Ask anyone in your team whether AI makes them faster and the answer is almost always yes. In the UK government's largest workplace trial, 20,000 civil servants given Microsoft 365 Copilot reported saving an average of 26 minutes a day [1]. Ask the finance director whether profit has moved and the answer is very different: in McKinsey's latest global survey, just 39% of organisations report any EBIT impact from AI at all [2]. Same technology, same year, two different stories. The gap between them is where most AI investment quietly disappears.

The personal gain is real

It helps to be clear that the individual story is not hype. In a well-known study of customer support agents, access to a generative AI assistant raised productivity, measured by issues resolved per hour, by 14% on average, and by 34% for novice and low-skilled workers [3]. Drafting, summarising, researching and first-pass analysis are tasks where a good assistant genuinely compresses effort.

Two cautions travel with those numbers. The government trial's savings were self-reported, and the authors note it was not possible to identify how the saved time was spent [1]. And the biggest gains in the customer support study went to the least experienced people, which matters for what comes next.

Why the organisational gain is so much harder

Saved minutes are not saved money. Twenty-six minutes a day is scattered across a dozen small tasks. Nobody can bank it, and unless someone decides what that capacity is for (more sales calls, faster quotes, fewer overtime hours), it dissolves into a longer lunch or a slightly more polished email. Individual time savings only become organisational value when a manager converts them into something measurable.

The bottleneck moves, it doesn't vanish. A business is a chain of handoffs. If AI helps one person produce a proposal in half the time, but approval, pricing sign-off and data entry still run at the old speed, the end-to-end process is no faster. The proposal simply waits in a different queue. McKinsey found that redesigning workflows has one of the strongest contributions to meaningful business impact, yet only about 6% of respondents qualify as AI high performers [2]. Most firms add the tool and leave the process untouched.

The benefit is uneven. If AI lifts less experienced staff far more than experts [3], then the "average" gain hides a lot of variation. Rolling out one licence to everyone and expecting a uniform uplift is how businesses end up with impressive pilot slides and flat results.

Tools learn nothing about your business. MIT's NANDA initiative attributes the failure of most enterprise pilots to a "learning gap" for both tools and organisations, and to flawed integration, rather than the quality of the underlying models. Its widely reported headline is that 95% of companies in its dataset see generative AI implementation falling short [4]. General-purpose assistants work brilliantly for one person at a keyboard but struggle when they can't see the company's data, rules and workflows. Note that this report was circulated as a research preview and has drawn methodological debate, so treat 95% as directional rather than precise.

Shadow AI fills the vacuum. The same research highlights widespread "shadow AI": employees using unsanctioned tools like ChatGPT outside official channels [4]. Personal productivity rises, but the business gains no shared data, no governance and nothing it can repeat or scale.

Nobody owns the outcome. Individual adoption needs a login. Organisational change needs a named owner, a baseline, a target and a decision about what happens to the freed capacity. In many SMEs none of those exist.

What to do about it

Start with a process, not a tool. Pick one end-to-end workflow that matters commercially, such as quote-to-cash or customer onboarding, and map where time and errors actually accumulate.

Set a baseline before you switch anything on. If you can't state today's cycle time, error rate or cost per transaction, you can't prove tomorrow's improvement.

Decide in advance where the freed time goes. Capacity should be assigned to a target: more throughput, faster response, or lower cost, with a named owner accountable for it.

Fix the plumbing. Connect the AI to clean, shared operational data so it works on your business rather than generic knowledge.

Bring shadow AI into the light. Offer sanctioned tools and clear guardrails so individual enthusiasm becomes shared capability instead of a hidden risk.

Treat individual gains as the on-ramp. Personal productivity builds confidence and skills, which is valuable. The mistake is stopping there and calling it transformation.

The bottom line

AI makes people faster almost immediately. Making a business faster takes redesigned workflows, connected data and clear ownership. The firms that understand the difference are the ones turning minutes saved into margin earned. If you'd like help identifying where your own gap sits, talk to InfinityX Digital at https://infinityxd.uk.

Sources & quotes

[1] UK Government, Microsoft 365 Copilot experiment: cross-government findings report, 2025. gov.uk

[2] McKinsey & Company, The state of AI in 2025: Agents, innovation, and transformation, 5 November 2025 (1,993 respondents, 105 countries). mckinsey.com

[3] Brynjolfsson, Li and Raymond, Generative AI at Work, NBER Working Paper 31161 (revised November 2023). nber.org

[4] Fortune, "MIT report: 95% of generative AI pilots at companies are failing", 18 August 2025, reporting MIT NANDA's The GenAI Divide. fortune.com

InfinityX Digital

Your value co-creators!

Start a conversation

London, UK — remote across Europe

(C) 2026 InfinityX Digital-LTD - Your value co-creators

Home-Work-Who we are-Services-Contact