The team is already showing you what they think. The question is whether you are close enough to the floor to see it.
