name

ai-opportunity-radar

description

Auto-generate the two "use the BCM" portfolio visuals from note 1 — an AI Opportunity Radar (AI use cases plotted on Value × Feasibility, coloured by Defend/Extend/Upend, pinned to BCM capabilities, zoned into Likely wins / Calculated risks / Marginal gains) and a 7-dimension AI Readiness gap chart (current vs future on Strategy / Value / Organization / People & Culture / Governance / AI Engineering / AI Data). Use this AFTER a BCM exists (it's the downstream step the [[bcm-creator]] map feeds), or whenever the user wants to prioritise AI use cases, map a use-case portfolio, score AI readiness, see the value/feasibility quadrant, plan where AI investment should go, or turn a capability map into an investment radar. Triggers: "AI opportunity radar", "prioritise use cases", "value feasibility quadrant", "AI readiness assessment", "put use cases on the BCM", "Defend/Extend/Upend", "draw a readiness radar". Renders precise data charts as self-contained HTML+SVG (deterministic), not raster.

ai-opportunity-radar

The downstream half of the BCM method: once you have a capability map ([[bcm-creator]]), this turns it into an investment radar and a readiness gap chart — the two visuals note 1 uses to decide where AI goes and whether you're ready. Full method context: ~/.claude/skills/bcm-creator/references/bcm-method-govern.md (bundled with the bcm-creator skill).

The two outputs

AI Opportunity Radar — every AI use case plotted on Value (y) × Feasibility (x), each dot coloured by Defend / Extend / Upend and pinned to a BCM capability. Zones: Likely wins (hi value + hi feasibility → do now) / Calculated risks (hi value, low feasibility → small bets) / Marginal gains (low value → don't distract). Rendered by scripts/render_radar.py.
AI Readiness gap chart — a 7-axis spider, current vs future, with a gap table. The point is the gap, not the score. Rendered by scripts/render_readiness.py.

Format = self-contained HTML + inline SVG (deterministic), NOT a codex raster. A radar is a scatter at exact (value, feasibility) coordinates and a readiness chart is a spider at exact scores — gpt-image-2 cannot place data points accurately, so a raster would be decorative-but-wrong. This is the same "precise data artifact" lane as the bcm-creator brief HTML, not the illustrative-map lane. (If the user explicitly wants a slide-style raster of these, that's a separate, lossy request — say so.)

Phase 0 — clarify (the scoring wall)

Most of the input is a strategic/internal judgement, not web-researchable (the data wall from [[bcm-creator]]): which use cases the org is weighing, their value/feasibility, the readiness scores. Pin via AskUserQuestion (skip what's already given):

Which BCM / company — and is there a bcm-creator brief to pull capabilities from?
Use cases — does the user supply them (+ rough value/feasibility), or should I propose candidates per capability from the BCM + industry-typical AI use cases?
Who scores — the user/org gives value/feasibility & readiness, or I put down estimates marked TBD for them to adjust? (Never present estimates as the org's real scores.)

Phase 1 — assemble the portfolio (apply the assessment lenses)

Build one portfolio.json:

use_cases[]: {name, capability (a BCM L1 box), value 1-5, feasibility 1-5, type: defend|extend|upend}. When proposing/sizing each, apply note 1's lenses (documented in bcm-method-govern.md):
- value = contribution to a business objective; feasibility = technical maturity + internal readiness.
- Defend / Extend / Upend tag — augment-with-stable-ROI / process-transformation-with-packaged-solutions / frontier-high-uncertainty.
- sanity-check with the 6 agent dimensions (Perception / Decisioning / Actioning / Agency / Adaptability / Knowledge — most need 2–3) and agency caution: a use case demanding end-to-end Level-1 autonomy is a Calculated risk at best (one bad auto-decision propagates), so its feasibility is low. Also note which techniques + data sources + integration it needs — thin integration ⇒ lower feasibility.
readiness[]: the 7 dims {dim, current 1-5, future 1-5}. Judge the gap; data readiness is contextual (don't require 100% clean).
notes[]: flag every estimate (TBD), the scoring wall, and the type-inference basis. Worked example: references/example-portfolio.json.

Phase 2 — render

python3 ~/.claude/skills/ai-opportunity-radar/scripts/render_radar.py /path/portfolio.json /path/radar.html
python3 ~/.claude/skills/ai-opportunity-radar/scripts/render_readiness.py /path/portfolio.json /path/readiness.html
open /path/radar.html /path/readiness.html

Sanity-check: radar dot count = use-case count; readiness axes = 7 (or however many dims given). Both are pure-stdlib (no matplotlib/numpy needed).

