Enablement
A 90-day AI training plan for employees, week by week
A week-by-week plan to take a team from curious to capable with AI, including who to train first, what to measure and how to stop the drop-off.
The short answer
A 90-day AI training plan should run in three phases: weeks one to three establish a baseline and train leadership and managers, weeks four to eight put frontline teams on their own live work in short repeated sessions, and weeks nine to twelve convert the use cases staff proposed into a small build queue. Measure capability by counting proposed use cases and tasks handed over per person, not by attendance or course completion.
Key takeaways
- Train managers before frontline teams. They are the group who can quietly block adoption.
- Short repeated sessions on real work beat one long webinar on sample data.
- Take a baseline in week one or you will never prove the programme worked.
- End the ninety days with a build queue, otherwise capability decays.
- Track use cases proposed per person per quarter as your headline metric.
Most AI training programmes end with a satisfaction score and no behaviour change. The fix is not better slides. It is sequencing the audience correctly, working on live tasks, and finishing with somewhere for the ideas to go.
Weeks 1 to 3: baseline and leadership
- Week 1: survey current tool usage honestly, including the unofficial personal accounts people already use. Record seat usage, not licence count.
- Week 1: agree the two metrics you will report on at day 90 and write down today's number for each.
- Week 2: half a day with leadership on where AI pays, what it costs to run, and what must never leave your tenancy.
- Week 3: manager workshop. Each manager maps their own team's week and quantifies recurring tasks in annual hours.
The manager session is the load-bearing one. Managers own processes, so they can both spot the automatable work and stall it. Give them the vocabulary to write an automation brief and they become the programme's distribution channel.
Weeks 4 to 8: frontline, on live work
- Two sessions per team, ninety minutes each, two weeks apart. The gap is where the learning happens.
- Everyone brings a real task with real inputs. No sample data, no fictional customers.
- Session one: hand over one recurring task and produce something usable in the room.
- Session two: bring back what failed. Failure analysis is the highest value hour in the programme.
- Close every session by capturing proposed use cases into one shared list with a named owner.
| Group | Time commitment | Goal at day 90 |
|---|---|---|
| Leadership | Half a day plus two reviews | Can challenge an AI business case on cost, error rate and review load |
| Managers | One day plus fortnightly check-ins | Can write a build brief a developer can act on |
| Frontline | Two 90-minute sessions per team | Has handed over at least one recurring task and proposed one more |
Weeks 9 to 12: convert capability into a queue
By week nine you should have a list of proposals from inside the business. Score them on volume, effort and cost of being wrong, pick the top two, and write proper briefs. This is where a training programme either becomes an operating change or evaporates.
- Rank the proposals with the managers who submitted them, not in a separate committee.
- Kill anything that cannot state its failure modes in one paragraph.
- Write briefs with inputs, rules, exceptions, escalation and a definition of done.
- Assemble 50 to 100 historical cases per candidate so the build has something to be scored against.
- Re-measure your two baseline metrics and publish the result internally, good or bad.
What to measure
- Use cases proposed per person per quarter. The clearest signal that capability moved.
- Recurring tasks formally handed over, counted per team.
- Weekly active usage of approved tools, not licences purchased.
- Hours returned against the week-one baseline, and what those hours were redeployed into.
- Number of proposals rejected on governance grounds, which shows people understand the limits.
Our delivery of this is described on the AI training for employees page, and it usually runs alongside or straight after an AI operating review. The build queue it produces feeds AI agents for business.
Skills are not the thing you add after the tools. They are the thing that decides whether the tools were worth buying.
Frequently asked questions
- Who should be trained first on AI?
- Managers, immediately after a short leadership session. They own the processes where automatable work sits, so training them first means frontline sessions arrive into teams whose manager already understands the point and has candidate tasks ready.
- How long should AI training sessions be?
- Ninety minutes, repeated, with two weeks between sessions on the same team. The gap lets people try the tools on live work and return with real failures, which is where the useful learning is. Full-day sessions produce fatigue and no follow-through.
- How do we measure whether AI training worked?
- Count use cases proposed from inside the business per quarter and recurring tasks formally handed over, both against a baseline taken before training. Attendance, satisfaction scores and course completions do not correlate with behaviour change.
- Do we need training if we are buying an off-the-shelf AI tool?
- Yes, and arguably more. Off-the-shelf tools fail on adoption rather than capability, and low seat usage is the most common outcome when licences arrive before skills. Budget training time in the same decision as the licence spend.
Work through this with us
brep runs AI operating reviews, workforce training and agent deployment for businesses of 10 to 250 people in the UK and US. One diagnosis, one workflow, someone accountable for keeping it working.