Back to course overview
Module 5Capstone 13 min

Capstone: The story & the fix

Build the readout package and the process recommendation: charts with claims for titles, the one-pager, Q&A prep, and the to-be map with AI seats assigned.

Findings become influence in this phase. Two deliverables: the readout package (how the organization hears the answer) and the process recommendation (how the fix gets installed and stays installed). Together they're what distinguishes an analyst from a report generator.

The readout package

  1. 1Spec and build 2-3 evidence charts. Titles are claims; the interesting series highlighted; axes honest. Run each through AI's what-will-be-misread critique.
  2. 2Write the one-pager: answer → beats → ask → confidence → appendix pointers. Then the 10-second test on a colleague who knows nothing. Rewrite until it passes; it usually takes two rounds.
  3. 3Run the persona Q&A with your real personas — your sponsor, your skeptic, and whoever owns the thing your finding implicates (there's almost always one; find them before the meeting does).
  4. 4Prepare the two re-framings of your answer sentence for the two most senior readers. Submit both; the delta between them is part of what's graded.

The process recommendation

  1. 1Map the process your finding lives in — swimlane, [UNKNOWN]s marked, annotated with your data. Walk it with one person who works inside it; note what the walk corrected (that correction is evidence the map is real).
  2. 2Score the relevant steps (frequency / rules / error-cost) and assign AI seats — including at least one step you deliberately keep human, defended in a sentence.
  3. 3Write the to-be page: fixes sequenced cheap-and-reversible first, each with owner and success metric. If your finding was inconclusive, the recommendation is the investigation to run next — scoped, owned, and sized.
The recommendation robustness trick

The strongest capstone recommendations survive their own analysis being wrong: 'even if the mailers aren't the cause, the flag audit and mandatory reason tags are cheap and catch the next spike early.' Where you can sequence your ask so step one pays off under multiple hypotheses, do it — decision-makers say yes to robust asks dramatically faster than to bets on your theory.