The core metric

Learning Velocity

AI is making intelligence abundant. The organisations that learn fastest will win.

Definition

Learning Velocity is the rate at which an organisation turns new information into better decisions, changed behaviour and improved outcomes.

Mark Cameron on stage addressing a leadership audience at The CEO Institute

The argument

A faster learning loop does more than save time. Each better question, decision and action improves the starting point for the next cycle.

Why it matters now

When intelligence becomes cheaper and more available, access stops being the advantage. The ability to absorb, apply and learn from it becomes decisive.

Small differences become different futures.

Same tools. Same starting point.

Learns each cycle

Adds more tools

NowSense → Reason → Act → LearnNext cycle

Evidence

Learning Velocity is Mark Cameron’s interpretive framework, not a validated or standardised metric. Published field studies show AI can change how quickly people learn and perform in specific settings; they do not measure Learning Velocity or prove the framework causes advantage.

Read the 88% / 5% methodology (2025 data) →

The model

Sense what is changing. Reason about what it means. Act while the insight is useful. Learn from the result. Begin again with a stronger frame.

Metaphor / example

Two organisations can adopt the same technology on the same day. The one that shortens every hand-off in its learning loop gradually creates a lead the other cannot close with tooling alone.

Leadership implication

Find where the loop is slow: noticing, deciding, acting or learning. Improve the constraint before adding more technology.

Common misunderstanding

Learning Velocity is not speed for its own sake. Fast action without reflection can compound error.

External evidence and interpretation

What the research shows—and what Mark reads into it.

Interpretive framework by Mark Cameron. Not a validated, standardised metric. Supporting research informs the argument but does not establish ownership of the idea or causality.

Published finding
In one company’s customer-support operation (5,179 agents), access to a generative-AI conversational assistant raised issues resolved per hour by about 14% on average in the November 2023 working-paper version, with much larger gains for novice and less-experienced agents. Later published versions may report revised figures.
National Bureau of Economic Research — Generative AI at Work (NBER Working Paper 31161, April 2023, revised November 2023) ↗
Mark’s interpretation
Mark reads this as one published example of a shorter learning loop: the assistant spread experienced agents’ know-how to newer agents faster. It is a single setting and does not measure Learning Velocity.
Published finding
The NBER Reporter summary (2024) describes the same study: gains concentrated among the least-experienced workers, as the system passed on behaviours learned from skilled agents.
National Bureau of Economic Research — The Economics of Generative AI (NBER Reporter 2024, No. 1) ↗
Mark’s interpretation
An illustration of organisational knowledge travelling faster—Mark’s interpretation, not the authors’ framing.
Published finding
In a 2023 field experiment with 758 BCG consultants, GPT-4 improved speed and quality on tasks inside the AI’s capability frontier, but the authors describe a ‘jagged’ frontier where AI falls short on other tasks—so human judgement about when to rely on it matters.
Harvard Business School / D^3 Institute — Navigating the Jagged Technological Frontier (HBS Working Paper 24-013, September 2023) ↗
Mark’s interpretation
Faster is not automatically better: learning loops need judgement about where AI helps, which is why the framework pairs speed with quality safeguards.

How to observe it in practice

Practical signals, not a validated scale. Use them to compare a team with its own baseline.

Decision latency: how long a decision waits between the moment it is needed and the moment it is made.
Evidence-to-action time: the elapsed time from a signal or finding reaching someone with authority to a changed action.
Feedback closure: whether the result of an action is captured and actually changes the next decision—and how long that takes.
Quality safeguards: track error, rework and harm alongside speed, so faster loops are not rewarded for compounding mistakes.
Compare a team with its own baseline over time; there are no universal thresholds, and numbers should prompt conversation rather than ranking.

Try the idea