AI Adoption & Change Management for Professional Services
Home » AI Adoption & Change Management for Professional Services

AI Guides

AI Adoption & Change Management for Professional Services

Most firms do not struggle to buy AI. They struggle to get it used safely, by the right people, on the right work. This guide covers shadow AI, de-risking a first rollout, getting fee-earners on board, scaling from one team to the whole firm, training and workflow design, and the myths that slow decisions.

Book a free AI Capacity Assessment
Take the AI Readiness check
Consultancy sector illustration

The short version

  • AI is already in your firm: 55% of employees use it for work, against 35% of businesses reporting formal adoption (ONS).
  • Banning it rarely works. The NCSC advises reducing shadow AI through culture, visibility and approved alternatives.
  • Start with one workflow that happens often and costs little if it goes wrong, and finish it before starting another.
  • Train before you choose tools, and bring partners in. They are often the least prepared group.
  • Measure quality, review time and real use, not just logins.

In this guide

AI adoption in professional services: where firms really are

Official data shows AI use spreading quickly but shallowly. The Office for National Statistics reports that the share of UK businesses with ten or more employees using at least one AI technology rose from around 12% in late 2023 to around 35% by June 2026. The average adopter uses only about 1.6 AI technologies, and analysis of the same data found just 10% of adopters describe their use as extensive. Larger firms lead, at 49% against 28% for the smallest firms in scope.

Professional services sit near the front. The LEAP Global Report 2026 found 31% of UK legal professionals use integrated or legal-specific AI daily or as a core part of their work, and 62% are active, regular users. Seventy-nine per cent of legal professionals in the UK and Ireland said AI saves their firm a moderate to significant amount of time.

The pattern is adoption that has widened without deepening. Plenty of people are trying AI. Few firms have turned it into a managed, repeatable part of how work gets done. That gap is where the risk and the opportunity both sit.

Shadow AI: the adoption you did not plan

The ONS found that 55% of employees say they use AI for work, against 35% of businesses reporting formal adoption. That difference is shadow AI: tools used without the firm’s knowledge or approval.

The scale is well documented. The NCSC’s September 2026 guidance cites Microsoft research finding that 71% of UK employees have used AI tools their employer has not approved. Deloitte’s 2026 workforce survey found 63% of workers have used AI at work and that 17% of generative AI users pay for their own tools, which it estimates costs UK workers around £1 billion a year. In legal, research published in May 2026 by Resultsense reported that 59% of legal professionals admit to using unauthorised AI on client work, while 68% of firm leaders believe they have full visibility.

Seniority does not protect you. A TrustedTech survey found 62% of UK senior leaders use unapproved AI tools, against 31% of staff below decision-maker level, and 22% of UK employees said they would keep using AI even if it were banned and disciplinary action were possible.

The NCSC’s advice is to reduce shadow AI rather than try to eliminate it. In practice:

  • Find out what is being used. Ask without blame. People hide what they fear will be punished.
  • Understand why. In the TrustedTech survey, 24% of shadow AI users said limited access to approved tools was a factor.
  • Provide something better. A safer tool that is also faster beats a ban.
  • Write the rules down. See our AI governance guide.

De-risking adoption, especially in a consultancy

De-risking does not mean going slowly. It means making each step small enough to check. In a consultancy or any advisory firm, where client confidentiality runs across many engagements, we suggest this order:

  • Choose one workflow using a simple test of volume and risk. Start with tasks that happen often and cost little if they go wrong, such as intake, document chasing, appointment booking and routine status updates.
  • Define done. Write what a good result looks like and which named person approves it.
  • Run it in shadow mode. The agent works alongside your team on real work, but nothing it produces goes out. People compare its drafts with what they actually did.
  • Launch with approvals on. Relax them only when the record shows they are no longer catching anything.
  • Monitor and retrain when your processes change.

Advisory firms have extra things to settle first: which client material may be used, how work for one client is kept apart from another, who owns the resulting intellectual property, and how AI use is described in engagement letters. A General Assembly survey of large professional services firms found most had abandoned at least one AI initiative because they lacked the necessary skills, which is a reminder that the people side deserves as much planning as the technology. Our article on de-risking AI agent adoption in a consultancy goes deeper.

Getting fee-earners to use the tools

Fee-earners are not resisting AI out of stubbornness. They are usually responding to something real: not enough time to learn a tool, doubt about whether the output can be trusted, fear of exposing a mistake, unclear rules about what is allowed, or tools that sit outside the systems where their work happens.

