A winery owner buys ChatGPT seats for the team in March. By June the licenses are mostly unused and the GM is convinced the staff is the problem. The diagnosis I usually hear is that the team is older, or skeptical, or stuck in their ways, or scared of being replaced.
The team is rarely the bottleneck. The bottleneck is usually the person the team has been asked to follow into the work.
Brennan McDonald wrote a piece called "Your AI Champion Is Doing More Harm Than Good" that names this pattern as the most-documented failure mode in AI rollouts. The framing is plain: when the chosen champion lacks peer credibility, the team's distrust attaches to the messenger before it reaches the message. Every demo, every Slack post, every Friday show-and-tell reinforces the wrong signal — this thing belongs to that person, and that person is not us.
That is the post in one sentence. The rest is what it looks like inside a thirty-person winery, why it happens, and how to recover when you have already picked the wrong person.
The credibility audit nobody runs
Most leaders pick a champion the way they pick a software admin. They look for the staffer who is already excited and already volunteering. That sounds responsible. It is the wrong filter.
The staffer raising her hand for AI usually fits one of two profiles. She is technically inclined and already three tools deep in her personal workflow. Or she is ambitious, knows AI is a career-defining wave, and wants the visibility of being the in-house expert. Both are real reasons to volunteer. Neither is the reason the team will follow her into the work.
Peer credibility is a different asset. It is the property of being the person other staff already go to when something is broken — usually neither the most senior nor the most technical, almost always the one whose advice gets taken in the break room. When that staffer says "I have been using this for my Friday wholesaler emails and it saves me an hour," the team hears a coworker. When the volunteer enthusiast says the same sentence, the team hears a sales pitch.
Hartz AI puts the cohort ratio at one champion per fifteen to twenty employees. For most wineries that means one to three champions total. The ratio is small enough that getting it wrong on the first pick is structural, not cosmetic. A bad first champion does not stall a rollout for a quarter. It poisons the well for the next attempt, because the team has already learned that AI is the thing the boss made Brittany do.
Why the enthusiast pick is so common
The enthusiast pick happens because it solves the leader's problem, not the team's. The leader needs someone to own the rollout. The enthusiast is willing. The match is clean on paper. The leader logs it as a win and moves on to the next agenda item.
The HBR data Keith Ferrazzi cited makes this more pointed. Seventy-eight percent of employees fear AI will take their job. Only twelve percent feel involved in their employer's AI deployment decisions. The team is showing up to a rollout they had no hand in designing, led by a coworker who got the role partly because she liked AI more than they did. That is the credibility gap, sitting in the room before the first prompt gets typed.
The leader sees an enthusiastic champion and a quiet team and reads it as "she is leading, they are catching up." Underneath that, a small private decision has already been made by the team. The decision is that this is her thing.
The decision is reversible. It is not easy to reverse.
The surface-compliance pattern
A team that has decided AI is somebody else's thing does not refuse the rollout. They comply on the surface. The pattern is documented. The forced champion or the unconvinced staffer opens the tool once a week, types a low-stakes prompt so the manager sees activity in the dashboard, closes the tab, and goes back to her actual work.
No habit forms. No second-order use kicks in. The 3x-a-week-for-8-weeks threshold Microsoft documented for habit formation never gets crossed, because the use is performance, not work. The leader looks at the dashboard at day forty-five and sees green check marks. The leader looks at the actual workflows at day sixty and sees the same handwritten visit recaps that existed in March.
I have seen this in adjacent industries the past two years. Salesforce's Agentforce push produced a string of cautionary tales: companies stood up agents, ran the demos, hit the activity metrics, and discovered six months later that the staff was treating the agents the way the cellar treats the calendar reminder system, which is to acknowledge it and then ignore it. The agents worked. The setup around the agents put a chosen champion in front of the rollout who did not have the trust to make the team try.
Surface compliance is the most expensive failure mode because it does not look like failure. It looks like adoption. The dashboards say things are happening. The leader budgets the next phase. The shadow AI use that has been sitting on personal phones the whole time keeps growing, because the staff has already routed around the official rollout. The gap between the 90% of workers using AI on their own and the 40% of firms with an official subscription, which MIT NANDA documented in 2025, is the gap a wrong-champion pick widens, not closes.
