Write the Interpretation Brief Before Any Metric Reaches an Exec

Workflow area: Measurement & ROI · Tier: Core · Hands-on time: ~30 minutes

The dangerous moment in engineering metrics is not collection; it is the forward. A VP screenshots your deployment-frequency chart into a board deck, the "productivity up 40%" reading detaches from the caveats you said out loud in the meeting, and three quarters later you are defending a trend against a baseline that was never real. The interpretation brief exists because numbers travel without their authors — so the caveats must be welded to the numbers, in writing, before the first forward.

When to use — and when not to

• Use it for every recurring metrics export that leaves engineeringmonthly DORA-style reviews, QBR packs, board-deck inputs, anything your engineering-intelligence platform emits toward executives.
• Use it especially during the AI rolloutthe rollout is itself the largest confound in every delivery metric you own this year, and the brief is where you say so before someone attributes the whole delta to the tools (or to your leadership).
• Use it when a number is about to be weaponizeda comparison between teams, a target being set on a metric that cannot bear one; the brief is your pre-emptive record of what the number can and cannot support.
• Skip it for internal team retrosthe team knows its own confounds; the brief is armor for numbers traveling to people without context, and writing it for every stand-up dashboard turns a discipline into a tax.
• Skip metrics you cannot brief honestlyif a number's confounds swamp its signal and you cannot say what it supports, the finding is that it should not be exported at all.

The recipe

1. 1.

   Fix the brief's shape — one page, four sections, every time. Consistency is what trains execs to expect interpretation with the numbers: What moved (the two or three real changes, plain language, no jargon); What explains it (confounds, each named and sized); What this supports (the strongest claim the data can honestly carry); What this does not support (the tempting over-read, pre-refused in writing). The fourth section is the one that saves you in Q3.

2. 2.

   Keep a running confound log so the brief writes itself. A dated list, maintained as things happen, not reconstructed at review time: holidays and leave seasons, reorgs and team changes, hiring and attrition, incident weeks, migration freezes, definition changes in the tooling ("deploy" recounted after the pipeline change), and every stage of the AI rollout — tool launches, training weeks, the pb-35 bake-off itself. Two minutes when it happens beats an hour of archaeology at export time; pb-37 uses this same log as measurement infrastructure.

3. 3.

   Draft with Claude, against the log.

   Copy

   Here is this period's delivery-metrics export [PASTE], last
   period's for comparison [PASTE], and our confound log for the
   window [PASTE].
   
   Draft a one-page interpretation brief, four sections: WHAT MOVED
   (plain language, top 3 changes only), WHAT EXPLAINS IT (map each
   movement to confounds from the log; where the log cannot explain
   a movement, say "unexplained — investigating", never invent a
   cause), WHAT THIS SUPPORTS (the strongest defensible claim),
   WHAT THIS DOES NOT SUPPORT (the over-read an exec might make).
   Audience: executives with no engineering-metrics background.
   No metric acronyms without a one-clause definition.
   

   The model's job is drafting and confound-mapping; the judgment calls in sections three and four are yours, and you edit them personally every time.

4. 4.

   Verify the brief against the raw data before it ships. Ten minutes, three checks: does every claim in "what moved" match the actual chart (models smooth over reversals); is every confound cited actually in the window; and is "what this supports" a claim you would defend, alone, in front of the CFO? Good output looks like:

   Copy

   WHAT MOVED: Change lead time fell ~30% (11.2d → 7.8d).
   WHAT EXPLAINS IT: Copilot rollout completed wk 2 (partial);
     two-week code freeze in the comparison period inflated the
     baseline (major); one team moved to trunk-based dev (partial).
   WHAT THIS SUPPORTS: Lead time is genuinely improving; the AI
     contribution is real but not yet separable from the freeze
     artifact. Re-assessable with one clean quarter.
   WHAT THIS DOES NOT SUPPORT: "AI made us 30% faster." Do not
     put this number in a board deck as an AI result.
   

5. 5.

   Staple the brief to the numbers — physically. The brief is page one of the export, the first slide before any chart, the top of the email above the dashboard link. Never send a raw export and promise interpretation later; the forward happens in the gap. When a number does appear detached downstream, the correction is gentle and unanswerable: "the brief that shipped with that chart covers this — here it is."

Make it yours

• Small team or startup: the brief is five sentences in the email above the dashboard link, and the confound log is a pinned note — the four-section discipline matters even at this size, because a founder forwards numbers to investors faster than any VP forwards them to a board.
• Large org or GCC: standardize the template across every team that exports metrics, so skip-level readers learn one format; roll team briefs upward — the org-level brief cites the team-level ones rather than re-deriving, and disagreements between levels are findings, not embarrassments. For GCCs, the parent company's dashboard pull is your most dangerous unbriefed export; get the brief into that pipeline first.
• Metrics-platform-heavy orgs: platforms (DX-, Jellyfish-, LinearB-, Swarmia-class) ship auto-insights that arrive interpretation-flavored but confound-blind — the brief's "what explains it" section explicitly confirms or overrides the platform's narrative, and where you override it, say so, or the platform's version wins by default in every meeting you are not in.

Quality check

• The brief ships with the numbers 100% of the timeone raw export forwarded upward undoes a quarter of discipline; audit your own outbox, not your intentions.
• Every confound is sized, not just listed"holidays (major), reorg (partial)" is interpretation; an unsized list is a disclaimer, and disclaimers get skipped.
• "What this does not support" names this quarter's actual temptationa generic caveat is filler; the section should make one specific over-read impossible to make innocently.
• Unexplained movements stay unexplaineda brief that explains everything is confabulating somewhere; "investigating" written plainly is what keeps the other sections credible.
• Execs quote the brief, not just the chartthe success signal is hearing your own caveat repeated upward without you; if numbers keep arriving in board decks stripped of interpretation, the stapling is failing and the format needs to move earlier in the pipeline.