I was at an industry roundtable in February. Mid-market professional services - law firms, consultancies, accountancy practices. The topic was AI adoption. Fifteen managing partners around the table, and within twenty minutes, the conversation had taken on that slightly competitive edge you get when nobody wants to be seen as the one who's behind.
One partner described a "firm-wide AI transformation programme." Another mentioned their "multi-agent workflow pilot." A third talked about custom-built AI tools for client-facing work. Heads were nodding. People were scribbling notes. The energy in the room said: everyone is doing this, and if you're not, you're in trouble.
Afterwards, over coffee, I had three separate conversations that told a very different story. The "firm-wide transformation programme" turned out to be six people with ChatGPT licences and no governance framework. The "multi-agent workflow pilot" was a proof of concept that hadn't moved past a slide deck. The "custom-built AI tools" were a vendor demo that hadn't been commissioned.
I'm not telling you this to embarrass anyone. I'm telling you because if you're making AI investment decisions based on what peers say at events like that one, you're calibrating against a sample that is - to put it politely - heavily curated. And that miscalibration has real consequences for how you spend money, where you focus, and how urgently you move.
So here's the honest picture. Based on what we're seeing across our own client base, published research, and a healthy dose of "what's actually in production versus what's in a PowerPoint" - this is what AI adoption really looks like at mid-market scale.
McKinsey's 2024 Global Survey found that 72% of organisations report using AI in at least one business function. Sounds impressive. But dig into the detail and it gets more specific. Gartner found that only 54% of AI projects make it from pilot to production. And a 2024 Thomson Reuters survey of professional services firms found that while 80%+ of respondents said they were "exploring" AI, fewer than 30% had deployed any AI tool at firm level with defined governance.
That gap between "exploring" and "deployed with governance" is where most mid-market B2B service firms actually live.
When we do AI readiness assessments, the tools with genuine, regular adoption tend to fall into three categories. None of them are particularly glamorous.
AI writing and communication assistance. Microsoft Copilot, ChatGPT, Claude. But in most firms, this isn't a formal programme. It's individual fee earners using these tools for drafting, summarising, and research on their own initiative - often paying for their own subscriptions. The firm hasn't deployed it. The firm hasn't governed it. People are just... using it. Which is both encouraging and slightly terrifying from a data security perspective.
AI-enhanced search and document retrieval. This usually comes through existing vendor features rather than standalone tools. Your document management system added an AI search capability. Your legal research platform bolted on an AI summarisation layer. It's not a strategic AI investment - it's a feature update you may not have even noticed arriving.
AI-assisted scheduling and administrative workflow. Calendar management, meeting notes, basic task routing. Mostly in operations and support functions rather than fee-earning work. And mostly through tools people were already paying for that quietly added AI features.
Notice what's not on that list. AI-powered knowledge management - the dream of every professional services firm - has meaningful deployment in a surprisingly small number of mid-market firms. Client-facing AI applications are rare outside the largest practices. And custom AI development at mid-market scale? I've seen maybe four or five genuine examples in the past year, and none of them were at firms below 300 people.
But everyone at the conference said they were doing this stuff...
I know. And some of them genuinely are. But the firms presenting about AI at conferences are, by definition, the firms with something to present. They're not a representative sample. They're an outlier sample. Calibrating your strategy against them is like benchmarking your fitness against the people who actually go to the gym at 6am. You'll just feel bad about yourself and make poor decisions.
Here's where it gets interesting - and, honestly, a bit boring. The use cases where firms report genuine, measurable productivity improvement aren't the ones generating LinkedIn posts. They're mundane. Almost disappointingly so.
AI-assisted document drafting and summarisation. Reducing time to first draft for standard documents - engagement letters, routine advice notes, internal briefings. Not replacing the expert review. Not producing final-quality output. Producing a first draft that's 60-70% there, which a qualified human then refines. Firms doing this well report time savings of 30-50% on those specific document types. That's real. That's measurable. And it's not remotely sexy.
AI-powered research for well-defined tasks. The emphasis is on "well-defined." When a fee earner knows exactly what they're looking for - background on a specific regulation, a summary of recent case law on a narrow topic, market data for a pitch - AI tools are genuinely faster and often more thorough than manual research. When the research question is vague or exploratory, the results are inconsistent at best.
AI-assisted report generation. Reducing time on routine reporting cycles - monthly client updates, internal performance reports, compliance summaries. Again, not replacing human judgement. Reducing the mechanical assembly time.
What distinguishes the firms getting results from those not? Annoyingly, it comes down to something very simple. The firms seeing measurable improvement defined what success looked like before they deployed. They had structured quality assurance rather than ad hoc "does this look right?" review. And they invested in genuine user training - not a one-hour webinar, but actual hands-on sessions where people learned to prompt effectively and understood the tool's limitations.
The firms getting results are not using more sophisticated tools. They're using straightforward tools more systematically. I keep coming back to this with clients because it's so counterintuitive to the narrative. The advantage isn't in the technology. It's in the discipline around the technology.
