AI enablement works as an operating discipline and fails as an event. A company that runs it as an event buys a day of training, watches a room light up, and finds three weeks later that the lift is gone. A company that runs it as a discipline gives the work an owner, a slot on the calendar, a line in the budget, and a number it reports.
What splits the two is not the quality of the session. It is whether anything exists to catch what the session started. Most of what gets written down as an AI enablement strategy describes the first thing while promising the second.
What an event-shaped program looks like
The shape is consistent enough to describe from memory. Licenses go out to 150 people. The usage curve rises for a week and flattens. Somebody asks for training, a session gets booked, and it scores high on the feedback form.
Then the calendar goes quiet.
That program produced an attendance list and a satisfaction score. It did not produce a changed workflow, a prompt library anyone opens twice, a second cohort, or a number anyone can take to a budget meeting. The session was real; the program around it was a purchase.
This is not a failure of judgment by the people who buy it. A workshop is the only format most procurement systems know how to process: it has a date, a price, a deliverable, and an invoice. A standing capability function has none of those and is harder to approve, so the event wins by default, not on merit.
The evidence that one session does not hold
Training is spreading faster than capability, and the gap is measurable.
Section’s AI Proficiency Report, published in June 2026 from a survey of 5,026 US knowledge workers, found that 62 percent say they have received AI training, up 40 percent in six months.
In the same population, 5.5 percent meet the report’s bar for proficiency. The largest group by a wide margin, 73.5 percent, sits in the tier the report calls experimenter.
That pair of numbers is the whole argument. Training reached most people. Competence reached almost nobody. An event is good at moving someone from novice to experimenter and has no mechanism for the step after that.
The employee-side data says the same thing from a different angle. BCG’s fourth annual AI at Work study, published in June 2026 with close to 12,000 respondents across more than a dozen markets, found that 36 percent feel they have received adequate upskilling while 72 percent say the skills they need have shifted.
Among frontline regular users of AI, 66 percent get limited or no guidance on what to do with the time the tools save them.
That last figure is the event model’s signature. A session can teach someone to draft a first pass in four minutes instead of forty. Only a program decides what the other thirty-six minutes are for.
BCG’s companion research on workforce transformation, from February 2026, sizes the gap. The companies it classes as furthest along are four times more likely to run structured AI-learning programs and to protect time for people to learn in. They plan to upskill more than half their workforce, against 20 percent at the laggards. The report makes the format argument directly: skills get built into daily work rather than through once-a-year workshops.
Access itself is a settled question. Stanford HAI’s 2026 AI Index puts organizational AI adoption at 88 percent. Nobody is competing on whether their staff can open a chat window.
The two models, side by side
The two models get described as different levels of commitment to the same thing. They are different things, and the difference shows up in six places.
| Event | Operating discipline | |
|---|---|---|
| Budget line | One-time, in the year it happens | Recurring, survives the planning cycle |
| Owner | The vendor, until the invoice clears | A named internal person with hours |
| Cadence | A date | A frequency |
| Source of material | The vendor’s curriculum | The team’s own live queue of work |
| What it measures | Attendance and satisfaction | Task quality, time, and who still uses it in month six |
| At license renewal | Nothing to show, so the ask repeats | A reported number, so the ask compounds |
Read the right column as a specification. Every row is a thing you either have or do not, and all six are checkable in an afternoon.
Why a six-step roadmap is still an event
The better vendor content in this category has diagnosed part of the problem. A Training Magazine piece from January 2026 calls it a sequencing error: teams define a learning outcome before an operational one, train before finding the real constraint, then measure course completion instead of decision quality. That critique is correct, and sharper than most of what ranks near it.
It is also still a roadmap, and a roadmap has a step six.
After step six there is no step. Nothing in the format says who runs the review in month nine, what happens when a third of the trained group leaves, or where next quarter’s material comes from once the consultant’s deck runs out. A six-step sequence is a longer event: better than a one-day one, and not a different category of thing.