Gotchas

Don't present estimated scores as the org's real numbers. value/feasibility/readiness are internal strategic judgements; pre-fill as TBD estimates and say so. A radar full of made-up scores is worse than an empty one.
The radar is a prioritisation aid, not truth — it sorts the conversation, it doesn't make the decision. Two people can score the same use case differently; that disagreement IS the useful output.
Pin use cases to a capability, not to themselves — same as the BCM rule; a use case with no capability to pin to is a smell (it's probably an outcome or a tool).
Readiness: gap, not score — a high current score with no future ambition isn't readiness, it's complacency; a big gap on Strategy/Governance (the oft-ignored dims) is the real blocker.
HTML+SVG, not raster (see above) — these are data charts; deterministic render is the correct tool, not a faked picture.
This skill is the use step; making the map itself is [[bcm-creator]], and the connecting method is ~/.claude/skills/bcm-creator/references/bcm-method-govern.md.

ai-opportunity-radar

The two outputs

AI Opportunity Radar — every AI use case plotted on Value (y) × Feasibility (x), each dot coloured by Defend / Extend / Upend and pinned to a BCM capability. Zones: Likely wins (hi value + hi feasibility → do now) / Calculated risks (hi value, low feasibility → small bets) / Marginal gains (low value → don't distract). Rendered by scripts/render_radar.py.

AI Readiness gap chart — a 7-axis spider, current vs future, with a gap table. The point is the gap, not the score. Rendered by scripts/render_readiness.py.

Phase 0 — clarify (the scoring wall)

Which BCM / company — and is there a bcm-creator brief to pull capabilities from?

Use cases — does the user supply them (+ rough value/feasibility), or should I propose candidates per capability from the BCM + industry-typical AI use cases?

Who scores — the user/org gives value/feasibility & readiness, or I put down estimates marked TBD for them to adjust? (Never present estimates as the org's real scores.)

Phase 1 — assemble the portfolio (apply the assessment lenses)

Build one portfolio.json:

use_cases[]: {name, capability (a BCM L1 box), value 1-5, feasibility 1-5, type: defend|extend|upend}. When proposing/sizing each, apply note 1's lenses (documented in bcm-method-govern.md):

value = contribution to a business objective; feasibility = technical maturity + internal readiness.
Defend / Extend / Upend tag — augment-with-stable-ROI / process-transformation-with-packaged-solutions / frontier-high-uncertainty.
sanity-check with the 6 agent dimensions (Perception / Decisioning / Actioning / Agency / Adaptability / Knowledge — most need 2–3) and agency caution: a use case demanding end-to-end Level-1 autonomy is a Calculated risk at best (one bad auto-decision propagates), so its feasibility is low. Also note which techniques + data sources + integration it needs — thin integration ⇒ lower feasibility.

readiness[]: the 7 dims {dim, current 1-5, future 1-5}. Judge the gap; data readiness is contextual (don't require 100% clean).

notes[]: flag every estimate (TBD), the scoring wall, and the type-inference basis. Worked example: references/example-portfolio.json.

Phase 2 — render

python3 ~/.claude/skills/ai-opportunity-radar/scripts/render_radar.py /path/portfolio.json /path/radar.html python3 ~/.claude/skills/ai-opportunity-radar/scripts/render_readiness.py /path/portfolio.json /path/readiness.html open /path/radar.html /path/readiness.html

Sanity-check: radar dot count = use-case count; readiness axes = 7 (or however many dims given). Both are pure-stdlib (no matplotlib/numpy needed).

Gotchas

Don't present estimated scores as the org's real numbers. value/feasibility/readiness are internal strategic judgements; pre-fill as TBD estimates and say so. A radar full of made-up scores is worse than an empty one.

The radar is a prioritisation aid, not truth — it sorts the conversation, it doesn't make the decision. Two people can score the same use case differently; that disagreement IS the useful output.

Pin use cases to a capability, not to themselves — same as the BCM rule; a use case with no capability to pin to is a smell (it's probably an outcome or a tool).

Readiness: gap, not score — a high current score with no future ambition isn't readiness, it's complacency; a big gap on Strategy/Governance (the oft-ignored dims) is the real blocker.

HTML+SVG, not raster (see above) — these are data charts; deterministic render is the correct tool, not a faked picture.

This skill is the use step; making the map itself is [[bcm-creator]], and the connecting method is ~/.claude/skills/bcm-creator/references/bcm-method-govern.md.