The shape of the problem.
Most generative AI rollouts hit the same wall around month four. The early adopters have logged in. The skeptics have not. Active-use charts look flat. Leadership starts asking whether the spend was worth it. The pattern shows up across McKinsey's State of AI research, which finds that generative AI use has spread fast at the individual level while sustained, function-level value capture remains rare.
What is actually happening underneath: a handful of workflows have been genuinely 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, not change management theatre.
Three metrics, not one.
- - Active use - weekly active users on sanctioned tools. 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. MIT Sloan Management Review's research on AI organizational learning reaches the same conclusion: companies that move past single-use-case AI are the ones treating it as a continuous workflow change, not a launch.
- - 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.
- Tool sprawl. Eight overlapping tools 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. Harvard Business Review's coverage of generative AI in the enterprise keeps surfacing the same finding - the teams that move are the ones where senior leaders are themselves on the keyboard.
- 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. Not pilots - 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. Working sessions, not training. 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, not just 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.
- - Our AI Alignment Playbook - 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 playbook.
Questions people ask.
What does 'generative AI adoption' actually 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. Not the proportion of people who have logged in once.
Why is generative AI adoption different from previous tech adoption?
Three reasons. The tools are general-purpose, so adoption is workflow-by-workflow rather than rollout-by-rollout. The 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, not one. 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. Anyone selling 30-day transformation is selling the launch, not the value.