Most winery AI plans I see are calendared by feel. The owner picks a thirty-day pilot window, or a ninety-day refresh cycle, because the planner is laid out in months and quarters and those are the available numbers. None of those numbers come from anything anyone has measured about whether an employee keeps using the tool past the second week. The plans look reasonable on a page and produce roughly the same outcome each time, which is a staffer who tried the tool, found it interesting for a Tuesday afternoon, and quietly went back to her old workflow by the end of the month.
There is a real number for this. Microsoft's Work Trend Index work on AI habit formation found that the threshold where employees stop bouncing off the tool and start compounding sits at three uses per week, sustained for seven to eight weeks. Below that line, the habit does not form, and the staffer reverts. Above it, something else happens — she reaches for the tool without thinking about it, the way she reaches for the email client.
That single threshold is what the 60-day play I run with consulting clients is built backwards from. Every choice in the eight weeks — one champion before a second employee is added, a retro at day sixty rather than day ninety — exists to get one staffer across that threshold and keep her there. If you want to know why a winery rollout that looked busy in week three was dead in week eight, the math usually starts here.
Where the number comes from
Habit-formation research has a long, mostly-disappointing history of round numbers. The twenty-one-day myth and the sixty-six-day study both get cited in productivity books and HR decks without much underlying data about the specific behavior they describe. AI use is different, because it is recent and because the tool vendors have been collecting telemetry on it from day one.
Microsoft's Work Trend Index is the cleanest of those data sets. It is built from anonymized usage signals across Copilot and the Office tools it sits inside, cross-referenced with self-reported survey data from users about how their work changed. The pattern they surfaced is structural. Employees who used the tool fewer than three times a week tended to drop off entirely within the next month, regardless of how positive their early experiences were. Employees who hit three uses a week and held it for seven to eight weeks tended to keep using the tool indefinitely, and at higher volumes than they started.
The way I think about that finding is that AI use sits in the category of habits that compete against existing well-grooved workflows. A staffer who has spent four years writing tasting room visit recaps the way she writes them has a path of least resistance already paved. The new path has to get paved by enough repetition that it becomes the easier route. Three repetitions a week for eight weeks is twenty-four repetitions, supported by data on real workers in real jobs rather than a tidy story about how the brain works.
What "three uses" means in practice
Three uses a week is a low bar that gets misread as a high bar all the time. A use does not need to be a thirty-minute deep-research session. A use is opening the tool on a real piece of work and getting an output back. Drafting a reply to a wholesaler email. Cleaning up a paragraph of the monthly newsletter. Turning a wine club cancellation note into something less robotic. Pulling three takeaways out of a long internal document.
Most of those tasks are under fifteen minutes. Three of them a week is forty-five minutes of AI time. A winery employee can hit the threshold on the equivalent of a long coffee break a week, spread across three sittings.
The reason it gets misread as a high bar is that leaders who have not used AI on real work themselves imagine the use cases as bigger than they are. They picture a project, or a complex workflow with a long custom prompt. What the threshold cares about is frequency rather than complexity. A staffer using AI to clean up three short emails a week is on track. A staffer using AI to build one elaborate quarterly report once a month is not.
The lesson I take from this is that the recurring small task is the unit a rollout should run on. The path gets paved by the Monday morning visit recap getting drafted with AI every Monday, the same prompt or its descendant, over and over until the staffer is taking the new route by default. Showcase projects and leader-impressing demos rarely produce that paving, because they happen too infrequently. The hated-task rule and wine club concierge starter folder posts get more specific about which kinds of tasks pave easiest. Cadence is what makes a recurring small task compound.
Why "less than that, no habit" is the load-bearing claim
The interesting half of the Microsoft finding is the lower bound. Employees who used the tool twice a week did not form the habit at half the rate of the three-a-week employees. They formed the habit at near-zero rate. The curve looks more like a step than a slope.
That maps to something I see in winery rollouts repeatedly. A staffer who uses the tool once on Monday and once on Thursday and skips the third use has missed the threshold entirely, and her next month looks like a slow drift back to the old workflow. The dashboard shows two uses a week and the leader reads it as "almost there." The behavior reads as "she tried it, it did not land."
The same threshold logic explains why a one-off training day does almost nothing. A staffer who sits through a four-hour training and then uses the tool zero times the following week has had one experience that does not compound. Twelve hours of training across a quarter beats four hours on one Tuesday by a large margin, because the twelve hours can be spread to hit the threshold.
I made the structural argument against the all-hands training day in why "all-hands AI training day" kills adoption. The habit-threshold math is the data layer underneath that argument. A single Tuesday in March cannot change how the team works in June because the threshold curve is a step rather than a line.
Why eight weeks, not four or twelve
The duration matters as much as the frequency. Three uses a week for two weeks is six total uses. The staffer has technically used the tool. She has not changed her workflow.
Three uses a week for seven to eight weeks is twenty-one to twenty-four uses. By the end of that window, the staffer has applied the tool to enough variations of her recurring tasks that the pattern has generalized in her head. She is no longer following a script. She is reaching for the tool on a task that resembles other tasks she has used it for, and adjusting on the fly. That generalization is what the habit is.
Twelve weeks is not better than eight. The marginal habit-strengthening effect of weeks nine through twelve is small compared to weeks one through eight. The plateau is real. The reason most plays stop at sixty days rather than ninety is that the additional thirty days produce a small bump in habit depth and a large opportunity cost — the time is more useful spent setting up the second employee.
