Intelligence Amplified logo: the letters AIA in teal with a gold sphereIntelligence Amplified

Measures of achievement & the impact of AI

A simple capability-based approach to measuring a firm's performance using a three lever model and the ratios between them as a lens to assess the impact of AI. AI moves all three, and keeps moving them while it is still improving.

Follow the steps, or drag the levels yourself. Performance is the number that matters, and it moves with the levels relative to each other, not with their size.

↺ repeats while AI keeps improving
How AI develops:

Step 1

Before AI

The firm delivers its work the established way: technology A in the curve, flattening at the top of its S-curve, where each extra effort brings a smaller improvement.

On a common measure of its output, it achieves 40. It could achieve 60 now, with the people and tools it has, if it really worked at it: better methods, better-trained people, better-managed work. And 100 is what it ought to achieve by developing its resources and removing constraints, within what is known to be possible.

Watch Performance is 0.40: the firm achieves 40% of what is possible. Hover over any term for what it means and how it is calculated.

Technological capability over time. Technology A, the established way of working, flattens. Technology B, AI, starts below it, crosses it at t1, stays below it through the critical decision area, overtakes it at t2 and climbs on. The dashed envelope curve E is what is feasible across both. A marker shows where the firm stands at this step.t₁t₂TimeTechnological capabilityenvelope curveEtechnology Athe established waytechnology BAIcritical decision areabefore AI

Technology A, the established way of working, is flattening. Technology B, AI, overtakes it at t₂; between t₁ and t₂ is the critical decision area, when a firm has to invest in both. The envelope E is what is feasible. With acknowledgment to Erich Jantsch.

measures on the same metric

indices for which unity is the maximum

normative planning
PotentialityPotentialityWhat the firm ought to be doing by developing its resources and removing constraints, within what is already known to be feasible. It rises when what is feasible rises: a new technology such as AI. Planning from it is normative planning.now 100100
planning by objectives
CapabilityCapabilityWhat the firm could be doing now, with its existing resources and constraints, if it really worked at it. Better methods, better-trained people and better management raise it; cutting budgets and letting good people go lowers it. Planning from it is planning by objectives.now 6060
programming
ActualityActualityWhat the firm is managing to do now, with its existing resources, under its existing constraints. Offloading work to agents raises it, as far as capability allows. Planning from it is programming.now 4040
LatencyLatencyHow much of what is possible the firm could do now: its latent capability. Cost reports do not show it, which is how it gets run down without anyone noticing.capability ÷ potentiality = 60 ÷ 100 = 0.60060100=0.600capability ÷ potentiality
ProductivityProductivityHow much of what the firm could do now, it is doing. This is the ratio cost reports already watch.actuality ÷ capability = 40 ÷ 60 = 0.6674060=0.667actuality ÷ capability
PerformancePerformanceHow much of what is possible the firm is achieving. It is latency times productivity, so it can never be higher than latency: agents alone cannot raise it past that.latency × productivity = 0.600 × 0.667 actuality ÷ potentiality = 40 ÷ 100 = 0.4000.400latency × productivity0.600 × 0.667actuality ÷ potentiality40100
ceiling: latency 0.600 · before AI 0.400

The levels rise steadily; the ratios dip at every re-estimate and climb back

round 1round 2round 3round 41234234234234levelsratios0.00.51.0potentialitycapabilityactuality
potentiality · capability · actuality (top)latencyproductivityperformance

Compare the paths

Each starts from step 2 of the first round, after AI has raised what is possible, and makes one round of moves.

Agents onlyoffload: actuality 52 · capability 60 · potentiality 1600.325latency 0.375 × productivity 0.867

Actuality rises, but only as far as capability. Productivity can reach 1; performance stops at latency.

People onlyamplify: actuality 40 · capability 100 · potentiality 1600.250latency 0.625 × productivity 0.400

Capability and latency rise, but nothing more is done yet, so performance does not move.

Bothamplify, then offload: actuality 76 · capability 100 · potentiality 1600.475latency 0.625 × productivity 0.760

People raise the ceiling; agents use it. The only path that moves performance past where it stood.

The levels are on an illustrative scale; only their ratios matter.

The levels are illustrative; the pattern is not. Each time AI improves, what is possible moves before anything inside the firm does, so performance falls before it can rise. Offloading work to agents raises it only as far as latency allows; amplifying people raises latency, and performance follows only when the work does. Only both, repeated each round, keep performance rising.

That is what a Diagnose engagement measures — one workflow, one team, a week or two: where the work stands on the three levels today, which of it can be written down part by part for agents and which needs people who understand the whole, and where the gap reopens as the models improve.

Talk to me about your practice