An operating discipline has no final step. It has a frequency, an owner, and a number, and it keeps producing all three after the people who set it up have moved on. That is also the distinction underneath the broader argument that buying licenses is not the same as building capability.
Five parts of a cadence that holds
Five things have to be true. Fewer than five and the program decays back into an event within two quarters.
A named owner with hours that exist
Not a committee and not a vendor. One named person with a defensible number of hours. At 50 to 500 people the honest figure is half a day a month: small enough to survive a bad quarter, large enough to run a review and prepare a session. An owner with no protected hours is a volunteer, and volunteers lose to their day jobs.
A rhythm, not a day
A quarterly session with a 45-minute monthly check between them beats one full day at the same cost. The reason is retrieval: capability forms when a skill gets used again under slightly different conditions, which a single day cannot do. The habit argument is made more fully in the case for building AI habits rather than buying more tools.
Work that comes out of the team’s own queue
The material for session two should be documents the team has to produce in the next two weeks. Generic exercises teach the tool. Live work teaches the job, and it also yields the before-and-after evidence that the measurement loop needs. This is the single cheapest upgrade available to an existing training plan.
A measurement loop short enough to survive a busy quarter
Two or three numbers, collected in under an hour. Share of the trained group still using the tools weekly, time on one named recurring task, and a quality check on a sample of output. Anything heavier gets skipped in the first bad month, and a skipped measurement is no measurement.
Governance on the same calendar, not a separate project
The NIST AI Risk Management Framework, the voluntary US standard published in January 2023, organizes its work as four continuous functions: govern, map, measure, and manage. It is written as a loop for a reason. Approved-use rules, data handling, and review of what people are pasting into these tools belong in the same recurring meeting as the capability review, because both decay on one schedule.
What the discipline costs to keep running
Start with the part that has a market price. A hands-on LLM fluency workshop run against the team’s own documents falls in a $2,000 to $15,000 range, depending on headcount and how much material is built from scratch. A follow-on automation project, once a team can specify one, runs $25,000 to $150,000.
The sustaining cost is mostly time rather than invoices: half a day a month for the owner, 45 minutes a month per participant, and a quarterly session. For a 60-person rollout that is around 150 hours a quarter across the whole company, or under three hours per person.
Weigh that against the license line it protects. BCG’s estimate is that about 10 percent of AI value comes from algorithms, 20 percent from the technology to implement them, and 70 percent from the people side. Most budgets invert that ratio, which is the arithmetic behind funding enablement before the next license renewal rather than after it.
Who runs it when nobody’s title says AI
At 50 to 500 people there is no Chief AI Officer and there is not going to be one. The work lands on a COO, Chief of Staff, Head of L&D, or functional VP as a second job, usually because they asked a good question in a meeting.
Design for that person. A cadence that needs a transformation office will not run. A cadence that needs half a day a month from someone who already has a full calendar might.
Most of the gain is not in the central owner anyway. In BCG’s February 2026 data, 88 percent of managers at the most advanced companies role-model AI use in their own decisions and daily operations, against 25 percent at the laggards. What that implies is that the owner’s real job is equipping the handful of managers whose teams are in scope to run a 20-minute version of the cadence inside meetings that already exist. Turning a license rollout into measurable team capability is largely a question of how much of the work you can push down to that layer.
When a single session is the right call
An event is a legitimate start and an illegitimate finish. Four cases where booking one session is the right decision:
- You need proof of appetite before asking for a budget line. One session with a real measurement attached is a cheap way to earn the second conversation.
- The content genuinely is one-time. An approved-use briefing after a policy change does not need a quarterly rhythm.
- The team is small enough that the cadence is already there. At 12 people, the standing agenda item in an existing weekly meeting is the program, and it costs nothing.
- You are choosing between vendors. A paid pilot session tells you more about a provider than any proposal will.
In all four the session is a step in something. The failure mode is only ever the session that is the whole thing.
