B2B AI PlatformB2B AI Platform
AI Productivity vs. Functional Capability
B2B AI Platform

AI Productivity vs. Functional Capability

Laura HolmesWords by Laura Holmes

AI productivity entered the boardroom through cost accounting; strategy arrived later. A drafting task that once consumed an afternoon can now be measured in minutes, and that saving is easy to multiply by volume and loaded cost. Capability became visible in a different moment, usually when an operations director had banked the saving and then found that the organisation could not do anything it had been unable to do before.

Consultancy.uk reported in November 2026 that Suzanne Coles, a partner at Curzon Consulting, applies a direct test to AI business cases: "What capability are we removing, and how will we replace the value it provides?" The article treats AI as a delivery success and still separates output from judgement. It argues that one can rise as the other falls, and that most reporting captures the first.

That distinction matters now because task-level gains have strong evidence and firm-level capability has thin evidence. The Department for Science, Innovation and Technology's assessment estimates that around 70% of UK workers are in occupations containing tasks AI could perform or enhance, while only around one in five firms use or plan to use AI. Exposure differs from adoption, and adoption differs from capability. The gap between those three things is where most AI programmes will eventually be judged.

What separates a productivity gain from a capability gain

A time saving reported by the person who received it is a perception, and perceptions of AI speed have been measured as wrong in both directions. In METR's randomised trial, sixteen experienced maintainers of large open-source repositories completed 246 real issues, with each issue randomly assigned to allow or forbid AI assistance, and when AI was allowed the developers took 19% longer, forecast a 24% speed-up, and after the trial still believed AI had made them 20% faster. The result is a snapshot of one setting and does not settle every profession. A stopwatch held by an enthusiastic user lacks the controls of a measurement system.

The California Management Review described five countervailing forces that can neutralise a task-level gain, including cognitive offloading, work that looks productive but has no strategic use, workflow saturation, delayed organisational redesign and neglected intangible inputs, and it reported a 2024 Upwork survey in which 96% of executives anticipated productivity gains from AI, while 77% of employees said AI had increased their workload and 39% spent more time reviewing or moderating AI-generated content. The point supports counting the review, filtering and coordination that automation moves elsewhere.

The Federal Reserve's 2026 note on the AI buildout found that sectoral productivity trends have stayed broadly similar across levels of AI exposure, which suggests micro-level gains have not yet become an aggregate shift. A firm can point to a faster team and still be unable to show that the business has become more capable of handling complexity, exceptions or unfamiliar work.

Source: Consultancy.uk's report.

Level Productivity question Capability question
Individual Has output per person risen? Can the person form a defensible judgement without AI and detect a weak recommendation?
Team Has cycle time fallen? Do colleagues challenge assumptions, combine expertise and resolve cross-functional problems?
Organisation Has unit cost fallen? Are better practices embedded in process, systems and decision rights across teams?

Those rows can move in opposite directions. An individual can produce more while the team makes worse decisions. A team can develop a strong local practice that never becomes part of the operating model. An organisation can report a lower cost per case while losing the informal checks that once caught exceptions before they became incidents.

Consultancy.uk draws a parallel from a post-merger integration programme in a global industrial technology group. Businesses that looked comparable at corporate level carried different operational meanings. Senior engineers understood the history behind each process, the exceptions that carried real weight and the points where functions had to cooperate. Opening up knowledge and aligning procedures did not produce a shared ability to judge trade-offs across the combined group. The lesson for AI is that a shared model output can narrow the range of assumptions colleagues test.

The learning problem is a design problem. Consultancy.uk describes an early-career engineer who starts with partial operating information, forms a view of a possible equipment failure and defends that view to maintenance, operations and customer-facing colleagues while a senior engineer challenges the reasoning. The tool can inject contradictory readings and unusual conditions. The learner later handles an unfamiliar case with less assistance to test whether the judgement transfers. That approach awaits evaluation, and it depends on experienced colleagues having time to supervise and challenge. If every productivity gain becomes a higher output target, the supervision will not happen.

Consultancy.uk also cites evidence that support agents kept productivity gains after AI help was withdrawn, which researchers treated as a sign of learning. The finding does not settle engineering, law or finance, and it supports designing for learning and testing transfer.

Forbes Technology Council frames the distinction in operational terms: productivity asks whether today's work can be done faster; capability asks whether the business can now attempt work that was previously impractical, uneconomic or impossible to scale. That second question can be measured at individual, team and organisational level.

A serious business case should identify the knowledge, judgement and relationships that the automated activity currently builds; decide which of those the organisation needs to keep; describe how new expertise will be formed and shared; track decision quality through error rates, exception handling, rework and performance on unfamiliar cases; and check whether better practice sits in the operating model or depends on one team. Those questions belong beside the financial assessment, and they should be revisited after implementation. If productivity improves while judgement, collaboration or exception handling deteriorates, leaders need to establish whether they have improved performance or moved cost and risk into a later period.

The practical test for a board is whether the team can reach a sound decision with the tool switched off. Where the answer is negative, the organisation has bought throughput and borrowed judgement. Functional capability is the ability to do new work well, repeatedly and without depending on a single person.

Sources

Source: AI may make businesses more productive, but not always more capable, Consultancy.uk

Recently filed

More on this

Institutional Capability: The Real AI Key
Institutional Capability: The Real AI Key
Making AI Predictive Modeling Work
Making AI Predictive Modeling Work
Architecting Workflows for LLM Integration
Architecting Workflows for LLM Integration