Map

AI Mapping Audit — Full Workspace Scan

Run a comprehensive AI Mapping Audit based on the research paper "Mapping AI into Production: A Field Experiment on Firm Performance" (Kim, Kim & Koning, 2026).

Background for the user

Tell the user:

This audit is based on a field experiment with 515 startups that found the #1 bottleneck to getting value from AI isn't capability — it's discovering where to deploy it. Firms that were shown how others reorganized around AI found 44% more use cases, completed 12% more tasks, were 18% more likely to acquire paying customers, and generated 1.9x higher revenue — all while needing 39.5% less external capital.

I'm going to scan your workspace, map where you're already using AI, and identify the high-value gaps where AI could be creating value but isn't.

Step 1: Discovery Questions

Before scanning any files, ask the user these questions to understand their business. These mirror the intervention from the research paper — the act of thinking through these questions is itself valuable.

Ask these one at a time (not all at once):

1. "Walk me through what your business does, step by step — from when a customer first finds you to when they pay you and get what they bought." (This maps their production chain end-to-end)

2. "Where in that process do you spend time moving information between tools — copying from a spreadsheet into an email, updating a CRM after a call, transferring data between apps?" (This finds "glue work" — the FazeShift pattern from the paper)

3. "What parts of your business do you currently outsource, or wish you could do faster or cheaper?" (This finds automation opportunities — the Ranger pattern)

4. "If you could instantly test three different versions of something in your business, what would it be?" (This finds parallel prototyping opportunities — the RyzLabs pattern)

5. "What decisions do you currently make based on gut feeling that you wish you had data for?" (This finds strategy and analytics gaps — the highest-value functions per the research)

Record the answers — they will inform the gap analysis in later steps.

Step 2: Venture Discovery

Use the Agent tool with subagent_type "Explore" to thoroughly scan the user's working directory and identify:

• All projects, businesses, and ventures
• Their current status (active, planned, dormant)
• Any existing AI/automation configurations
• Revenue-generating vs. developmental projects

Be thorough — check subdirectories, read READMEs, check for config files, package.json files, .env files, GitHub Actions, automation scripts, etc.

Step 3: AI Usage Mapping

For each discovered venture, map current AI usage across these 10 production functions:

For each function, assign one of these statuses:

• Saturated — AI is deeply integrated, diminishing returns from more
• Active — AI is used but could be expanded
• Minimal — Some AI usage but mostly manual
• Absent — No AI involvement at all

Step 4: Gap Analysis

Read these two reference files:

• ${CLAUDE_PLUGIN_ROOT}/skills/ai-mapping-audit/references/examples-from-other-firms.md — examples of how firms deploy AI in each function
• ${CLAUDE_PLUGIN_ROOT}/skills/ai-mapping-audit/references/case-studies-from-paper.md — the 4 real case studies from the Harvard/INSEAD research (Gamma, RyzLabs, FazeShift, Ranger)

For each function marked "Minimal" or "Absent", generate specific recommendations:

• What Claude/AI could do in this function
• How other firms have deployed AI here (from the examples and case studies)
• Which case study pattern applies (process redesign like Gamma? Parallel prototyping like RyzLabs? Eliminating glue work like FazeShift? Services-first like Ranger?)
• Expected impact (revenue, time saved, capital reduction)
• Implementation complexity (Low/Medium/High)

O-Ring Warning

Check for the "partial automation trap": if AI is used in only one step of a multi-step process while the rest stays manual, flag it prominently:

Warning: Partial automation detected. AI is being used in [function] but the steps before and after it are still manual. Research shows that automating just one step in a chain preserves the bottleneck rather than relieving it. The biggest gains come from rethinking the entire process around AI — not just speeding up one piece.

Cross-reference with the user's answers from Step 1, especially their description of their step-by-step process and where they identified glue work.

Step 5: Priority Scoring

Score each opportunity on three dimensions (1-5 scale):

• Revenue Impact: How much could this increase revenue or reduce costs?
• Capital Reduction: Does this reduce need for external investment or hiring?
• Ease of Implementation: How quickly can this be deployed with existing tools?

Calculate a composite score: (Revenue Impact x 2) + Capital Reduction + Ease of Implementation

Sort opportunities by composite score, highest first.

