The short answer
The AI Scorecard takes three of the four Product Alpha Scorecard axes, Efficiency, Margin and Moat, and applies them to the AI execution of a portfolio or a portfolio company. The AI Playbook details the formulas, thresholds, audit and steering tools, and a first hundred days sequence. It answers a simple question: does an asset's AI protect the multiple, or quietly destroy it?
Key takeaways
- Three axes: Efficiency (ARR per R&D head), Margin (selling price against inference cost), Moat (fine-tuning, semantic cache, faithfulness).
- Margin threshold observed on the case: 4x is critical, 8x is healthy.
- The weekend test: could a competitor with the same API key rebuild the product? If yes, the moat does not exist yet.
- One hundred days can move two of three axes into the healthy zone, not three: say so in the Board memo.
A field case: Day 1 versus Day 100
A ConTech group (PE-backed) built through five acquisitions in three years, backed by a European fund, roughly €18M in ARR, twelve months from a planned exit. The management deck had an AI slide. Nobody could say whether it protected the multiple or quietly ate it.
| Axis | Day 1 | Day 100 |
|---|---|---|
| Efficiency (ARR per R&D head) | €466,667: watch list | €506,667: healthy, crosses the €500,000 threshold |
| Margin (selling price against inference cost) | 4x (€0.40 against €0.10): critical | 8x (€0.40 against €0.05): healthy. SaaS margin on the line from 75% to 85% |
| Moat | RAG on a public LLM, no fine-tuning nor semantic cache: critical | Fine-tuning documented, semantic cache live but young: watch list |
The 100 day sequence
- Weeks 1 to 2: baseline audit, interviews before numbers.
- Weeks 3 to 4: quick wins, temporary cap on the most expensive AI usage, inference cost line added to financial reporting.
- Weeks 5 to 10: smaller model for simple requests, semantic cache for the 30% of near-duplicate queries, fine-tuning on two years of support correction data.
- In parallel: three support roles freed by the automation redeployed to onboarding, no layoffs.
Two axes crossed into healthy, one moved from critical to watch and stayed there. No promise of everything turning green in one cycle: what matters is being able to say which number moved, by how much, and what is still to do.
What the Playbook contains
- Its own three-dimension derivative of the Product Alpha Scorecard.
- Its detailed first hundred days.
- Its concrete audit and steering tools.