Nobody talks about this at conferences, obviously. But in private conversations - and particularly during our AI readiness assessments - a pattern emerges of tools and approaches that were tried and abandoned. Understanding what didn't work is at least as valuable as knowing what did.
Client-facing chatbots. Almost universally underperforming in professional services contexts. The problem is fundamental: the queries clients bring to professional services firms are specific, nuanced, and often emotionally charged. A chatbot trained on your FAQ page cannot handle "I think my business partner has been diverting funds, what are my options?" with the sensitivity and precision required. I've spoken to several firms that launched chatbots, watched client satisfaction scores go down, and quietly removed them within six months. One managing partner described it as "the most expensive way we've ever made our website worse."
AI document drafting for complex, high-stakes work without adequate human oversight. This one stings because the promise was so appealing. For complex work - detailed legal opinions, nuanced advisory reports, bespoke compliance frameworks - AI drafting tools produced errors at a rate that generated more rework than the original drafting time would have taken. Not because the tools are bad. Because the work requires contextual judgement that the tools don't have, and firms deployed them without building in sufficient review steps.
'AI strategy' documents that produced no operational change. This might be the largest single category of AI investment that has produced zero measurable return. A firm commissions an AI strategy. A document is produced. The document contains a roadmap, use cases, and recommendations. The partnership approves it. And then... nothing happens. The document sits on SharePoint. The roadmap isn't resourced. The use cases aren't piloted. Six months later, someone asks "what happened with our AI strategy?" and the answer is an uncomfortable silence.
If your AI strategy doesn't include a funded pilot with a named owner and a deadline within 90 days, it's not a strategy. It's a document.
Let me put some numbers on this, because the gap is wider than most people assume.
Microsoft and LinkedIn's 2024 Work Trend Index found that 75% of knowledge workers say they use AI at work. But the same research found that 78% are bringing their own AI tools - meaning the organisation hasn't formally deployed or governed them. Salesforce's 2024 research found that while 67% of IT leaders said their organisation had an AI strategy, only 28% had implemented AI in any customer-facing application.
The pattern is consistent across every piece of research I've reviewed: stated adoption significantly exceeds measured, governed, productive adoption. Firms claim AI programmes that, on examination, consist of a handful of individuals using consumer AI tools without oversight, governance, or measurement.
This matters for you specifically. If you're a managing partner who's been feeling the pressure - everyone seems to be using AI, we're falling behind, we need to accelerate - the honest picture is more nuanced than that internal monologue suggests.
You are probably further behind than the conference circuit implies. And you are probably less behind than you fear.
The honest average for a mid-market professional services firm right now is: some individual AI tool use (mostly ungoverned), minimal formal AI governance, and limited evidence of productivity improvement at the firm level. If that sounds like your firm, you're not an outlier. You're the norm.
That's not a reason for complacency, mind. The gap between the average and the firms that are getting this right is a genuine competitive gap that will widen. But it is a reason to make decisions based on your own reality rather than someone else's conference presentation.
When we run AI readiness assessments, we typically see firms fall into one of three places. The recommended actions are genuinely different for each.
You're actively experimenting. Some individuals in your firm are using AI tools. You're exploring use cases. Maybe you've run an informal pilot or two. But there's no governance, no defined success criteria, and no measurement. The move here isn't a strategy document - it's formalising what you're already doing. Put basic governance in place: clarity on what tools are approved, what data can be shared with them, and who's responsible. Define success criteria for the two or three most promising use cases. Run a structured pilot with a 90-day deadline and a named owner. That's it.
You're watching but not acting. You're aware of AI. You've read the articles (including, presumably, this one). You may have attended a webinar or two. But nothing's been deployed. Start with the practical use cases specific to your sector - document drafting, research summarisation, routine reporting. Pick one. Launch one pilot. This quarter. Not next quarter. This one. The longer you wait, the more the gap compounds.
You're ahead of the average. You've deployed AI tools with some governance and you have measurable results. Good. The priority now is expanding from individual productivity tools to firm-level workflow automation - which requires infrastructure, integration, and a level of data readiness that most firms haven't addressed yet. Our guide on agentic AI covers the multi-agent architecture that makes that kind of firm-level automation possible, if you want to understand what that infrastructure investment actually involves.
If you're not sure which category your firm falls into - or you suspect different parts of the firm are in different categories, which is common - we've put together an AI adoption self-assessment that maps five adoption indicators against these three maturity levels. It takes about ten minutes and it'll give you a clearer starting point than any conference panel.
The firms that will gain the most from AI aren't the ones moving fastest or talking loudest. They're the ones moving deliberately - starting with the mundane use cases that produce measurable results, building governance before scale, and making investment decisions based on their own reality rather than someone else's conference presentation.
The gap between talk and action isn't a problem. It's an opportunity. While your competitors are discussing what they're going to build, you could be quietly deploying what actually works.
And in twelve months, when the conference circuit has moved on to the next shiny thing, the firms with working AI tools, trained users, and measured outcomes will have the genuine competitive advantage. Not because they were first. Because they were honest about where they started.
This article was published in 2025. Given how quickly the AI adoption landscape is evolving, we plan to review and update the data and recommendations in this piece every six months.