A tool is something you own. A habit is something you have. Your renewal quote prices the first and says nothing about the second. That gap is the whole problem with how most companies have approached AI enablement so far: the money went to access, the value depends on repetition, and nobody wrote a line item for repetition.
The good news is that the access problem is genuinely solved. BCG’s AI at Work: Why Strategy Matters More Than Tools (June 3, 2026), based on close to 12,000 frontline employees, managers and leaders across more than a dozen markets, found 74% of frontline employees now use AI every day or a few times a week — up 23 points in a year. The bad news is in the same report: only 36% of respondents feel they have received adequate upskilling, and 66% of frontline employees get limited or no guidance on what to do with the time AI saves them.
A Tool You Own Versus a Habit You Have
The two get used interchangeably in board updates. They differ on every dimension a buyer controls.
| Buying a tool | Building a habit | |
|---|---|---|
| Unit of measure | Seats, logins, assigned courses | Frequency on a named task |
| Who owns it | IT and procurement | The employee and their manager |
| What it costs | A predictable annual number | Attention from people with day jobs |
| Time to show up | Weeks | Months |
| Failure mode | Paid seats nobody opens | Use on the easy tasks only |
| Success signal | Deployed on schedule | The absence of the tool is felt |
That last row is the test worth remembering. A habit exists when removing the tool would be noticed within a day. Until then you have a subscription and a hope.
The behavior, not the deployment, is what correlates with results. McKinsey’s State of AI in 2025 found the respondents attributing at least 5% of EBIT to AI — roughly 6% of the sample — were nearly three times as likely as other respondents to have fundamentally redesigned their workflows, and that redesign had one of the strongest contributions to business impact of every factor tested. Redesigning a workflow is not a procurement act. It is a change in what people do on a Tuesday.
Leaders know where the problem is. Deloitte’s State of AI in the Enterprise, a survey of 3,235 leaders across 24 countries fielded in August and September 2025, found insufficient worker skills named as the biggest barrier to integrating AI into existing workflows. Not model quality. Not licensing.
Our case against buying licenses and calling it progress makes the broader argument. This piece is narrower: how to tell what is actually wrong, and what to buy instead of more seats.
What a Flat Usage Chart Actually Means
Most companies in this situation have one dashboard and one conclusion from it: usage is low, so we need more training. That is one of three diagnoses, and the other two do not respond to training at all.
| What you see | What it usually means | What fixes it |
|---|---|---|
| People logged in once, then stopped | No trigger. Nobody can name the moment in their week when this is the tool they reach for. | Attach the tool to one recurring task per person, with a time and a cue |
| People use it for low-stakes work only | No trust. They tried it on something that mattered, could not verify the output, and quietly retreated. | A verification routine they can run in under two minutes, taught on real work |
| Usage concentrates in one or two teams | No permission. The rules about what data may go in are unwritten, so abstaining is the safe choice. | A one-page written rule with worked examples of allowed and disallowed inputs |
The diagnosis matters because the three fixes cost very different amounts and only one of them looks like a session. Sending a no-trust team to another demo makes things worse: they know what the product does. What stops them is that they cannot tell a good answer from a plausible one in their own domain.
Gallup’s AI indicator shows how much of the remaining gap is reinforcement rather than ability. As of its May 2026 reading, 15% of US employees use AI daily in their role, 30% a few times a week or more, and 52% a few times a year or more. The variable that moves those numbers is unglamorous: 36% of employees in AI-integrating organizations strongly agree their manager supports their team’s use of AI, and those who do are 1.7 times as likely to use AI a few times a week or more. A manager who asks about it on Monday outperforms a curriculum.
The sequencing error underneath most flat charts is covered at length in the real cost of a rollout nobody sequenced — the short version is that licenses almost always arrive before anyone has decided whose job it is to make them matter.
The Habit Mechanics an AI Enablement Program Has to Build
Habits are not a mindset. They have parts, and the parts are specific enough to put in a scope of work.
A cue that already exists in the week. Not “use AI more,” but the Monday pipeline review, the Thursday board pack, the intake email that arrives forty times a week. The cue has to happen whether or not anyone is thinking about AI — that is what makes the behavior repeat without a reminder.
