Bmc Ai Assisted Draft Starting

---
name: bmc-ai-assisted-draft-starting
description: "Starts a BMC session from an AI-generated first-draft canvas and brand audit instead of a blank page — the team's real work becomes editing, rejecting, and enriching the draft rather than generating it from nothing, which is faster and surfaces disagreement earlier."
---

# BMC AI-Assisted Draft Starting

## Purpose

A blank canvas is slow to fill and tends to produce vague first drafts,
because a team spends its first real energy just generating plausible-
sounding content rather than debating it. This skill inverts the
order: feed the target company's own public material to an AI model
before the session, get a fast first-draft canvas and brand read, and
have the team's actual work be editing that draft — accepting,
rejecting, correcting — rather than generating one from scratch. Use
this as a pre-session preparation step, immediately before
`bmc-session-facilitation-design`.

## Anchored in research

The source practitioner, identified independently: an executive advisor
on strategy, innovation, and communication, a certified Business Model
Generation (BMG) facilitator/coach, a contributor to *Value Proposition
Design* (the well-known Osterwalder/Strategyzer book), and a workshop
leader for Strategyzer's own "Accelerating Innovation with AI" program
— i.e. a real, Strategyzer-affiliated practitioner, not an unverifiable
name. The specific "start from edit" phrase and exact workflow
described here could not be independently located in public material
beyond the owner's own account — if the owner attended or watched this
specific workshop personally, that's a reasonable first-hand source; if
secondhand, describe it more cautiously as a technique demonstrated in
a Strategyzer AI workshop rather than a verified direct quote. Either
way, the underlying mechanism (AI-generated first draft, human edits
rather than generates) is sound practice independent of the specific
attribution.

## Method

1. **Gather public material on the target company before the session** —
   its website, any public product pages, visible pricing, marketing
   copy, and (if available) recent public statements or press.
2. **Generate a fast first-draft brand audit and BMC sketch from this
   material.** Ask explicitly for: a short brand-positioning read
   (what the company seems to claim about itself, in its own words vs.
   what actually seems differentiated), and a draft canvas across all
   nine blocks, each entry marked with a rough confidence level (clearly
   stated in the public material vs. inferred vs. genuinely uncertain).
3. **Bring the draft to the session as a starting point, not a
   deliverable.** State to the team explicitly, before anything else:
   this is a fast, AI-generated first pass built only from public
   material — it will contain errors, blind spots, and things the team
   knows are wrong. Its only job is to save the team from a blank page.
4. **Run the session as an edit pass, not a generation pass.** For each
   block, the team's first move is to accept, reject, or correct what's
   already there — not to start fresh. This surfaces disagreement
   faster than blank-page brainstorming does, because reacting to a
   concrete (even wrong) claim is easier than generating one from
   nothing, and a wrong AI guess often provokes a sharper, faster
   correction than an open question would.
5. **Pay closest attention to confident-sounding AI content that's
   actually wrong** — the biggest risk of this method isn't obviously
   uncertain content (the team will naturally scrutinize that), it's
   plausible-sounding, confidently-stated content that's subtly
   incorrect and gets accepted without real scrutiny because it reads
   well. Explicitly ask the team, block by block: "if this were
   actually wrong, would we have caught it?"
6. **Treat low-confidence-marked entries as session priorities**, not
   afterthoughts — these are exactly the blocks where the team's real,
   non-public knowledge adds the most value over what AI could infer
   from public material alone.
7. **Don't skip `bmc-canvas-diagnostic-reading` at the end** just
   because the canvas started from a draft — an AI-originated starting
   point doesn't reduce the need for the pack's normal quality checks;
   if anything, the DR-05 "defensive canvas" signal (a team accepting
   everything too readily) deserves extra attention here, since editing
   an existing draft can create a false sense that the content has
   already been vetted.

## What this skill does NOT do

- Doesn't replace `bmc-session-facilitation-design`'s session structure
  — it's a preparation step that changes what the team starts editing,
  not a replacement for how the session itself is run.
- Doesn't work well without genuine public material to draw from — a
  company with almost no public footprint (early-stage, pre-launch, or
  deliberately private) won't produce a useful draft this way; use the
  normal blank-canvas approach instead.
- Doesn't remove the need for real customer/market evidence — an
  AI-generated draft is built from public material only, which is a
  different and much weaker evidence source than the validated,
  evidence-graded canvas `bmc-canvas-diagnostic-reading` expects by the
  end of the process.

## Refinement notes

- How much does this actually speed up a real session, in your
  experience — is it the first 20 minutes, or does it change the whole
  session's pace?
- What's the clearest example of a confidently-wrong AI draft entry
  that a team almost accepted without scrutiny?
- Does this work as well for a company you (the facilitator) already
  know well, or is its real value specifically for an unfamiliar
  target?

## Continue from here

- Use immediately before: `bmc-session-facilitation-design/SKILL.md`.
- Still required afterward: `bmc-canvas-diagnostic-reading/SKILL.md` —
  an AI-originated starting draft doesn't reduce the need for the
  pack's normal quality checks.
- Related in another pack: `../../../../ai-strategy-and-governance/skills/workshop-to-agent-productization/SKILL.md`
  — a different application of AI-assisted content reuse, for
  productizing existing material rather than drafting new analysis.
- This pack's shared guardrails: `../../CLAUDE.md`

## References

- `../../references/bmc-source-material-notes.md` — source material background
- `../../../../ai-strategy-and-governance/references/ai-native-reshuffle-heuristics-research.md` —
  selection and grounding notes for this skill and its siblings
- `../../CLAUDE.md` — this pack's shared guardrails