A partner at a fifteen-person consultancy told me last month that her senior people were spending the first hour of every workday lifting yesterday’s meeting notes into project files, then another hour mining old decks for content to drop into this week’s proposals. She knew, roughly, what the lost time was costing the firm. She did not know what to do about it without inviting either a Slack-bot free-for-all or a six-figure procurement exercise. That gap between knowing and doing is where AI in consulting actually lives.
This is a primer for owner-operated UK consulting firms, the five to fifty person practices that bill on time and outputs and feel the cost of inefficiency in margin rather than headlines. The shape of the question is not whether to use AI. The shape is which jobs to point it at first, who owns each one, and what to refuse to let it touch.
What does AI actually save consultants time on?
The honest answer is research, drafting and synthesis, not judgement. A Harvard Business School field experiment with consultants found generative AI lifted task completion rates by about 25% in the studied workflow, with quality improving on tasks inside the model’s frontier. AI shines where the work is document-heavy, repeatable, and reviewable in the same afternoon.
The four jobs that pay back fastest in a small consultancy are proposal drafting from approved templates and past bids, internal knowledge search across prior decks and case notes, meeting transcription with action extraction and client recap drafts, and first-pass research summaries with named sources you then verify. None of these are the work clients pay for in isolation. All of them are the work that slows down the work clients pay for.
Why does this matter for an owner-operated firm?
The commercial point is capacity per consultant, not headline cost savings. If a senior gets ninety minutes back a day on note tidying and proposal scaffolding, that time goes into more bids out, more chargeable hours, or more space to sell. For a five to fifty person firm, that is the difference between a busy quarter and a profitable one, and it opens a real window before adoption levels out.
The trap is treating the productivity gain as instant revenue. It is capacity. Capacity has to be sold or it leaks back into Parkinson’s law and the same work fills the new time. The firms getting commercial value from AI are the ones with a partner who has decided in advance what the freed capacity is for, whether that is two extra proposals a month, one extra delivery slot per consultant, or shorter project cycles at the same price.
Wingenious’s roundup of UK AI adoption notes that around 15% of smaller organisations have adopted AI compared to 52% of UK businesses overall. Treat that adoption gap as a window. The five-person consultancy that gets two extra bids out a quarter on the back of better proposal scaffolding does not need a procurement business case to justify the move. It needs a partner willing to own it and a quarter to prove the numbers shift.
Where will you actually meet it in delivery?
You will meet AI first in the inbox and in the second draft. Meeting transcription tools produce a recap that needs editing, not posting. Internal search tools surface three past decks when you ask for a methodology summary, and you pick the one closest to the new context. The pattern is that AI gives you a competent eighty-percent draft and the consultant earns the fee by being right about the missing twenty.
Consultancy.uk’s coverage of UK firms describes an executive AI sponsor inside the consulting firm who governs which use cases are live, which are in pilot, and which are off-limits. That role matters more in a small firm than in a large one, because there is no enterprise risk function to absorb a mistake. A named partner, even part-time, who owns the AI use case map is the difference between a firm that scales the wins and a firm that discovers a confidentiality breach on a Friday afternoon.
When should the firm say no to AI?
There are three lines worth holding firmly. The first is confidential client data. The ICO is clear under UK GDPR and the Data Protection Act 2018 that the firm remains responsible for compliance regardless of the tool, including lawful basis, transparency and accuracy. The second is regulator-facing output, where the FCA’s DP5/23 sets expectations on governance, explainability and model risk that apply when advising FCA-regulated clients.
The third is unverified marketing claims. The CMA’s AI foundation models work flags the risk of AI-generated performance claims that the firm cannot evidence, and for a consultancy that monetises trust, an unprovable claim on a website is a slower problem than a confidentiality breach but a harder one to undo. The NCSC’s guidance on generative AI adds the operational angle, that prompt injection, data leakage and weak integrations all increase the firm’s attack surface as adoption grows. None of these are reasons to refuse AI. They are reasons to install governance before scale.
What does a sensible starting move look like?
A realistic first quarter has three ingredients. Pick two use cases, name one partner to own each, and set one measurable KPI per use case before the tool goes live. The KPI set worth tracking is narrow, proposal turnaround time, research time per project, meeting-note processing time, content reuse rate, and consultant billable utilisation. If the numbers move within six weeks, scale. If they do not, retire the use case.
The two use cases worth starting with for almost any UK consultancy are proposal scaffolding from past bids, and meeting-note processing with action extraction. Both have clear baselines, both are reviewable by the consultant within a day, and both stay inside firm-controlled material rather than touching live client data. Tools to consider exist across the price band, from a Copilot or ChatGPT Enterprise seat through to a structured retrieval setup over the firm’s archive, and the choice matters less than the partner who owns the outcome.
The single piece of advice worth taking from this primer is that the productivity gain is real, the regulatory risk is real, and both are managed by the same move, which is putting one person in charge of each use case with one KPI attached. The firms scaling AI well in UK consulting are the ones where a partner can tell you on a Monday what AI is doing for them, what it is not allowed to touch, and what the numbers say about whether it is earning its keep. Tool stack is secondary, ownership and measurement come first.
If you want to talk through how that map looks for your firm specifically, book a conversation.