Repetition on work the person actually owns. Practicing on a sample dataset teaches the interface. Practicing on the contract you have to turn around by Friday teaches judgment, and judgment is what transfers.
Reinforcement that is social rather than technical. Someone has to notice. The Gallup finding above is a reinforcement finding: manager attention is the reward signal, and a program with no manager briefing has no reinforcement mechanism.
Friction removed before the session, not after. If the workflow depends on a system the tool cannot reach or a permission nobody granted, the habit cannot form regardless of enthusiasm. Two days of plumbing beforehand beats a second training day.
Those four parts map onto the staged progression in our five stages from pilot to default behavior, and the ordering constraints between them are set out in the rules for sequencing a program. Each of the four is something you can ask a vendor to deliver, and three of the four are missing from most proposals.
The Timeline Nobody Puts in the Proposal
The most-cited study on how long habits take is Lally et al. (2010) in the European Journal of Social Psychology, which followed 96 volunteers performing a self-chosen daily behavior and found a median of 66 days to reach a plateau in how automatic it felt, with a range from 18 to 254 days. Treat that with caution: the sample was small, self-reported and drawn from university volunteers, and it has not been replicated at scale. It is an order of magnitude, not a law.
Even as an order of magnitude it is inconvenient. Two months of repetition sits badly against a ninety-minute vendor session and a quarterly business review, and that mismatch is the most common reason a program gets declared a failure while it is still working.
What a well-run engagement can honestly promise on each horizon:
| Horizon | What should be true |
|---|---|
| The next Monday | Every participant has one named task they will do differently this week, written down |
| Three to four weeks | The easy half of those tasks have stuck; the hard half have surfaced a blocker worth fixing |
| Two to three months | Use on the named tasks happens without prompting, and a second wave of tasks is being proposed by the team rather than by you |
| Two quarters | The pattern library is being edited by people who did not attend the original session |
A proposal implying behavior change inside a week is selling a session and calling it a transformation. One that commits to nothing inside a quarter is selling a retainer. What the first twelve months should contain, month by month, is laid out in our roadmap breakdown.
What to Require From Whoever Runs the Session
Six requirements separate a habit-building engagement from a demo with catering, written as contract language.
- Pre-work on our material. The facilitator reviews real documents and real recurring tasks before the session and builds the exercises from them.
- One named task per participant. Every attendee leaves with a specific recurring piece of their own work that they will do with a model in the loop, and a cue for when.
- A written pattern library, handed over. What worked, by task, in a file we keep and can edit after the engagement ends.
- A manager briefing. The people who assign the work learn what to expect, what to ask for, and what a reasonable turnaround now looks like.
- A follow-up clinic three to four weeks later. Scheduled before the first session, not offered afterward. This is where the hard half of the tasks get unstuck.
- A baseline and a re-measure. One number on one task before, the same number after, using a method we can repeat without the vendor.
Items 2, 3, 5 and 6 get cut when a buyer pushes on price, and they are the four that build the habit. Cutting them turns a capability purchase into an awareness purchase at a discount. What the two words on a quote actually buy is examined line item by line item in what you are actually buying.
Measuring Habits Instead of Logins
License dashboards answer a question nobody asked. Four measures answer the real one, and none of them need new software.
Named-task coverage. Of the people who attended, what share can state and demonstrate a recurring task they now do with a model? A headcount, collected by asking. It is the leading indicator and it moves within weeks.
Cycle time on one task. Pick the most repeated task in the group and time it before and after on a handful of real instances. One task measured honestly beats a portfolio of estimates.
Unprompted second use. Count the tasks the team proposed itself after the engagement. Zero means you bought a session. Several means the habit is generalizing, which is the part that compounds.
The absence test. Would removing the tool tomorrow be noticed by Wednesday? Qualitative, and the most honest signal available.
Why seat-level reporting flatters you is the subject of our look at what usage dashboards overstate: logins measure curiosity, and curiosity is already high.