The recovery move
If the rollout is three months in and the dashboard looks like compliance rather than work, the move is to re-pick the champion. Not fire the original one — quietly re-pick.
The first step is honest. Sit with the current champion, alone, and tell her the truth. The role is not working the way you hoped, and the reason has nothing to do with her. The rollout needs a different shape, and her time is better spent on her actual job while you reset the pattern. Most people in this position are quietly relieved. Being the AI champion when the team is not following you is exhausting and lonely. The exit is a gift if it is offered cleanly.
The second step is the credibility audit you should have run the first time. Stand in the break room on a Tuesday afternoon and watch. The staffer who keeps getting pulled into the side conversation about the difficult guest or the inventory miscount is your champion. The candidate is almost never the volunteer.
For a tasting room team, this is often the lead host who knows the regulars by name. For a DTC team, the coordinator who handles the awkward subscription cancellation calls. For a cellar, the production assistant the winemaker answers first when she texts. Peer trust shows up in patterns the staff already runs. The leader's job is to see it, not to assign it.
The third step is to transfer the work. Take whatever the original champion built — the starter folder and the routine of three uses a week — and walk the new champion through it for ninety minutes. Hand it over. Then leave her alone with it for three weeks before checking in. The mistake here is to over-manage the handoff, which signals to the team that the original problem was the person and that there is a new chosen one. The handoff should feel low-key. The right framing for the team avoids any reset announcement and instead has the new champion mention, the next time the topic comes up, that she has been using the folder Diego built and it is saving her about ninety minutes a week.
I am writing about this with a fictional demo winery in mind. Not Really Wines is the openly-fictional winery we run as a working showcase at notreallywines.vercel.app, and the scenario maps onto its team cleanly. If Maren had picked Diego, the head winemaker, as her first champion because he is technically inclined, the wholesale team would never follow him into a club retention workflow. The right first champion at NRW would be Riley, the tasting bar lead, who hears the most customer voice and whose advice the rest of the front-of-house team already takes. The right second champion is whoever Riley starts texting on Wednesdays asking how she handled a tricky email — almost never the one who raised her hand.
The handoff move shows up in practice as the most common single intervention I make when a winery owner comes to me three months into a stalled rollout. The diagnostic conversation takes an hour. The handoff takes a week. The dashboards usually stop being green-check theater and start showing real second-order use within thirty days, because the new champion is finally a person the team is willing to copy.
What this means for what you do next week
If you are setting up a rollout from scratch, the lesson is to run the credibility audit before you assign the role. Sit in the break room for a week and watch who staff go to when something breaks. Pick that person. Then resist the urge to dress the role up with a title or a project name. Champions work when they look to the team like a coworker doing a thing, rather than the boss's chosen lieutenant.
If you are partway in and the rollout has the surface-compliance shape — green check marks on the activity dashboard, no observable change in the underlying work — assume the champion pick is the place to look first. Earlier than the tool, earlier than the prompt library. The instinct to blame the team or the tool is almost always wrong on the first try.
The team is rarely refusing AI. They are watching their coworker carry it and deciding whether to follow. The pick is the rollout. Most of the rest of the play follows from getting that one decision right.
For more on the affirmative side of this question — who the right first champion turns out to be for most wineries, and why that pick is almost always the same role — see the companion to this post: why your tasting room manager is the right first AI champion. For the broader set of moves to avoid, the post on five things wine industry leaders should stop doing in AI rollouts covers the moves that often pair with the wrong-champion pick.
A note on the webinar
I am running a webinar that walks through this exact question — how to pick the right first champion and how to recover when the first pick was wrong. The 60-day play we walk through is built around the credibility audit and the eight-week habit window that turns a single staffer's use into a team rhythm.
If your winery already pays for AI licenses and the rollout has stalled around one person, the webinar is built for that exact situation. Check the webinar schedule for the next date and to register.
Part of a 20-post series on employee AI adoption for wineries — see the full series under AI Adoption.