Step 6: Generate the AI Map (Markdown + HTML)

Create TWO files in the user's working directory:

File 1: AI-MAP-[date].md (text-readable report)

Markdown file with:

1. Executive Summary — Total ventures scanned, current AI saturation %, number of gaps found, number of "press-send" wins
2. Venture Inventory — List of all discovered ventures with status
3. The AI Map — A table showing each venture × each function with status indicators (Saturated/Active/Minimal/Absent)
4. Top 10 Opportunities — Prioritized list with scores and specific implementation steps
5. Press-Send Wins — The subset of opportunities where infrastructure already exists (Ease = 5, Score ≥ 15)
6. The Paper's Key Insight — Remind the user that gains are largest at the 90th percentile (AI amplifies winners, doesn't rescue losers)
7. 30-Day Action Plan — Week-by-week implementation sequence starting with highest-impact items

File 2: AI-MAP-[date].html (full visual report)

Generate a self-contained, dark-theme HTML report. Use the template at:

${CLAUDE_PLUGIN_ROOT}/skills/ai-mapping-audit/templates/heat-map-template.html

The template includes a donut chart (SVG), color-coded stat cards, per-venture progress bars, a CSS-grid heat map, a radar/spider chart, a horizontal bar chart for the top 10, a vertical timeline for the 30-day plan, icon callouts, and a research footer. All visuals are pure HTML+CSS+SVG — no external assets, no JavaScript.

Step 6a — Calculate the donut chart

After classifying every (venture × function) cell, count each status across all cells.

total       = ventureCount × 10
sat_count   = cells marked Saturated
act_count   = cells marked Active
min_count   = cells marked Minimal
abs_count   = cells marked Absent

SAT_PCT = round(sat_count / total × 100)
ACT_PCT = round(act_count / total × 100)
MIN_PCT = round(min_count / total × 100)
ABS_PCT = round(abs_count / total × 100)

SATURATION_PERCENT = round((sat_count + act_count) / total × 100)

The donut uses a circle with r=70, so circumference = 2π × 70 ≈ 440. Each segment's length is its share of 440. Stack the segments by adding the previous lengths to the next segment's dashoffset:

DONUT_SAT_LEN    = (sat_count / total) × 440
DONUT_ACT_LEN    = (act_count / total) × 440
DONUT_MIN_LEN    = (min_count / total) × 440
DONUT_ABS_LEN    = (abs_count / total) × 440

DONUT_SAT_OFFSET = 0
DONUT_ACT_OFFSET = -DONUT_SAT_LEN
DONUT_MIN_OFFSET = -(DONUT_SAT_LEN + DONUT_ACT_LEN)
DONUT_ABS_OFFSET = -(DONUT_SAT_LEN + DONUT_ACT_LEN + DONUT_MIN_LEN)

Round all dasharray/offset values to 2 decimal places when substituting.

Step 6b — Build per-venture progress bars

For each venture, compute its own saturation:

venture_score = (sat_in_venture × 1.0) + (act_in_venture × 0.6) + (min_in_venture × 0.25)
venture_pct   = round(venture_score / 10 × 100)
functions_active = sat_in_venture + act_in_venture

Pick the bar color class based on venture_pct:

• 0–15% → r (red)
• 16–30% → y (yellow)
• 31–60% → b (blue)
• 61%+ → g (green)

Emit one card per venture into {{VENTURE_BARS}}:

<div class="venture-card">
  <div class="vc-name">VentureName</div>
  <div class="vc-meta">FUNCTIONS_ACTIVE / 10 functions covered</div>
  <div class="vc-bar-wrap"><div class="vc-bar b" style="width:42%"></div></div>
  <div class="vc-pct"><span>AI Saturation</span><strong>42%</strong></div>
</div>

Step 6c — Build the heat map grid

The heat map is functions × ventures (10 rows, N columns where N = venture count). Replace {{HM_COLS}} in the CSS grid with the venture count.

Replace {{HEATMAP_VENTURE_HEADERS}} with one <div class="hm-vh">VentureName</div> per venture, in the same column order you'll use for the cells.

Replace {{HEATMAP_ROWS}} with 10 rows in this exact function order: Product Dev, Strategy, Marketing, Sales, Customer Success, Finance, Operations, Content, Data, System Admin.

<div class="hm-row">
  <div class="hm-fn-label">Product Dev</div>
  <div class="hm-cell hm-saturated">SATURATED</div>
  <div class="hm-cell hm-active">ACTIVE</div>
  <div class="hm-cell hm-absent">ABSENT<span class="star">★</span></div>
  <!-- one cell per venture, in the same order as the headers -->
</div>

Cell class + label mapping:

• Saturated → hm-saturated / "SATURATED"
• Active → hm-active / "ACTIVE"
• Minimal → hm-minimal / "MINIMAL"
• Absent → hm-absent / "ABSENT"

Add <span class="star">★</span> inside any cell whose (venture × function) intersection appears in the top-10 priority list with composite score ≥ 12.

Step 6d — Calculate the radar/spider chart polygon points

The radar has 10 axes (one per function), with the first axis pointing straight up. Score every function on a 0–1 scale based on the BEST status that function reaches across all ventures (or the average — pick one and be consistent):

Absent    → 0.0
Minimal   → 0.25
Active    → 0.6
Saturated → 0.9

For each function i (0–9), compute the radar point:

angle_deg = (i × 36) − 90        # -90 so first spoke points up
angle_rad = angle_deg × π / 180
x = 250 + (score × 180) × cos(angle_rad)
y = 250 + (score × 180) × sin(angle_rad)

Function index order (must match the spoke positions in the SVG): 0=Product Dev, 1=Strategy, 2=Marketing, 3=Sales, 4=Customer Success, 5=Finance, 6=Operations, 7=Content, 8=Data, 9=System Admin.

