Using ops method

Using the Ops Method

Operations analyses fail in predictable ways: answering a question nobody asked, computing on data nobody validated, reporting a metric nobody defined, and ending without a decision. The method below exists to make those failures impossible by construction. Follow the sequence; do not skip steps because the task looks small. Small tasks skip the brief; wrong answers follow.

The sequence

Step	Question it answers	Use	Artifact
1. Brief	What decision does this analysis feed?	templates/analysis-brief.md	Filled brief, confirmed by the requester
2. Data contract	What data, what columns, what quality?	templates/data-contract.md	Contract + quality-gate results
3. Honest metrics	What is the true number, under which definition?	route: delivery → otif-analysis · forecasts → forecast-accuracy-review · portfolio/stock → abc-xyz-segmentation, safety-stock-review	Metric ladder / scoreboard
4. Drivers	Where does the problem concentrate?	root-cause-pareto	Ranked drivers + sub-analysis of #1
5. Stress test	Where does the standard answer break?	the stress sections inside the routed skill	Trust statement per segment
6. Decision memo	What should the reader do?	templates/decision-memo.md	One-page memo with counter-metric

For a recurring deliverable (weekly/monthly), wrap steps 3-6 with weekly-ops-report instead of a one-off memo.

The constitution (always-on rules)

These override convenience in every step. The full versions with rationale live in rules/data-hygiene.md; the short forms:

1. Report every dropped row. Filtered, deduplicated or failed-parse rows are findings, not housekeeping.
2. Denominators and date anchors are stated, never implied. An unlabeled percentage is a rumor.
3. Never average ratios. Recompute from numerator and denominator sums; a mean of percentages is almost always wrong.
4. Benchmark against doing nothing. A forecast beats naive or it subtracts value; an initiative beats the baseline or it is theater.
5. Reconcile one headline number to raw data before presenting. Single pass, no intermediate tables; a mismatch stops the delivery.

Routing notes

• Requests mentioning on-time, OTIF, late orders, teslimat → otif-analysis.
• Forecast accuracy, MAPE, demand planning value → forecast-accuracy-review.
• Which SKUs matter, segmentation, rationalization → abc-xyz-segmentation.
• Buffer, safety stock, service level → safety-stock-review (route through abc-xyz-segmentation first if classes are unknown).
• Downtime, defects, complaints, "biggest cause" → root-cause-pareto.
• "Summarize for management", recurring KPI reporting → weekly-ops-report.
• Mixed or unclear → start at step 1; the brief will disambiguate.

Failure modes this method prevents

• Skipped brief → technically correct answer to the wrong question; the requester "clarifies" after you present.
• Skipped contract → silent schema drift; last month's numbers no longer reproduce.
• Skipped stress test → the recommendation works on average and fails exactly where it matters (volatile SKUs, peak weeks).
• Skipped memo → an analysis that everyone praises and nobody acts on.

Output expectation

Every engagement ends with the decision memo (or the recurring report): finding in one line, evidence with numbers, what the evidence does not say, options with trade-offs, a recommendation, and a counter-metric with a re-check date.