What tends to work:

  • Teach the basics before choosing the tool. Practical, tool-neutral literacy covering what AI can and cannot do, how to check its output and your professional obligations makes any later tool easier to adopt.
  • Start with their own tasks. Use real, low-risk work from the team’s own week, not a generic demo.
  • Make the approved route the easy one. If the approved tool is slower than a free one, people will go round it.
  • Use peers. One respected person in each team who uses the tool well is worth more than a firm-wide email.
  • Be honest about jobs. ONS data shows fewer than one in 15 adopting businesses reported reducing headcount because of AI. Say plainly what the firm intends.
  • Bring partners in. In the General Assembly survey, 74% of legal respondents named partners as the least prepared group. Leaders who model good practice, and who stop using unapproved tools themselves, set the tone.

Scaling from one team to the whole firm

Scaling is where most efforts stall. In RSM’s 2026 middle-market survey, professional services firms that reported moderate or limited success with AI pilots most often pointed to data quality (45%) and integration problems (41%), not to the AI model itself.

A sensible path from one team to the whole firm:

  • Prove one workflow in one team. Measure the time it saves and the errors it catches.
  • Add a different workflow, or a second team. Do one at a time. Twelve half-built pilots save nobody any time.
  • Standardise what worked. Turn the policy, permissions table, logging and approval rules into a firm-wide template.
  • Give it an owner. One named person is responsible for the approved tool list, the logs and the monthly review.
  • Keep reviewing. Processes change, so agents need retraining. That is what an ongoing support subscription is for.

Training and workflow design

Training works best in layers:

  • Everyone: what AI can and cannot do, what must never go into a tool, and who to ask.
  • Users of a specific agent: what it handles, what it hands over, and how to approve or reject its work.
  • Reviewers: how to check outputs, citations and figures, and what to do with an error.
  • Owners: how to read the logs, spot drift and decide when to retrain.

Workflow design matters as much as training. Before building anything, draw the workflow on one page in four steps: what the agent reads, what it decides within rules you set, what it does with the permissions you give it, and what it records. Put a person’s name against each approval and write down the exceptions. If the sketch does not make sense on one page, the workflow is not ready to automate.

Seven AI myths that slow decisions

“We are too small for AI.” Adoption is lower among smaller firms, but not negligible, and a small firm can decide and move faster than a large one. Our cost guide shows what a first build typically costs.

“A ban will keep us safe.” The NCSC says shadow AI is unlikely to disappear and recommends reducing it through culture, visibility and secure alternatives.

“Free tools are fine for drafting.” The SRA warns that both free and paid tools carry risks depending on terms, settings and architecture, and that client information in an open tool can waive privilege for good.

“AI means redundancies.” ONS data shows fewer than one in 15 adopting businesses reported cutting headcount.

“It makes things up, so it is useless.” A general chat tool will always produce an answer. An agent built to answer only from your sources, to show where an answer came from and to say “I don’t know”, behaves differently. See AI agents explained.

“We need a big programme.” Gartner forecasts that over 40% of agentic AI projects will be cancelled by the end of 2027 because of rising costs, unclear value or inadequate risk controls. Start with one workflow that earns trust.

“Our people will resist.” In a Studio Graphene survey of UK business leaders, 61% said frontline staff are more comfortable with AI than the senior team. Resistance is often at the top.

For more, see our article on AI myths in 2026.

Frequently asked questions

How long does it take to adopt AI in a small firm?

Building a first agent takes our clients from 7 to 21 business days depending on scope. Adoption takes longer, because it depends on training, review habits and trust. Plan for a first workflow to settle over weeks, not days, and to be reviewed every month.

Who should own AI adoption?

One named partner or director who is accountable, supported by a day-to-day lead who understands the workflow. Spreading ownership across everyone usually means no-one owns it.

How do we measure adoption?

Look at active use of the approved tool, the share of the target workflow it handles, error and rework rates, review time, and whether unapproved tool use is falling. Logins alone tell you little.

Related reading

Plan a first workflow that earns trust

Law firms and accountancy practices can book a free 30-minute AI Capacity Assessment to find the first workflow worth automating. Everyone else can take the free 8-question AI Readiness check.

Book a free AI Capacity Assessment
Take the AI Readiness check

Sources

Survey figures are quoted as published by each source. Some surveys cover the UK and US or specific sectors and sample sizes, so treat them as indicators, not forecasts.

More guides

Next steps

Related articles