A founder of a 14-person professional services firm asked me last month why his accountant was getting so much more out of AI than he was. Same kind of business on paper, similar size, similar client base. He had paid for the same Copilot licences. He had read the same articles. And yet his accountant was billing out a redesigned workflow while he was still using AI to clean up emails.
The honest answer is that AI does not deliver the same value to every business, even within the same broad sector. It rewards certain conditions, and where those conditions are present it compounds. Where they are absent, it sits there looking helpful and changing very little.
This piece sets out what those conditions are, what the evidence says about where AI is currently creating most value, and how to read your own firm against that pattern.
Which industries are actually pulling ahead with AI?
The pattern is sharper than the headlines suggest. A 2025 World Economic Forum analysis found that sectors able to use AI intensively are growing productivity at roughly 4.8 times the rate of sectors where AI is harder to apply. Value capture is heavily concentrated in information services, finance, and certain manufacturing niches. Online retail and digital advertising sit at the top because they run on data and feedback at huge scale.
The global AI market sat at about 244 billion dollars in 2025 with projections towards 800 billion by 2030, but that headline figure obscures how uneven the distribution is. The Competition and Markets Authority’s 2023 report on AI foundation models concluded that value pools around firms with large digital data assets and compute-intensive workflows. The same forces that made search and advertising winner-take-most are now pulling AI value into the same kinds of organisation. Smaller firms in data-light sectors do not get nothing, they just get a different shape of gain.
What conditions decide whether AI creates real value?
Three conditions, working together, do most of the explanatory work. The first is data intensity, the firm already runs on structured digital data that AI can plug into. The second is repetitive information-heavy work, drafting, reading, routing, summarising. The third is clear feedback loops, fast and measurable outcomes that let models improve. When all three are present, AI compounds. When one or more is missing, gains are real but smaller and slower.
Finance is the textbook example of all three lining up. Banks like HSBC and Barclays already had structured transaction data, repetitive monitoring work, and a brutally clear feedback signal (fraud or not fraud). The same logic applies in reverse, a UK plumbing firm with paper invoices, bespoke jobs, and slow feedback will not see the same uplift no matter how good the tools become, until something in the data and process picture changes.
Why feedback loops matter more than people think
Online retail platforms run millions of micro-experiments daily. Click-through rate, conversion rate, basket abandonment, every action produces a signal the model can learn from. That is why Amazon and Google have been able to compound AI gains for over a decade. A consultancy producing five complex proposals a month does not have that kind of feedback signal, so AI helps with the drafting but cannot tune itself the way an ad platform does.
This is why the gap between industries widens rather than narrows. Sectors with tight feedback compound. Sectors without it improve linearly. Over five years, linear and compounding diverge dramatically.
Where is AI making the biggest measurable difference today?
The strongest evidence sits in functional rather than sector terms. McKinsey’s 2023 state-of-AI survey found supply chain and inventory functions reporting the highest revenue uplift, with more than 5% revenue growth for a majority of respondents using AI in those areas. Customer support is the other standout, a controlled trial documented agents using AI handling 13.8% more enquiries per hour with a small quality improvement on top.
Experimental studies of generative AI on office tasks show average performance gains around 66% on complex work, but almost no benefit on tasks that are already fully automated or tightly standardised. The pattern is consistent. AI helps where humans were doing cognitive work that had repetition and structure inside it. It helps less where the work was either already mechanised or genuinely artisan.
For UK service firms specifically, the Law Society and ICAEW have both documented 10 to 30% time reductions on document review and drafting in legal and accounting work. That is real money for a 20-person firm, but it is not the order-of-magnitude shift you hear about in online retail.
What does this mean for a UK service-firm owner?
If you handle high volumes of digital documents, similar emails, and repeatable reports, and your work produces measurable outcomes like response times or chargeable hours, you sit in the group where AI can shift the economics of your firm. Modest investment in copilot tools you probably already pay for will give you 10 to 50% time savings on drafting and admin work, with gains compounding once processes catch up with the tool.
If your value is mostly physical, on-site, bespoke, or your records are paper-based, the gains are real but they sit one layer back, in quoting, scheduling, marketing, and documentation rather than in the core delivery. The right move is not to chase generic productivity claims, it is to digitise enough of the front-office workflow that AI has something to grip.
The regulatory frame you can’t skip
UK regulators have been clear about where the friction sits. The ICO requires lawful basis, transparency, and a DPIA for high-risk AI uses, particularly anything involving personal data, staff monitoring, or scoring customers. The FCA expects authorised firms to treat AI as part of their systems and controls regime, with board-level accountability under SM&CR. The NCSC has flagged data-rich service SMEs as prime targets for AI-enabled cyber attacks.
For firms serving EU clients, the EU AI Act classifies AI used in employment decisions, credit scoring, and access to essential services as high-risk, with obligations on risk management, data governance, and human oversight. These constraints do not stop adoption, they shape where the early wins are. Internal productivity tools and drafting assistants move fast. AI making substantive decisions about customers or staff moves slowly and needs proper governance to move at all.
What should you actually do this quarter?
Run a workflow audit, not a tool audit. Walk through a typical week with your operations lead and mark every recurring activity that involves high volumes of similar documents, emails, or data. Mark every triage point, the moments where someone is routing, prioritising, or producing a template-shaped output. Those are your AI-amenable workflows. They will sit mostly in customer service, finance and admin, sales operations, and recurring reporting.
Start with the tools you already pay for. Microsoft 365 Copilot, Google Workspace AI, and the AI features built into your CRM and practice-management software cover 80% of the realistic value for a 5 to 50 person firm. Custom models and bespoke implementations rarely make sense at this scale unless you have a clear repeatable use case with thousands of similar transactions behind it.
Set a copilot-not-autopilot policy from day one. AI drafts, summarises, and suggests. A human reviews and decides, particularly for anything regulated, client-facing, or involving personal data. Write the policy down, share it with the team, and revisit it quarterly. The firms getting hurt by AI right now are not the ones moving slowly, they are the ones moving fast without a frame.
If you want a peer to walk through your specific workflows and pick the two or three that will actually move the needle, book a conversation.
