bmc-sensemaking-question-mapping

Builds the BMC as a structured set of open, falsifiable questions per block instead of filled-in answers, and distinguishes 'we lack data' gaps from 'we haven't interpreted contradictory signals' gaps

SKILL.md

GitHub

BMC Sensemaking Question Mapping

Purpose

When raw information is cheap and abundant — increasingly true as AI makes lookup and summarization nearly free — a canvas built purely from answers a well-prompted AI assistant could produce from public information is low-value: it hasn't required any team judgment. The scarce, valuable work has shifted from COLLECTING information to INTERPRETING it and asking the questions that actually move a decision forward. This skill uses the BMC as a question-generating map rather than an answer-collecting template, applied early — before or during the first canvas draft, not after the canvas already looks finished.

Anchored in research

Grounded in Karl Weick's organizational sensemaking theory — the established, decades-old body of work on how organizations interpret ambiguous information, as distinct from simply gathering more of it. Applied here to the current, active 2026 discourse on AI's effect on this distinction: as AI increasingly commoditizes "sensing" (retrieving and summarizing information), the differentiating leadership and strategy skill shifts toward "sensemaking" (interpreting what information actually means, and knowing which questions to ask in the first place) — a theme covered independently across multiple 2026 sources on sensemaking as a leadership skill and on human-AI sensemaking specifically.

Method

  1. Rewrite each of the nine blocks as an open question with an explicit "what would change our mind" clause, instead of a declarative answer. Not "Our customer segment is small businesses with 10-50 employees," but "Is the segment actually small businesses with 10-50 employees, or is that a comfortable assumption? What evidence would tell us we're wrong — a specific number of failed sales conversations, a specific competitor winning that segment instead?" Do this for every block that isn't already backed by tested evidence.
  2. Distinguish two different kinds of blank, because they need different fixes:
    • A sensing gap — the team simply lacks data. The fix is to go get more information (a search, a report, a data pull). If the honest answer to "why don't we know this?" is "we haven't looked," it's a sensing gap.
    • A sensemaking gap — the team already has information but hasn't interpreted what it means. The fix is facilitated interpretation, not more data collection. If the honest answer is "we have three contradictory signals and don't know which to believe," more data will not resolve this — the team needs to reason through the contradiction directly. Misdiagnosing a sensemaking gap as a sensing gap is the most common failure mode this skill exists to catch: teams keep "researching" something they already have enough information about, because interpreting it is uncomfortable and gathering more data feels like progress.
  3. Apply the "could an AI have produced this" test to each filled-in block. Ask: could a well-prompted AI assistant have written this block's content from public information alone, with no team judgment involved? If yes, that block hasn't earned its place on a strategy artifact yet — it's sensing-level content, useful as raw material but not yet a team's actual position. Push it toward genuine interpretation: what does THIS team, with THIS specific context and judgment, believe this information means for the business?
  4. Prioritize which questions to resolve first by the cost of getting them wrong, not by ease of answering. Before running any experiment or research effort, ask: if we get the wrong answer to this specific question, what does it cost us — time, money, opportunity, credibility? Sequence toward resolving the highest-cost-of-being-wrong questions first. This is the same logic as this pack's "Clueless Corner" hypothesis prioritization (see bmc-tool-switching-decisions's hypothesis quality decision) — applied here one step earlier, to which QUESTIONS get asked at all, not just which hypotheses get tested once they already exist.
  5. Use this early and revisit it, don't run it once and move on. By the time bmc-canvas-diagnostic-reading runs its evidence grade check (DR-04), the canvas should already be past the raw-sensing stage for its most important blocks — this skill is what gets it there. Re-run the "could an AI have produced this" test whenever new information arrives, since a block that was genuinely team-interpreted last week can quietly slide back into generic-answer territory if it isn't revisited.

What this skill does NOT do

  • Doesn't tell the team what the right answer is — it only distinguishes which blanks need more information and which need interpretation, and forces the interpretation to actually happen for the ones that need it.
  • Doesn't replace direct customer research or the other data-gathering skills in this pack — sensing gaps still need real information; this skill's job is making sure teams don't mistake a sensemaking gap for one.
  • Doesn't have a numeric scoring rubric the way bmc-canvas-diagnostic-reading does — this is a qualitative reframing technique, not a scored diagnostic; use it as a lens applied throughout the session rather than a one-time checklist.

Refinement notes

  • What's the clearest real example you've seen of a team mistaking a sensemaking gap for a sensing gap — endlessly "researching" something they actually needed to just decide on?
  • How do you personally run the "could an AI have produced this" test with a client without it feeling like an accusation that their work is shallow?
  • Is there a cleaner way you've found to sequence which questions get asked first (Step 4) than pure cost-of-being-wrong?

Continue from here

  • Use early: alongside the first canvas draft, before bmc-canvas-diagnostic-reading's evidence grade check (DR-04).
  • Related: bmc-tool-switching-decisions/SKILL.md's "Clueless Corner" hypothesis prioritization — the same cost-of-being-wrong logic, one step later in the process.
  • This pack's shared guardrails: ../../CLAUDE.md

References

  • ../../references/bmc-source-material-notes.md — source material background
  • ../../references/bmc-resilience-heuristics-research.md — selection and grounding notes for this skill and its siblings
  • ../../CLAUDE.md — this pack's shared guardrails