Concatenate all 10 points into a points string:

"x0,y0 x1,y1 x2,y2 ... x9,y9"

Substitute into {{RADAR_CUR_POINTS}}.

For {{RADAR_TOP_POINTS}} (the 90th-percentile reference polygon), use score = 0.85 for every function — this represents the "top performers" from the research. The polygon will be a near-perfect circle, making the gap between blue (current) and green (top performers) immediately visible.

Step 6e — Build the horizontal bar chart for top 10 opportunities

For each opportunity in the top 10, compute the bar width:

bar_width_pct = (composite_score / 20) × 100

Apply the top class (red gradient) to the top 2 highest-scoring opportunities — these are the visual "press send" callouts.

Emit into {{OPPORTUNITY_BARS}}:

<div class="bar-row">
  <div class="bar-label"><span class="rk">1</span>VentureName × Function</div>
  <div class="bar-track"><div class="bar-fill top" style="width:90%"></div></div>
  <div class="bar-score">18/20</div>
</div>

(Use bar-fill without top for ranks 3–10.)

Step 6f — Build the priority detail list

{{PRIORITY_LIST}} is the same top 10 opportunities with full explanation:

<div class="priority-detail-item">
  <div class="pdi-head">
    <span class="rank">1</span>
    <span class="pdi-title">VentureName × Function: short action description</span>
    <span class="pdi-score">18/20</span>
  </div>
  <p class="pdi-body">Why this matters in plain language. Which case study pattern applies (Gamma's process redesign / RyzLabs's parallel prototyping / FazeShift's glue-work elimination / Ranger's services-first). The first concrete step to take this week.</p>
</div>

Step 6g — Build the press-send list

Filter the priority list to opportunities with composite_score ≥ 15 AND Ease = 5. Emit into {{PRESS_SEND_LIST}}:

<div class="ps-item">
  <strong>VentureName × Function:</strong> What to activate
  <small>Why it's "press send": infrastructure already exists because [specific reason]</small>
</div>

Step 6h — Group the timeline into 4 weekly themes

Group the top opportunities into 4 weekly buckets by theme:

• Week 1 — Activate (w1, red): Press-send wins. Anything with Ease=5 and infrastructure already in place.
• Week 2 — Customer (w2, orange): Customer-facing opportunities (Sales, Marketing, Customer Success).
• Week 3 — Finance (w3, yellow): Finance & Pricing, Data & Analytics, Operations.
• Week 4 — Strategy (w4, green): Product Dev and Strategy & Planning — the highest-value functions per the research.

If a theme has no relevant top-10 opportunities, fill it with the next-highest-scoring opportunity that fits.

Emit into {{TIMELINE_WEEKS}}:

<div class="timeline-week w1">
  <span class="tw-tag">Week 1 · Activate</span>
  <h3 class="tw-title">Highest-impact press-send wins</h3>
  <ul class="tw-tasks">
    <li>Specific action 1 from this theme</li>
    <li>Specific action 2 from this theme</li>
  </ul>
</div>

Repeat for w2, w3, w4.

Step 6i — Substitute the remaining placeholders

Why both formats

• Markdown is easy for the user to read in their editor, version-control friendly, can be re-processed by other tools
• HTML opens directly in any browser with a styled visual heat map — much easier to share with co-founders, investors, or team members who don't read markdown
• The HTML file is self-contained (all CSS inline, no external assets) so it works offline and can be emailed as an attachment

After generating both files, tell the user:

I've generated two files in your working directory:

1. AI-MAP-[date].md — text report with the full analysis (open in your editor)
2. AI-MAP-[date].html — visual heat map (open in any browser by double-clicking it)

The heat map color-codes every venture × function intersection — green where AI is active, red where it's absent. Stars (★) mark the highest-impact gaps to fill first.

Your top opportunity is: [briefly describe #1 from the priority list]

Want me to run /the-44-percent-rule:map-score to drill into the prioritization, or /the-44-percent-rule:map-venture [name] to do a deep audit of a specific project?

Important Notes

• Focus recommendations on areas where AI is ABSENT, not where it's already active
• Be specific — don't say "use AI for analytics", say "connect Beehiiv MCP get_post_stats to a weekly Claude agent that identifies which content topics drive the most ad revenue"
• Reference the paper's findings throughout to ground recommendations in evidence
• The biggest wins come from product development and strategy — weight those functions higher
• Always check: if a venture only uses AI for writing/content, it has a mapping problem