AI enablement

What does AI enablement actually include?

Four jobs, and in a small firm all four land on the owner. One clarification first, because the word is overloaded: this is not sales enablement, which is a separate discipline about equipping a sales team and shares nothing with the meaning below.

Data design. How the work really gets done lives in people’s heads, and AI cannot reach it there. This is the least interesting of the four and it decides the other three. Gallup found that within organizations that make AI available, 88% of employees who strongly agree that AI fits their existing systems and processes use it frequently, against 55% of those who do not. Fit is most of the difference, and getting it means writing the process down before buying the agent.

Process redesign. Attaching a tool to an unchanged process is the default move because it asks nothing of anyone. It is also why most rollouts produce a faster version of the old workflow and no change in the numbers.

Change management. Three things sit here and only one is training: the hours people need, written permission to use AI on real work, and a decision about what the freed time is for. Slack found that 45% of workers have no explicit permission to use AI, which is a sentence an owner can fix in a morning.

Prototyping. The job no vendor definition contains, and the one where the learning happens. Build something small, put it in front of a person, and change it based on what breaks.

Is AI enablement the same as AI training?

No, and the difference is where each one ends.

Training is a purchase with a completion date. It finishes when the sessions are over and somebody reports attendance. Enablement has a business condition on the exit instead, so it is not finished until a number moves.

The evidence says training is not the constraint most firms think it is. In an interview study of ten experienced professionals using Copilot at work, no participant named formal training as their main way of learning it. Eight of the ten learned by trial and error and six by swapping tips with colleagues. Ten people is a small study, so read it as a pattern rather than a rate, but it points the same direction as the spending data: more curriculum does not move the thing that is stuck.

Why do most AI enablement programs stall?

Because they are sold as readiness, and readiness has no finish line.

Read the definitions on the first page of a search for the term and they end at prepared, equipped, ready. Each one describes real components, and none of them says what has to be true for the work to be over. A program with no exit condition can run, and be invoiced, indefinitely.

Gallup measures the result: 47% of U.S. employees say their organization has integrated AI, while 30% use it regularly. Their own summary is that implementation does not guarantee use. That gap between deployed and used is the whole territory the word is supposed to cover.

What does AI enablement look like in practice?

A loop with five steps and no fixed length. It runs until the number moves, then stops.

Name the number first. One workflow, one measure that would change if this worked. A firm that cannot name it is not blocked on tooling, and finding that out costs nothing.

Write down what only lives in people’s heads for that workflow. Not the whole firm. The checks a manager runs without thinking, the exceptions, the conditions that make a case unusual.

Build the smallest version and put it in front of one person. Not a pilot with a launch date. A thing somebody uses on real work this week.

Make it cheap to report that it broke. This is the step most firms skip and the one that carries the loop. When someone says what they tried and what went wrong, the firm learns two things at once: what people want the tool to do, which nobody wrote down, and what has to be fixed next. A failed attempt is a roadmap that arrived for free.

Change it and hand it back. Read which of the four causes killed it, then cut, rebuild, or point it at a smaller question. Repeat until the number moves.

Nobody has a general formula for which build works, which is the honest reason this is a loop rather than a sequence of phases. The firms that get somewhere are the ones that can run the loop cheaply enough to run it several times.

Who owns AI enablement in a small firm?

Not an enablement lead, and usually not a committee.

At 25 people the workflow layer and the training layer are the same two people, and one of them is the owner. What works instead is the person who already does it without being asked: someone who helps colleagues improve their workflows, writes down what works, and builds small tools to kill repetitive work.

One condition decides whether that person is enough. They have to be ahead of the people reporting problems to them, and those people have to believe a report gets fixed. Take that away and reporting a problem is just admitting you could not make something work, so people stop reporting and quietly go back to the old way. That silence is what makes adoption numbers look better than they are, since 78% of AI users already bring their own tools to work, 80% at small and medium companies.

What does AI enablement cost?

More than a licence and less than a transformation program, and the expensive part is not the software.

The cost is the failed builds. A prototype that does not reach production still bought information: which data was missing, whether the problem was worth solving, whether anyone would use the output, whether the results held up. Firms that treat those as waste stop after the first one. Firms that budget for them get to the version that works.

The hours are the other real cost, and in a service firm they compete with billable work. That is a scheduling decision an owner has to make on purpose, which is the argument in nobody gave them the hours.

How do you know AI enablement worked?

A number in the business changed.

Pick it before starting: cost to deliver on one service line, hours a recurring task consumes, turnaround time a client complains about. If no number can be named, that is the finding, and it comes before any tool decision.

Seat counts, licence utilization, and course completions are not that number. They measure whether the thing was bought and attended, which is exactly the readiness trap the term keeps falling into.

Where does AI enablement sit on the Delivery Model Ladder?

It is the work of moving from Stage 1 to Stage 2 on the Delivery Model Ladder.

Teaching people to prompt on top of an unchanged process is Stage 1, Enhanced: cost to deliver may fall, and nothing else moves. Stage 2, Augmented, is a rebuilt workflow with the freed time assigned a job. Enablement is the name for the distance between them, and the failed prototypes are what crossing it costs.

The failure modes on the way are catalogued in the Production Gap. The one this term runs into most often is the Owner Ceiling, because all four jobs are the owner’s and none of them is getting staff to want it.

When is a firm not ready for this?

Three conditions, stated plainly.

It needs someone with authority, not an enthusiast with initiative. An enthusiast can carry the tools and cannot carry the decisions.

It needs a nameable number. Without an exit condition the loop never terminates and quietly becomes the readiness program it was supposed to replace.

And below a certain size there is no workflow to redesign. A firm of three people is splitting work between the owner and AI, which is a different problem.

The full argument, with what the loop cost to run in practice, is in AI enablement is a loop, not a roadmap.