Four weeks is not enough either. I have run truncated versions of the 60-day play at thirty days when a winery owner insisted on a faster timeline. The thirty-day version produces a staffer who knows what the tool can do, then watches her go quiet by month two. Eight weeks is the duration that converts curiosity into reflex.
Mapping the 60-day play to the threshold
The 60-day play I run with clients lays out four phases. Each is shaped around hitting and holding the threshold, not around what looks impressive on a project plan.
Foundation (days 1-14, weeks 1-2) is where the staffer opens the starter folder for the first time and gets to her first concrete win on a real task. The target is three uses in week one and five in week two. The play is deliberately quiet at this stage. No rituals beyond the leader-uses-it-out-loud move. No metrics beyond frequency. The math reason is that the foundation phase is just getting the staffer to the lower edge of the threshold. Layering on too much structure at this point produces compliance behavior, which counts toward the dashboard and not toward the habit.
Habit (days 15-30, weeks 3-4) is where the threshold is held. Three uses a week, every week, for two weeks. This is the most important phase of the play and the most boring to describe. Nothing flashy happens. The staffer keeps doing the same recurring tasks with the tool. The Friday show-and-tell starts here as a single demo, often just the champion and the leader in the room. What is changing in this phase is that the use is becoming default behavior. The dashboard would look identical between week three and week four. The internal experience is different — the staffer is no longer thinking about whether to open the tool. She is thinking about what to ask it.
Multiplier (days 31-45, weeks 5-6) is where the second employee gets brought in. The reason this happens at week five rather than week one is that the threshold has to be visible before a second employee will copy it. A staffer watching her coworker — usually the tasting room manager the team already trusts — use the tool three times a week for a month, while she still does the task by hand, will eventually ask how it works. A staffer watching her coworker open the tool sporadically in week two will assume the experiment is going to fade. The second employee post has more on why the multiplier phase is where most rollouts live or die.
Operationalize (days 46-60, weeks 7-8) is where the threshold goes from a personal habit to a team rhythm. By the end of week eight, the Friday wins log has entries from multiple people and the prompt library has five to ten attributable entries. The day 60 retro is where you name the next three to four employees who will come into the play in the second sixty days. Habit at the individual level needs structural reinforcement to survive the leader's attention drifting to the next priority. The retro doc is that reinforcement.
Each phase is two weeks for a reason. Two weeks is long enough to establish a beat and short enough that a stalled phase can be diagnosed before it drags. Four phases of two weeks add up to the eight the threshold needs.
What hitting the threshold looks like on a Friday
If you ran this play and it worked, here is what a Friday afternoon at day fifty-six looks like inside the building.
Your champion has used the tool three times this week. So has the second employee, who joined in week five. Both of them have one specific task they used it on that they can describe in one sentence. There is a prompt in the shared library that the champion wrote and the second employee is using almost daily. Two other staffers — neither of whom is officially in the play yet — have asked how to get set up. One of them has been borrowing the second employee's phone to watch her work for fifteen minutes after shift on Wednesdays.
The Slack/Salesforce Workforce Index research that came out in April and May of 2025 surveyed over five thousand workers and found that daily AI users reported a 64% lift in self-reported productivity, a 58% lift in focus, and an 81% lift in job satisfaction. The satisfaction number is the one that matters most at day fifty-six, because it is what predicts whether the habit survives month three. By the time you can read the satisfaction signal in your champion's voice, the play has worked. She is reaching for the tool because the work feels lighter, not because the dashboard says so.
If the play did not work, the Friday at day fifty-six looks different. The champion logged in twice this week. The second employee was supposed to start in week five but the leader has been busy and the handoff slid. No prompts have made it into a shared library because there is no shared library. The dashboard says activity is up. The work has not changed.
I argued the broader version of the metric problem in stop measuring AI usage rate. The threshold math is the reason the surface-level activity dashboard misleads. The dashboard adds up logins. The habit is built from three uses a week sustained for eight, and the dashboard cannot tell the two apart. A staffer who logs in eight times on a single Tuesday and then nothing for ten days adds eight to the count without crossing the threshold. The number you want is per-staffer per-week, sustained for eight.
One sentence to put on the rollout plan
If I were going to put a single sentence on a winery's sixty-day rollout plan to anchor the threshold question, it would be a target rather than a metric. "By day sixty, one named staffer will have used AI three times a week for eight straight weeks, on one named recurring task." That sentence is harder to satisfy than the version that talks about activation percentages or training completion. It is also the version that maps onto whether the rollout is going to be alive in October.
The thing about thresholds is that they reward patience and punish hurry. A leader who tries to hit the threshold with four employees at once will hit it with none. The leader who picks one staffer, protects her time, runs the Friday ritual starting week three, brings the second employee in at week five, and runs the retro at day sixty will get the threshold with one staffer and a path to the next three.
A note on the webinar
The hour we run walks through the full sixty-day play and the math behind the eight-week window. It is built for winery owners and GMs who already pay for AI licenses and are wondering why the team is using them less than expected. If that is where you are, check the webinar schedule for the next date — it is built for that situation. The threshold rule is the one part of the play I would not change if I had to redesign it from scratch tomorrow.
Part of a 20-post series on employee AI adoption for wineries — see the full series under AI Adoption.