The short answer
Adoption is the gap between what a trained team can do and what it actually does day to day. Close it with a 30-60-90 rollout: prove one workflow in the first month, standardise and document in the second, and expand to adjacent teams in the third. Measure hours saved and output quality on named workflows, and watch for the stall points — no owner, no guardrails, no workflow redesign.
Key takeaways
- Training is the start of adoption, not the end of it.
- 30 days: one workflow, one team, one measurable result.
- 60 days: document what worked and make it the default way of working.
- 90 days: expand to adjacent teams with the proven pattern.
- Most stalls trace back to a missing owner or an unchanged workflow.
There is a gap between a team that can use AI and a team that does. Training closes the first gap. Almost nothing in a typical program closes the second — which is why so many businesses have upskilled staff and unchanged operations.
Adoption is a separate piece of work with its own design: an owner, a measure, a rhythm, and a short list of stall points to watch for. Here's how to run it.
Upskilling is a capability. Adoption is a behaviour.
| Upskilling | Adoption | |
|---|---|---|
| Question answered | Can they do it? | Are they doing it? |
| Evidence | Session artefacts, confidence | Changed workflow, changed number |
| Timeframe | A day or two | Four to twelve weeks |
| Owner | Facilitator | Internal manager or AI owner |
| Fails when | Content is generic | Nobody follows through |
A 30-60-90 day adoption rollout
Adoption needs a rhythm, not a launch. This is the shape we use with clients, sized for one function at a time.
Days 1–30: prove it on one workflow
- Pick one workflow with a number attached — hours per week, turnaround days, volume.
- Record the baseline before any training. Rough is fine; missing is not.
- Run the session, rebuild the workflow live, agree the guardrail page.
- Name the owner and put the day-30 checkpoint in the diary immediately.
Days 31–60: make it the normal way of working
- The cohort runs the rebuilt workflow on real work, not pilots.
- Owner holds one short checkpoint: what's working, what broke, what got quietly abandoned.
- Fix the friction — access, prompts that need refining, review steps that are too heavy.
- Re-measure the baseline number and compare.
Days 61–90: widen and hand over
- Add the second workflow, or the adjacent function.
- Have a cohort member co-run the next session — this is the handover.
- Report the three numbers upward: baseline, current, adoption rate.
- Decide explicitly whether to widen, deepen, or stop.
We publish an interactive version of this as the 30-60-90 day AI plan, which is easier to adapt than starting from a blank page.
Metrics that survive scrutiny
Pair one output measure with one usage measure. Output alone can be one heroic individual; usage alone can be activity with no result.
| Type | Metric | Watch out for |
|---|---|---|
| Output | Hours per week on a named task | Self-reported drift — ask the same question the same way |
| Output | Turnaround time on a deliverable | Seasonality; compare like periods |
| Output | Rework or revision count | Quality falling while speed rises |
| Usage | Share of cohort using AI on real work weekly | Counting logins instead of work |
| Usage | Number of rebuilt workflows still in use | Workflows quietly reverting to the old way |
Three numbers, reviewed monthly, is enough. A dashboard nobody maintains is worse than a spreadsheet someone does.
Six stall points, and what unsticks them
- No owner. Everyone is responsible, so nobody chases it. Name one person and give them hours, not just the title.
- No baseline. You can't prove a change you never measured. Capture the number before training even if it's approximate.
- Guardrail vacuum. Careful people opt out to avoid risk. One written page fixes more adoption problems than more training does.
- Tool sprawl. Five overlapping subscriptions and no default. Pick one primary tool per job for the first 90 days.
- Bolt-on workflows. AI added as an extra step rather than replacing steps. Redesign the process — see AI training for managers.
- Celebrated win, undefined next step. Momentum dies in that gap. Name the next workflow the same week you announce the first result.
Building the internal capability to keep going
The goal is not permanent dependence on an external facilitator. By day 90 you want an internal owner who can scope a workflow, run a session, and hold a checkpoint. That is the difference between a training spend and a capability.
Where clients want that ownership built deliberately — including systems, guardrails and internal advisors — that's the enablement side of our work, described on our Solutions page.
Frequently asked questions
What's the difference between AI upskilling and AI adoption?
Upskilling is what people can do after training. Adoption is what they actually do four weeks later on real work. Most programs measure the first and assume the second.
How long does AI adoption take?
Expect a first measurable change within 30 days on one workflow, and a habit that survives without prompting somewhere between 60 and 90 days. Anything faster is usually one enthusiast rather than a team.
What are the best metrics for AI adoption?
Pair one output measure with one usage measure: hours or turnaround time on a named workflow, alongside the share of the team using AI on real work weekly. Either one alone is misleading.
Why do AI initiatives stall after the training?
Usually no owner, no measure, or no guardrail. Without an owner nobody chases it, without a measure nobody knows it worked, and without a guardrail cautious people opt out entirely.
Should we appoint an internal AI owner?
Yes, and give them time rather than just a title. A few hours a week from someone respected in the business beats a committee, and it is what lets capability survive after the external help leaves.
How do we keep momentum after the first win?
Make the win visible, then immediately name the next workflow. Momentum dies in the gap between a celebrated result and an undefined next step.
What to do next
Before your next training session, write down one number and who owns it. That single act is the difference between upskilling and adoption. For the program design around it, start with AI training for business.