If you can't fit it on one page, you don't understand it well enough yet. That's the idea behind the A3.
What it is
The A3 is named after the paper size, about 11 × 17 inches. Toyota used it for problem solving, proposals and status reports. The Lean Enterprise Institute and John Shook's book Managing to Learn made it widely known in the West.
A problem-solving A3 typically flows left to right:
| Left side | Right side |
|---|---|
| Background. Why this matters | Countermeasures. What we'll try |
| Current condition. What's happening now, with facts | Plan. Who, what, when |
| Goal. What we want, measurable | Follow-up. How we'll check it worked |
| Analysis. Root cause |
It's PDCA on one sheet. See Plan-Do-Check-Act.
The A3 isn't just a form. At Toyota, it's a conversation tool. The author takes it to the people affected, gets their input, and revises it. The page gets better each time.
Why people skipped it
Writing a good A3 is hard. Fitting everything on one page takes many drafts. The person with the problem is usually too busy to write it well.
How a bot does it
- Give the bot the facts. Data, complaints, notes, the current process.
- Have it draft the whole A3 in the standard sections.
- Have it keep it short. Each section a few lines or one small chart.
- Have it list questions each stakeholder might ask.
- People take the draft and discuss it. The bot revises from their comments.
The draft gets to the conversation faster, and the conversation is the point.
The bot brief
Draft a one-page A3 for: [problem or proposal].
Inputs: [data, notes, process documents].
Sections: Background, Current condition, Goal, Analysis, Countermeasures, Plan, Follow-up.
Rules: the whole A3 must fit on one page. Each section 2-5 lines or one small table.
Current condition and Goal must contain numbers from the inputs, with sources.
Analysis must state a root cause with evidence, or say HYPOTHESIS.
Plan: owner and date for each action (leave owner blank if unknown).
After the A3, list the 3 questions each of these people is likely to ask: [stakeholders].Starting point only. Test it on your own data and model, and adjust as needed before you rely on it.
An example (condensed)
Topic: Too many bot-drafted quotes need rework.
- Background: Quotes are drafted by a bot and approved by an estimator. Rework delays replies.
- Current condition: 18 of 50 quotes in September needed edits (36%). Average delay: 1 day.
- Goal: Under 10% needing edits by December.
- Analysis: 12 of 18 edits were freight charges. The skill has no freight table; the bot estimates freight.
- Countermeasures: Add the carrier's freight table to the skill's sources.
- Plan: Estimator supplies table by 10/30. Skill updated and tested on 50 old quotes by 11/6.
- Follow-up: Track edit rate weekly through December.
One page. Anyone in the building can read it in two minutes.
Where bots differ
The A3's value at Toyota comes partly from the author's struggle to write it. That struggle is how they learn the problem. A bot skips the struggle. So the person who owns the problem should still edit the draft heavily and take it to people themselves. The bot gets them to a good first draft fast. It doesn't replace their understanding.
What this means for you
- Use an A3 for any problem or proposal worth a meeting.
- Let a bot write the first draft from facts.
- Take the draft to the people affected. Revise.
- Keep it to one page.