Start here
The technology is not the problem
Most professional services firms have trialled AI. Almost none have turned those pilots into something the whole team relies on. The technology is not the barrier. The absence of a structured rollout plan is.
Expert firms should not lose to faster, AI-enabled competitors simply because they did not act in time — yet that is exactly what happens when pilots stay scattered, wins stay siloed, and governance gets bolted on after the fact. What follows is five moves drawn from OpenAI’s leadership guidance, applied to how UK consultancies and advisory firms actually operate. Work through them in order. Skip one and the others will not hold.
Why now
Three numbers behind the pressure
Before you build the plan, three numbers are worth sitting with.
- More than 5.6 new frontier models have been released since 2022.
- The cost to run GPT-3.5-class models has dropped 280 times, and AI adoption is spreading four times faster than desktop internet did in its day.
- Early movers are already growing revenue up to 1.5 times faster than the peers watching from the sidelines.
The risk is not that AI turns out to be overhyped. The risk is that your competitors figure out the operational playbook before you do.
These moves map directly to the 5 Steps for AI Leadership: Align, Activate, Amplify, Accelerate, Govern. The order is non-negotiable. Firms that skip to Accelerate without Alignment are the ones complaining six months later that nothing stuck.
Move one — Align
Make your “why” clear before anything else
AI adoption fails quietly when employees cannot answer why their firm is investing in it. Without a clear leadership narrative, people treat AI as optional — something to use if they find time, or avoid because they are not sure it is allowed.
Write a two-paragraph AI position stating the competitive reason, the client-expectation reason, and the growth reason for using AI. Then make it visible. Moderna’s CEO mandated that staff use ChatGPT twenty times a day — an extreme example, but the principle holds: the signal from the top decides whether AI feels like optional experimentation or a genuine shift in how work gets done.
Pick one measurable adoption goal and bake it into your planning KPIs. Most firms miss this entirely — they install tools, circulate a few use cases, then wonder why adoption flatlines. What gets measured gets done, so track three things: time saved per workflow, time to production for new initiatives, and adoption rates by team. Run a quick pulse survey first to understand where appetite and anxiety sit before you ask anyone to change how they work.
Move two — Activate
Training that happens in the flow of work
Nearly half of employees say they lack the training to use AI effectively, and they rank training as the single biggest factor in whether adoption succeeds — that is from OpenAI’s own research on enterprise rollouts. The pattern we see constantly in UK firms: licences purchased, no clear guidance given. In some cases people use ChatGPT in secret, worried their manager will think they are cutting corners.
That is not an AI problem. It is a leadership problem.
Three things close the gap.
- Role-based training in context. Not abstract theory. Show your specific team how AI applies to the work they are doing this week — proposal drafting, research synthesis, client reporting.
- A champions network. Identify two or three AI optimists with technical aptitude, give them resources and a mandate. OpenAI runs a champion network through its enterprise academy; you can build a version inside your firm with very little overhead.
- Protected experimentation. Run a no-code hackathon on the first Friday of each month and fast-track the winners. Notion used this approach before launching Notion AI internally. The best ideas come from the people doing the work — give them space to surface them.
Link AI engagement to performance conversations. If it matters, it needs to be measured.
Move three — Amplify
Stop letting wins die in silos
One team figures out a prompt that saves three hours on a standard deliverable. They use it every week. Nobody else knows it exists. This is the most common failure mode in professional services AI adoption: individual wins compound for one person and disappear when they leave. You have wins — they just do not compound.
Fix it with three things.
- A central knowledge hub. Confluence, Notion, or SharePoint — the tool does not matter. What matters is that it holds your approved prompts, training resources, and AI usage policy, and that it is surfaced where people actually work. If it lives in a folder nobody opens, it does not exist.
- An internal AI wins newsletter. Short. Weekly or fortnightly. Share the problem, the workflow, the hours saved — ChatGPT can help format and automate it. Aim for three firm-wide wins a month. Document once, reuse ten times.
- A community of practice. A Slack channel or Teams space where champions seed discussions and share resources. No budget required, just a moderator and a commitment to keep it active.
Move four — Accelerate
Remove the friction between pilots and production
Most AI pilots die between proof of concept and production — not because the idea was bad, but because the approval process took three weeks, the data access request was denied, or nobody owned the decision to scale.
Four accelerators close the gap.
- Unblock tools and data access. If teams wait weeks for approvals or IT signoff, give them a clear process to submit recommendations. Bottlenecks here kill momentum faster than anything else.
- A simple intake process. One form, a clear prioritisation rubric, transparent decisions about what gets resourced. This removes the ambiguity that stalls good ideas.
- A centralised AI hub. A single place to test, share, and scale use cases. Not every pilot needs to become a permanent workflow, but the best ones should have a clear path to production.
- A cross-functional AI council. A small, exec-sponsored group with authority to unblock and fast-track. Keep compliance and risk in the loop from the start, not bolted on at the end.
BBVA’s approach is instructive: they built a central network, moved fast with clear alignment, and doubled down on the teams generating the most measurable saving. Rewarding winning teams creates the social proof that shifts firm-wide behaviour. Track time to production, approval times, and whether the highest-impact work is getting resourced.
Move five — Govern
Clarity, not control
Governance is the move most firms either skip or overengineer into a roadblock. Neither works. Good governance does one thing: it tells people what is safe to try, what needs escalation, and what data they should never put into an AI tool. It should enable speed, not slow it down.
- A simple usage policy. One page. What is allowed, what needs approval, what is off-limits — including data-handling rules. A clear answer removes the anxiety that keeps people from experimenting.
- A policy copilot. A custom GPT that answers policy questions in plain English and routes edge cases, removing the bottleneck of waiting for a compliance answer before you can move.
- Quarterly reviews. Light audits to keep rules current as the technology and your use cases evolve. If your governance process is slowing launches, fix the process.
The plan
Your 30-60-90 days
- First 30 days. Publish your AI position. Set the adoption goal. Appoint your champions. Schedule the first monthly hackathon.
- Days 31-60. Launch the knowledge hub. Start the internal wins newsletter. Open the intake process with a clear rubric. Stand up the AI council.
- Days 61-90. Move one or two pilots to production. Reward the teams generating impact. Run your first governance review.
The real question
An operational discipline, not a technology project
AI is not a technology project. It is a new way of working — the same shift the internet represented, moving at four times the speed. The firms that look back on this period with confidence are the ones that treated adoption as an operational discipline: align the team, activate it properly, amplify the wins, accelerate from pilot to production, and govern with clarity rather than restriction.
The KWA framework applies at every layer — you are building Knowledge (your hub, prompts, policies), Workflow (standardised, repeatable processes), and eventually Agents (AI doing real work in production). Most firms are still stuck at Knowledge, and the gap between Knowledge and Workflow is exactly where this plan closes.
Twelve months from now, one of two conversations is happening in your boardroom. Either “look at what AI has done to our margin,” or “why did we not move sooner?” Which one is a choice you make now, not later.