An incident can be fixed in code while its reasoning remains scattered across chat threads and post-mortems. An A3 report puts the problem, evidence, root cause, and proposed countermeasures into a one-page record that others can use to understand and trust the fix. The page limit is a constraint on clarity, not a requirement to use paper: a Markdown file, slide, or shared document can serve the same purpose. section 8.6
Give each part a distinct job:
- State what happened and why it matters, then quantify the current impact with measures such as latency, error rates, or users affected. section 8.6
- Set a measurable goal. “p99 latency stays below 200 ms after every migration” is more useful than “improve latency.” section 8.6
- Show how the root cause was reached, using evidence from tools such as 5 Whys, a fishbone diagram, an Is/Is Not table, or a Pareto chart. Do not present only the conclusion. section 8.6 section 8.9
- Specify countermeasures and an implementation plan. Distinguish a quick fix from changes that prevent recurrence or detect it automatically. section 8.6 section 8.8
API latency after a migration
A deploy is followed by 40 minutes of elevated latency, 847 timeouts, and three customer escalations. A 5 Whys analysis traces the slowdown to a migration that added an index without running `ANALYZE`; the query planner lacked statistics and chose a full table scan. The A3 separates running `ANALYZE` now from adding it to the migration template and checking query plans in CI. That distinction makes the immediate repair and the longer-term safeguards visible together. section 8.6
A3s also work before an incident. To propose automating a slow staging deploy, describe the time lost, the proposed pipeline change, its cost, and how success will be measured. This gives someone who did not observe the friction enough context to decide asynchronously. Translate the impact for the reader—delivery time for a product manager, recovered capacity for an engineering manager—without changing the underlying evidence. section 10.4
Keep the analysis blameless: people must be able to describe what actually happened for root cause techniques to work. Before sharing an A3, ask whether each countermeasure makes recurrence impossible or merely less likely. Label temporary fixes as temporary, share the record, and follow through on the proposed changes. section 13.2 section 8.8
Listing 8.5 A markdown A3 report template that can be checked in
# A3 Report: API Latency Incident 2026-02-12 ## Background API p99 latency spiked to 8s for 40 minutes following the 14:32 deploy. ## Current State 12,000 requests affected. 847 timeouts. 3 customer escalations. ## Goal p99 latency stays below 200ms after every migration. ## Root Cause Analysis 5 Whys traced to a migration that added an index without running `ANALYZE`. The query planner had no statistics for the new index and chose a full table scan. Response time was 200x slower than expected. ## Countermeasures | # | Action | Layer | Status | |---|--------|-------|--------| | 1 | Run `ANALYZE` on affected tables | Fix | Done | | 2 | Add `ANALYZE` to migration template | Prevent (standard work) | This sprint | | 3 | Add query plan assertion to CI | Detect (automated check) | Next sprint | ## Implementation Plan - **#1** — @dbadmin — completed 2026-02-12 - **#2** — @platform-team — PR by end of sprint 14 - **#3** — @platform-team — spike in sprint 15 ## Follow-up Monitor p99 after each migration for 4 weeks. Alert if p99 exceeds 200ms within 30 minutes of any deploy. ## Learnings Index creation without a statistics update is a latent failure. The query planner trusts stale stats and will choose a table scan over a new index it does not know about.
The check-in-ready Markdown template shows how incident measures, a traced cause, and fix/prevent/detect actions fit into one report. section 8.6
Table 10.5 A3 stakeholder adjustments to land the pitch right
|
Stakeholder
|
They care about
|
Translate to
|
|
Product Manager
|
Feature velocity, time to market
|
Lead time reduction, cycle time improvement
|
|
Engineering Manager
|
Team capacity, retention, predictability
|
Developer hours recovered, reduced frustration
|
|
CTO/VP Engineering
|
System reliability, scalability, risk
|
MTTR improvement, reduced incident frequency
|
|
Finance/Exec
|
Cost, ROI, risk mitigation
|
Dollar cost of waste, payback period
|
|
Security
|
Vulnerabilities, compliance
|
CVE exposure window, remediation speed
|
The stakeholder table helps translate the same improvement into relevant measures without reducing the proposal to a request for engineering time. section 10.4

