Enable
70%
of organizations use generative AI in at least one function. Agent use is still in the single digits.
Stanford HAI, AI Index 20261The problem is not that your team lacks AI tools. The problem is that the workflow still depends on manual coordination.
Companies issue licenses, run a lunch-and-learn, and then wonder why cycle time did not move. People will try a chat window. They will not change how a case gets from request to result unless the path is redesigned and expected.
How I see it
Adoption is a workflow change, not a license rollout
Adoption work starts with a job, not a tool. Pick the steps where people already lose time: drafting from messy context, searching for the last decision, preparing a packet, writing the same update into two systems.
Then make the new path the easy path. Provide the prompt or the workflow in the place the work already happens. Show a completed example. Name what is not allowed. Measure the step, not the login count.
Managers have to ask for the new artifact. If the packet can still arrive as a messy email, people will keep sending messy email. Adoption without a changed definition of done is optional.
This is the first of the three paths: improve the work you already have. It is the right move when the process is sound and a few steps create the friction. It is the wrong move when the process itself is the problem.
How the engagement runs
What you actually buy.
A service is a sequence with an artifact at each step. It is not a transformation program you purchase as a bundle.
Change the expected path
The new artifact is how the work is done, not an optional assistant beside the old way.
Done is redefined for a named step.
A license count or a lunch-and-learn.
Train on live cases
People produce the new artifact on real work, with boundaries and a quality standard.
Completed cases meet the standard.
A generic demo of the tool.
Make managers accept the new done
If managers still accept the old packet, training will not hold.
The old path is no longer a successful submission.
Inspiration for the whole firm and no change in what gets approved.
Do not buy this for
These engagements fail for predictable reasons.
Rolling out a tool to everyone at once
A company-wide Copilot launch creates anecdotes. A function-level change to one step creates a baseline.
Training on features instead of cases
People remember how to rewrite a paragraph. They need to see their actual case done the new way.
Leaving quality undefined
If nobody says what a good AI-assisted packet looks like, you will get volume and inconsistency.
Before we start
Questions that decide whether to engage.
Which step in an existing workflow should look different in 30 days?
If the answer is 'people should use AI more,' adoption has no target.Where will the new path live: in the system of work, or in a side chat window?
Side windows get abandoned when the week gets busy.What will a manager reject if it arrives the old way?
If nothing will be rejected, the old path remains the real path.What is not allowed: data, actions, or claims?
Adoption without boundaries creates shadow risk.How will you know the step got faster or cleaner?
License reports will not tell you.
How we size it
Adoption that countsshare of cases using the new step × measured change in time or quality on that step
High usage with no change in the step is not adoption of a better workflow. It is curiosity.
This is the wrong engagement if
Do not start this engagement yet.
- The process itself is the bottleneck and a copilot will not fix the queues.
- Leadership wants a company-wide launch more than a changed step.
- There are no examples from the team's actual work.
- Security will not permit the work data inside the tool, and nobody will say so.
A useful next step
Bring the team that is already using AI unevenly.
We will pick one step, define the new artifact, and make the old path harder to use.
Discuss an AI opportunity