What AI Enablement Costs When It Is Priced Around Habits
Market rates for hands-on fluency work land roughly between $2,000 and $15,000 per engagement, depending on cohort size, pre-work on your material, and whether a follow-up clinic and a re-measure are included. Custom automation built after a team is fluent runs in a different band, roughly $25,000 to $150,000.
Put those next to the license line. A company of a couple of hundred people with assistant seats across most of the staff is already carrying a five-figure annual bill for access alone, and spending nothing on repetition against a renewal that size is the expensive choice rather than the thrifty one. The automation spend only pays off after the habits exist: a team that does not use a model daily cannot tell you which process is worth removing.
Where the money should go, in order: fix the friction, build the habit on tasks people already own, measure one task honestly, then automate what the team itself nominates. Our playbook for moving from license rollout to measurable capability covers that sequence end to end.
Frequently Asked Questions
What is the difference between AI tools and AI habits?
A tool is access — a seat, a login, an assigned course. A habit is a repeated behavior attached to a specific recurring task and a cue in the week. Tools arrive in weeks through a procurement cycle; habits form over months through repetition on real work. Returns show up on the second, not the first.
Why do employees stop using AI tools after a few weeks?
Usually one of three reasons: no trigger (they cannot name the moment the tool belongs to), no trust (they could not verify an answer on work that mattered and retreated to low-stakes use), or no permission (the data rules are unwritten, so abstaining is safest). Each needs a different fix, and only the first resembles training.
How long does it take for AI use to become a habit?
Plan in months, not weeks. The most-cited habit study, Lally et al. (2010), found a median of 66 days to reach a plateau in automaticity, with a wide range of 18 to 254 days and a small self-reported sample. In practice, expect a changed task the week after a session, visible sticking at three to four weeks, and unprompted use around the two-to-three-month mark.
Is AI training a waste of money?
No, but it is wasted when it teaches features instead of attaching a model to work people already own, when it ends at the session, and when no manager follows up. BCG’s 2026 data found only 36% of employees feel adequately upskilled despite high tool access — a design problem, not an argument against spending.
Who should own AI enablement in a company without a Chief AI Officer?
Whoever owns the workflows being changed, with a named executive sponsor and the people managers as the reinforcement layer. Placing it only in IT produces deployment without behavior change; placing it only in L&D produces completion rates. The owner needs authority over how work is assigned.
How do we measure AI adoption without buying another analytics tool?
Four measures, all collectable by asking: the share of participants who can state and demonstrate a recurring task they now do with a model, cycle time on one repeated task before and after, the number of new tasks the team proposed itself, and whether removing the tool would be noticed within a day.
What does a habit-building engagement cost compared with a standard workshop?
Market ranges for hands-on fluency work sit between roughly $2,000 and $15,000 per engagement. The spread is explained less by session length than by pre-work on your documents, a manager briefing, a follow-up clinic, and a before-and-after measurement. Those four items are what create repetition, and they are the first things cut on price.
Can we build AI habits internally instead of hiring help?
Yes, if one person has real time for it and credibility with the teams involved. The failure pattern is assigning it as a second job to someone with no authority over how work gets allocated. An outside facilitator mainly compresses the first cycle and produces the written pattern library an internal owner then maintains.
Should we mandate AI use to speed adoption?
Mandates reliably produce logged interactions. They do not reliably produce changed work. The data points toward visible manager support instead: employees who strongly agree their manager backs AI use are 1.7 times as likely to use it weekly or more, per Gallup’s 2026 readings. Require the task, not the tool.
What is the first thing to do if usage is already flat?
Diagnose before you spend. Ask ten people what moment in their week the tool belongs to. If they cannot answer, you have a trigger problem, not a skill problem, and another session will not fix it.
Where to Start
Pick one recurring task, one team, one cue. Measure the task before anyone touches it, then build the habit on that task and nothing else for a month. One changed Monday is worth more than a company-wide awareness campaign.
For a second opinion on which of the three causes is flattening your chart, book a free AI-readiness call. We will look at your usage pattern and tell you whether you need a session, a one-page rule, or two days of plumbing — including when the answer is that you do not need us.
Arthur Wandzel
