Using AI for A3 problem solving: what to hand over, box by box

AI can draft the A3 boxes that compile what the team already knows, and challenge the boxes that carry the team’s reasoning. Box names vary by template; in the order of John Shook’s A3:

• Background: AI drafts it from the context you give; it cannot know on its own why the problem matters.
• Current condition: AI compiles the trend and the Pareto from the data, citing the cells; the team goes and sees.
• Target (goal): set by the team; AI checks that it names a KPI, a baseline, a value and a date.
• Analysis: the author’s work, with the team; AI sorts each step into fact, assumption or logical jump.
• Countermeasures: decided by the team; AI builds the table and flags root causes left without a countermeasure.
• Implementation plan: AI lays out the table; owners and dates come from the team.
• Follow-up: AI compares actual with target; the team decides when the A3 is closed.

These A3 prompts for an AI assistant use only what you give them, treat empty cells as missing and say what is missing instead of filling it. Replace the parts in [brackets]. The last prompt is for the mentor, not the author.

Draft the Background box of an A3 from the context we paste here: [paste]. In three lines, say why this problem matters now to the business (customer, safety, quality, cost or delivery), using only what the context states. If the context does not say why it matters, ask instead of writing it.

This sheet holds [defect / delay / scrap] data for [period]. Draft the Current condition box of an A3: the trend, and a Pareto of the main defect categories (or whatever the data is stratified by), citing the cell range each figure comes from. Describe, do not explain: no causes in this box. Treat empty cells as missing, not zero, and list them. End with what we should go and observe on the floor to confirm the picture.

Here is our target: [paste]. Check it without changing it: does it name a KPI, a baseline, a target value and a date, and is it consistent with the current condition in the sheet? List what is missing as questions. If we gave no target, write [TARGET: to be set by the team, from X to Y by <date>].

Here is our 5-whys chain: [paste]. Challenge it like a lean coach. For each 'why', say whether it is a verified fact, an assumption or a logical jump, and what we should go and see on the floor to confirm it. Flag as a logical jump any step that blames a person or ends in 'lack of training' without data to show it. Do not propose a root cause yourself.

Here are the root causes we validated and the countermeasures we decided: [paste]. Create a sheet named 'Countermeasures' with these columns: countermeasure, root cause addressed, owner, due date, expected effect on the target KPI, how we will check it. Fill only what our notes state; leave owner, due date and expected effect empty unless we gave them. For any validated root cause that no countermeasure addresses, write a question for the team, not a countermeasure. Flag any countermeasure that is only training or a reminder to be careful, and ask what changes in the process instead.

Here is our A3 follow-up plan: [paste], and the KPI values since the countermeasures started are in [range]. Compare actual with target, month by month, treating empty cells as missing. Say whether the target is met, not met, or not yet measurable. List the questions for our next check: what to verify on the floor, and where else the countermeasure could apply. Do not declare the A3 closed.

Act as an A3 mentor preparing a coaching session. Read this A3: [paste]. Write the questions you would ask its author, box by box, in the order of the story: where the data is thin, where a step jumps, where the target or the countermeasures do not follow from the analysis. Questions only, no answers and no rewritten A3.

The A3 is a learning tool before it is a report. John Shook’s Managing to Learn (Lean Enterprise Institute, 2008) describes it as the support of a dialogue between an author and a mentor: the value is in the questions the author has to answer, the floor visits they trigger, and the reasoning that gets corrected on the way.

An assistant asked to write the analysis skips that learning and returns a plausible cause that nobody verified. Plausible is the dangerous word: a well-written 5-whys hides its jumps. Used as a sparring partner, the same assistant does the opposite: it points at the jumps and sends the team back to the floor.

A fictional supplier sees scrap on a CNC machining line rise over a quarter, with the overtime it takes to make up the lost parts. The team pastes three lines of context into the background prompt and gets a draft that names the scrap cost and the overtime, and asks whether delivery dates are at risk.

The current-condition prompt returns a trend and a Pareto from the scrap log: burrs first, dimensional defects second, with the cell ranges behind each share. It also lists three weeks where the night-shift log is empty, the weeks a new operator started: a gemba question, not a data-cleaning task. On the floor, the team finds the log simply was not filled, and handles that gap separately.

The 5-whys challenge marks “the new night operator was not trained” as a logical jump: nothing in the data says so, and the burrs appear on the day shift too. The team goes to the machine, finds the cutting tool well past its usual life and no tool-change frequency in the standard, and writes that as its own analysis. Drafting the countermeasure table then takes a few minutes; the follow-up itself waits for the data.

A good A3 also answers a strategic question: which of this year’s priorities does this problem stand in the way of? When the initiative it serves links back to it, a strategy review can see which priorities are blocked on an open root cause.

In hoshin.app, that link sits on the initiative itself, next to its status and its KPI, so the A3 is one click away from the review that discusses it. The other lean prompts, for the X-Matrix, the 8D and the PDCA, are gathered in our guide to AI prompts for lean.