The shape of the problem.
A generative AI launch can stall a few months in. The early adopters have logged in. The skeptics have not. Active-use charts look flat. Leadership starts asking whether the spend was worth it. McKinsey's 2026 State of AI survey shows the same gap at scale: eight in ten respondents say AI has improved their own productivity, while only 37% report any contribution to their company's EBIT.
Underneath, a handful of workflows have been rebuilt with AI in them, and the rest are unchanged. The flat curve is real, because no one in the middle has been given a redesigned workflow to step into. Adoption is workflow work.
The three metrics.
- - Active use: weekly active users on sanctioned AI. The vanity metric. Necessary, not sufficient.
- - Depth: average number of redesigned workflows a user participates in. The honest middle metric. If depth is stuck at one, the program is not compounding.
- - Outcome: the business result tied to each redesigned workflow. Time-to-brief, concept-to-approval, cost-per-asset, customer NPS. One per workflow.
The four most common stall points.
- AI sprawl. Eight overlapping AI assistants and no clear primary. Users default to ChatGPT for everything and never learn the workflow-specific ones.
- No workflow redesign. AI gets bolted onto the old process. Time saved disappears into more rounds of the same work, not better work.
- Leader exemption. The CMO talks about AI without using it. Direct reports notice; adoption stalls at the manager layer.
- Generic training. Prompt-craft workshops with no link to the work people are paid to do. People learn, then return to their unchanged inbox.
A 90-day pattern that works.
- Days 1-15. Pick three workflows. Choose workflows that already have an owner and a metric. Concept-to-brief, brief-to-creative, creative-to-approval is a good starting trio for a marketing team.
- Days 16-45. Rebuild each workflow with AI in the loop. Run them as working sessions. Owner is on the keyboard. New steps documented at the end of each week.
- Days 46-75. Roll the redesigned workflows to the rest of the team. Pair experienced users with newer ones. Measure depth as well as active use.
- Days 76-90. Read the outcomes against the original metrics. Cut what did not work. Pick the next three workflows. The cycle repeats.
Related reading.
- - AI adoption strategy: the eight-step plan that sits above the 90-day pattern.
- - AI change management framework: the human side of the workflow redesigns above.
- - AI readiness assessment: the read that should come before the 90 days.
- - The AI Profit Partnership: the done-with-you engagement that runs the 90-day pattern with your team.
- - Our book, The Elephant in the Algorithm: the longer argument behind the guide.
Questions people ask.
What does 'generative AI adoption' mean?
The proportion of work, in a defined team or workflow, where generative AI is now part of how the work is done: producing first drafts, summarizing inputs, structuring decisions, generating variants. Logging in once does not count.
Why is generative AI adoption different from previous tech adoption?
Three reasons. Generative AI is general-purpose, so adoption happens one workflow at a time. AI models change quarterly, so the workflow that worked in Q1 needs rework in Q3. And value is hard to attribute, so adoption gets measured by logins instead of outcomes.
What is the right adoption metric for generative AI?
Three numbers. Active use (how many people use it weekly), depth (how many workflows it touches per user), and outcome (the business result the workflow was supposed to move). Active-use alone is a vanity metric.
How long until adoption pays off?
First useful results in 6-12 weeks for a defined workflow. Compounding value at the team level takes two to three quarters, because the workflow itself has to be rebuilt. A 30-day transformation covers the launch, and the value comes later.