A five-question test
Five questions, answerable in a few minutes, that separate a program from a purchase better than any maturity model.
- Who has this on their calendar next month, by name?
- What number do you report, and to whom?
- Where does the material for the next session come from?
- What is the budget line called, and does it recur?
- What happens when a third of the trained group leaves?
Four or five clear answers and you have a discipline. Two or fewer and you have an event with a plan attached, which is the most common state and the most fixable. A fuller definition of what the term covers and excludes is in our guide to what AI enablement covers.
Frequently asked questions
What is an AI enablement strategy?
An AI enablement strategy is the plan for converting purchased AI access into demonstrated capability on real work. A complete one names an owner, a cadence, a source of practice material, a measurement loop, and a recurring budget line. A plan that names only a training date is a procurement decision.
How is AI enablement different from AI training?
Training changes what a person knows. Enablement changes how a function works, and keeps working when that person leaves. Training is one input to enablement, alongside workflow changes, usage rules, measurement, and manager behavior. Buying only that input is the most common version of the mistake.
How often should AI enablement sessions run?
A quarterly working session with a short monthly check between them is the lightest rhythm that holds, and it beats one full day at equal cost. The monthly touch matters more than its length: 45 minutes of applied practice on current work beats three hours of new material, because a skill has to be used again to stick.
Who should own AI enablement at a company of 50 to 500 people?
One named person with protected hours: usually a COO, Chief of Staff, Head of L&D, or the VP whose function gains most. The job is less about delivering content than equipping line managers to run short versions of it in meetings that already exist.
What does it cost to keep an AI enablement program running?
The visible cost is delivery, in a $2,000 to $15,000 market range per session depending on headcount and customization. The sustaining cost is time: about half a day a month for the owner plus 45 minutes a month per participant. Across a 60-person rollout that is around 150 hours a quarter, or under three hours per person.
How do you measure whether AI enablement worked?
Three numbers, collected in under an hour. The share of the trained group still using the tools weekly in month three, elapsed time on one named recurring task before and after, and a quality review of a sample of output. Attendance and satisfaction scores measure the session, not the capability.
Does a one-day workshop still have a place?
Yes, as an opening move. One session is right when you need proof of appetite before asking for a budget, when the content is genuinely one-time, when an existing meeting already carries the rhythm, or when you are comparing vendors. It is wrong as the entire program.
What if we have no L&D function at all?
Most companies this size do not have one, and the cadence does not need it. It needs an owner with half a day a month and managers willing to spend 20 minutes of an existing meeting on applied practice. Buy the delivery, keep the rhythm in-house.
How long before an AI enablement program shows a result?
Expect a measurable change on one named task within a quarter, and a visible shift in how a function works across two or three. Anything promising a transformation in six weeks is selling an event; anything promising nothing for a year has no measurement loop.
What is the first thing to change in a program that already stalled?
Change where the practice material comes from. Running the next session on documents the team has to produce in the following two weeks costs nothing extra, lifts retention, and yields the before-and-after evidence the budget conversation needs.
Key takeaways
- An event moves people from novice to experimenter and has no mechanism for the step after. Section’s June 2026 survey found 62 percent trained and 5.5 percent proficient.
- The six markers of a discipline: a recurring budget line, a named internal owner, a frequency rather than a date, practice material drawn from live work, a measurement loop, and a number reported at renewal.
- A six-step roadmap is a longer event. The format has a final step, and capability programs do not.
- Design the cadence for someone doing this as a second job: half a day a month for the owner, 45 minutes a month per participant, quarterly sessions.
- Line managers carry more of this than the central owner does. BCG recorded 88 percent manager role-modeling at the most advanced companies against 25 percent at the laggards.
If you are holding a flattened usage dashboard and a training line nobody has approved yet, the useful next step is a short diagnostic conversation rather than a session booking. Book a free AI-readiness call and we will work through which of the five parts you have and which one to fix first.
Arthur Wandzel