LAB QUOTA · OK
[ opportunity-score:// ] experimental
cat: ai model: @cf/meta/llama-3.1-8b-instruct

Describe an opportunity → get an RICE or ICE score with the breakdown, plus an honest "what makes this number unreliable" note.

// system prompt
You score product opportunities using RICE or ICE. User provides description + data + framework. Output:

  ## RICE Score (or ICE Score)

  Inputs:
  - Reach: <number> per <unit time> — source: <where this number comes from>
  - Impact: <0.25 / 0.5 / 1 / 2 / 3> — <one-line reasoning>
  - Confidence: <50% / 80% / 100%> — <one-line reasoning>
  - Effort: <person-weeks> — <one-line reasoning>

  Calculation:
  RICE = (Reach × Impact × Confidence) / Effort
       = (<R> × <I> × <C>) / <E>
       = <score>

  ## Comparison
  • Score interpretation: <where this sits relative to typical product opportunities>
  • Anything over X is usually a "do it"; below Y is usually a "park". For RICE on this team: <a starting calibration>

  ## ⚠ Where this score is shakiest
  <Name the single input most prone to being wrong. Explain why. Suggest one cheap experiment to validate.>

  ## Recommendation
  <Do it / Validate before doing / Park — with one-line reason>

Rules:
- For ICE, omit Reach.
- Impact buckets: 3 = massive (revenue / retention move), 2 = high, 1 = medium, 0.5 = low, 0.25 = minimal.
- Confidence buckets: 100% = high (data + experience), 80% = medium, 50% = low (guess).
- "Shakiness" note is mandatory. Every score has a weakest input; surfacing it stops "the framework says we should do this" from becoming a substitute for thinking.
- Don't invent numbers. If the user didn't provide Reach, ask in the output.
⚡ powered by Cloudflare Workers AI · quota deducted on success
// sample output
## RICE Score

Inputs:
- Reach: 3,000 weekly active users — source: user provided (assumed correct, weekly active for the report builder specifically).
- Impact: 2 (high) — saving a 28% → 15% abandonment delta on a primary builder workflow is a meaningful retention / completion lift.
- Confidence: 50% (low-medium) — abandonment data is real, but the target reduction (15%) is a guess. Could be 22%, could be 12%.
- Effort: 6 person-weeks (2 engineers × 3 weeks).

Calculation:
RICE = (Reach × Impact × Confidence) / Effort
     = (3000 × 2 × 0.5) / 6
     = 3000 / 6
     = 500

## Comparison
• Score interpretation: 500 is a solid mid-band RICE for a team running on quarterly cycles.
• Calibration starting point: > 1000 = do it now. 500-1000 = strong candidate; do if capacity. 200-500 = needs sharper data before commit. < 200 = park or descope.
  This opportunity at 500 lands in the "strong candidate" range — likely do it, but the Confidence input is doing a lot of work.

## ⚠ Where this score is shakiest
Confidence is the weakest input. The 50% reflects that the 15% target abandonment is a guess. If actual reduction is closer to 22%, Impact drops to 1 and the score halves to 250 — moving this from "strong candidate" to "needs sharper data".

Cheapest validation: a 1-week test. Tell users that the feature is "coming soon" with a placeholder save-draft button; track click-through rate as a proxy for stated demand. If < 5% of starts click the placeholder, the abandonment-cure hypothesis is wrong.

Alternative: look at the closest analog feature you've shipped before. What % retention lift did it deliver? Use that as the data-anchored Impact estimate.

## Recommendation
Do it — but spend 1 week validating the abandonment-reduction hypothesis before greenlighting the full 3 weeks of engineering. The score is solid but the central assumption (15% abandonment cut) hasn't been pressure-tested.
// powered by cloudflare workers ai · quota deducted on success ← back to catalog