MIT NANDA's research arm published a number in August 2025 that should be on the wall of every winery owner who has bought a ChatGPT Plus subscription. Of enterprise generative AI pilots that have run since 2023, ninety-five percent produced no measurable impact on revenue, cost, or P&L. The MIT phrasing is precise: no measurable impact, period.
The pitch decks do not mention this number. The LinkedIn posts where someone claims their team transformed with AI in six weeks do not mention it either. The number describes how often the AI work that wineries are paying for fails to produce anything the GM can point to at the end of the quarter.
The harder thing to look at is what the 5% that did work share. It has nothing to do with the tool.
Where the 5% live
The MIT report calls it the GenAI Divide. The 5% that produced measurable impact had something else in common: they set up the work around the tool so the tool could do its job. Budget and model choice were second-order. Two patterns dominated.
The first: bought from vendors or built with partners, instead of trying to build internally. Bought-or-partnered pilots succeeded at 67%. Internal builds succeeded at 22%. Internal builds were not failing because the engineers were worse. They were failing because internal builds keep getting reassigned, rescoped, and rebuilt while the work is still warm. A 30-person winery does not have a platform team that outlasts that churn. The vendor relationship is the discipline.
The second: systems that hold onto what the employee did yesterday and adapt to it. The pilots that survived past month two were the ones where the model had access to the team's writing, customers, and workflow. The 95% were pilots where the employee opened a blank chat window every morning and started from scratch.
A blank chat window with no memory of yesterday is why some staff log in once, feel like nothing carries over, and quietly stop logging in. The tool worked. The setup around the tool was incomplete.
Why the tool isn't where the gap is
BCG published a related number that I think about more often than the 95. They studied AI implementations across hundreds of firms and found that the top performers split their effort 10/20/70: 10% on the algorithm, 20% on technology and data, and 70% on people, process, and workflow redesign.
Most winery owners I talk with have inverted that ratio. They paid for licenses and expected the licenses to do the work. The 70% — the part where the leader sits down with one employee and rebuilds how a recurring task gets done — gets assigned to "we'll figure it out," and then nothing happens for six months and the tool gets the blame.
The 70% is the part nobody sells you. Nobody runs an ad for "we'll sit with your tasting room manager for two hours and write down what she does every morning, then build her a folder that has all of it in one place." That is the work. That is also the thing that turns a winery's AI subscription from a budget line item into a Monday morning that runs four times faster than it did the month before.
What this looks like at a thirty-person winery
A 30-person winery is not a Fortune 500 case study. It does not have a center of excellence, a head of AI, or a quarterly OKR cycle. It has a GM, a winemaker, a DTC director, a tasting room lead, a club concierge, a couple of cellar hands, and a marketing person who is also running the wholesaler ops. That is the surface. The 95% number tells you something specific about how AI fails at that surface.
What fails: licenses for everyone, a Slack channel called #ai-prompts that nobody posts in, a Notion page with thirty starter prompts that nobody clicks, an all-hands training day in February that everyone forgets by April, a quarterly "AI strategy" review that produces a Google Doc with no owner.
What works: one person, one task she does every week, one folder built with her for ninety minutes, three uses a week for eight weeks. After that, a second person. Then, eventually, a third. The shape of the play is the inverse of how every SaaS vendor wants to roll out their seat license.
McKinsey's 2025 State of AI study added a number that sounds like a constraint but is closer to a permission slip for smaller wineries. Only 29% of sub-$100M-revenue firms have reached the scaling stage of AI use, against about 50% of $5B-plus firms. The implication some leaders draw from this is "we're behind." The implication I draw from it is that the firms quietly running an actual play are still a small minority. A 30-person winery is not losing a race; most wineries have just not been shown the method yet.
Why "the team won't use it" isn't about the team
There is one more piece of the picture worth pulling forward, because winery owners often hand me the wrong half of it first. The MIT report also noted that 90% of workers report using personal AI tools daily for personal tasks, while only 40% of firms have an official subscription in place for staff to use. The teams that look like they are refusing to touch AI are already using it — on their phones, between customer visits — they just are not using it on the work the winery pays for. I write more on that gap in a later post in this series.
The official rollout looks like a failure because the leader is measuring the wrong tab. The shadow usage has been there for months. The setup question is how to move the AI use that already exists on the team onto the work the winery pays for.
That is a different problem than the one most leaders walk into the rollout with. It is a much smaller problem. It is also the gap the 5% closed.
A small wine business has an advantage here over a Fortune 500 firm with a center of excellence. The leader is in the room with the staff every week. There is no distance to bridge, no internal politics filtering the truth, no AI committee scheduling a review of the survey of the audit of the rollout. The leader can walk over to the tasting room manager on Tuesday afternoon, pull up a chair, and ask what she is already using AI for in her personal life. The answer is almost never zero. The work after that conversation is to take what the staffer is already comfortable doing and pull it inside the part of her week that the winery is paying for. No new rollout required.
The cheapest first move
A winery owner reading this who wants to do one thing differently this month should stop running the all-hands rollout and pick one person.
Pick the person other people already go to when something is broken. That is rarely the most technical staffer on the team, and rarely the most senior. It is the tasting room manager who knows everyone's birthday, or the DTC coordinator the wholesaler reps email back. Whoever it is, the asset to compound is peer credibility.
Find the recurring task that person does every week and does not love doing. Sit with her for ninety minutes and build her a folder containing the task description, three or four samples of how the work usually reads, and the rough shape of the output you want. Hand her the folder. Ask her to use it three times a week for the next eight weeks. Do not ask her to teach anyone. The launch meeting and the Slack announcement can wait.
The reason the eight-week window matters is documented. Microsoft's Work Trend Index research has a habit-formation threshold of three uses a week for seven to eight weeks. Below that, employees bounce off. Above it, the use turns into a habit and the second-order effect kicks in: the staffer starts asking the tool things she would never have sat down to write out before. That second-order use is where the time savings live. The first eight weeks are where the habit forms. The value follows the habit.
That is the play. The 95% number is what happens when a winery skips the eight weeks.
A note on the webinar
I am running a webinar that goes deep on this exact pattern. The hour is a live demo of building a Monday-morning starter folder for one frontline winery role in real time, ending with something a GM could hand off the next day. The second half walks through the research-backed 60-day play that holds the rollout together past month two, the phase where most AI initiatives quietly fold.
If your winery already pays for AI licenses and those licenses are not getting used, 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.