The Task Autopsy: A 60 Minute Workshop to Find Your Team's First Real AI Use Case

Most AI training ends the same way. People leave energised, full of ideas, then Monday morning arrives and nothing in the actual work changes. The gap is not knowledge. It is translation: nobody spent an hour connecting what AI can do to what the team concretely does every week.

The Task Autopsy closes that gap in one hour. The workshop borrows from design thinking, journey mapping and dot voting, and applies them to a single recurring task. The deliverable is not a list of ideas. It is one AI assist tested live on a real piece of work, with a named owner.

Objective

Identify where AI genuinely helps in one recurring team task, test it live, and leave with one working use case and a named owner.

Participants

4 to 8 people who actually perform the task. No observers. If a manager attends, they participate as an equal.

Duration

60 minutes.

Materials

A whiteboard or a large sheet of paper, sticky notes in two colors, markers, one laptop connected to your AI tool of choice, projected so everyone can see the screen, and a visible timer.

The run sheet

Step 1, choose the victim (5 min). Before the session, ask the team to nominate recurring tasks that feel heavier than they should: the weekly report, the client onboarding email, meeting preparation, the monthly newsletter. In the room, the team picks one. Rule: it must happen at least weekly and take at least 30 minutes each time.

Step 2, dissect (15 min). Map the task step by step on the whiteboard, one sticky note per step. Push for honesty: not the official process, the real one, including the hunt for the previous version and the reformatting. Most teams discover the task has 12 to 15 steps when they thought it had 5.

Step 3, diagnose (10 min). With the second color, mark each step with one of three labels. Judgment: needs a human decision or context. Grunt: repetitive transformation, drafting, formatting, summarising. Glue: waiting, chasing, transferring between tools. AI helps most with grunt steps, sometimes with glue, rarely with judgment. This reframe matters: the goal is not to automate the task, it is to relieve the heavy steps and protect the judgment steps.

Step 4, dot vote (5 min). Each person gets two dots to place on the grunt or glue steps where help would matter most. Take the winner.

Step 5, test live (15 min). One person drives, screen projected. The team writes a prompt together for the winning step, using real material from the last time the task was done. Run it. Critique the output honestly: what is usable, what is wrong, what is missing. Iterate twice. This is the moment the workshop earns its hour: people see a real output on their real work, with its real flaws.

Step 6, commit (10 min). Name one owner who will use the assist on the next real occurrence of the task, and set a 10 minute follow up in two weeks to check two things: whether the assist was actually used, and whether the output is good enough to keep.

Debrief questions

Which steps surprised the team during the mapping? Where did the judgment versus grunt labels spark debate? What did the live test reveal that the discussion did not? What conditions would make this assist the default way of doing that step?

Common failure modes

Choosing a task nobody in the room actually performs: the map turns theoretical. Mapping the official process instead of the real one. Letting the most senior voice pick the step instead of the dot vote. Testing with invented examples rather than real material, which makes output quality impossible to judge. Skipping the commit step, which turns the session into a pleasant conversation. And trying to fix the whole task instead of one step: one assist that survives contact with a real Monday beats five ideas that stay on the whiteboard.

In MAKIA terms, this exercise works on Knowledge and Impact at the same time: the team builds capability by testing on real work, and the result is measurable because the task recurs every week.

Try this week

You do not need the full workshop to start. Take one recurring task of your own, list its steps in 10 minutes, then label each one: judgment, grunt or glue. The labelling alone usually reveals the first candidate step for an AI assist. Under 30 minutes, no meeting required.