🪴 Brain

AI 70-20-10 Rule

3 min read

AI 70-20-10 Rule

🗒️ Description

A framework from Boston Consulting Group research: real value from AI is distributed roughly as follows:

  • 10% — the algorithm itself (model, SOTA benchmarks)
  • 20% — the deployment technology (MLOps, infra, integrations)
  • 70% — redesigning people and processes (culture, skill-up, change management, middle management)

A TechRadar article (Jason Harvey, 2026-04-10) shows that CEOs discuss AI on earnings calls at record frequency (top-5 topics per CEO Radar, ~5000 calls), but only 21% of executives pass 4+ out of McKinsey’s 10 strategy tests — a 40% drop vs. a decade ago. In other words: a lot of noise, weak execution.

The takeaway: AI is not an IT project — it’s a new chapter in managing human capital.

🔗 Links

🧩 Key claims

  • AI as a board-level priority, not “yet another IT project”.
  • Middle managers as execution lever — 88% of managers role-model AI use in daily work (per the BCG study cited).
  • Innovation culture + training + change management matter more than the stack itself.
  • Strategy without execution = hallucination (especially in the AI era, where it’s easy to mistake a demo for a product).
  • Competitive advantage > cost cutting — AI delivers the most value when it changes how customers are served, not just trims costs.

🧩 Case study — wood manufacturer (80 years in business)

  • Problem: old visual scanners → false positives, waste, hours of manual inspection, 20,000 pieces/day.
  • Partners: global machine-vision leader + local partner for production testing.
  • Solution: machine vision with deep learning + 3D sensors, continuous learning.
  • Results: 8× precision, £500,000+ annual savings, export to new markets, better supply-chain visibility, addressed the shortage of qualified workers (new team members only annotate images).

☘️ Impact on my work

  • At PLSoft / consulting: instead of selling an “AI feature”, sell a process redesign — that’s where the 70% of ROI is. Ties into Autonomous Sales Agent Playbook (AI as a sales lever, not a replacement for salespeople).
  • At Qamera AI / product: onboarding + adoption > feature creep. 70% of the value = the client’s team actually starting to use it.
  • Resonates with Agentic Coding and Harness Engineering — swapping the tool (Claude Code) alone is only ~30%; the rest is changing the developer’s workflow.

📖 Further reading

Template: knowledge_note_info

Graph View

Backlinks