Fifty-one percent of companies already have AI agents in production. You've probably seen that stat, or one like it, floating around LinkedIn. And if you're a managing partner or CTO at a mid-market professional services firm, there's a decent chance it made you feel slightly behind.
So you had a conversation. Maybe with a vendor, maybe with a consultant, maybe with that one partner who keeps sending articles to the group chat. And somewhere in that conversation, someone said something like: "You need to get your data house in order first." Or: "You really need a proper AI strategy before you do anything."
That advice isn't wrong, exactly. But it's the kind of advice that sounds responsible while quietly ensuring nothing ever happens. Because "get your data in order" is not a task with a finish line. It's a direction of travel that can absorb twelve months and six figures before anyone thinks to ask what you're actually preparing for.
I've written about what AI practically means for firms like yours in a separate piece - worth reading if you're still getting your head around the landscape. But this article is for the people who've moved past curiosity and into the question of: where do we actually start?
The biggest misconception I see - and I see it constantly - is that AI readiness is a binary state. You're either ready or you're not. The whole organisation needs to hit some undefined threshold before anyone's allowed to do anything interesting.
That's not how it works.
We think about AI readiness as a profile across four dimensions. Not a single score. A profile. Because a firm can be strong in two areas and weak in two others, and that's not a reason to wait - it's a reason to start in the areas where you're strong.
I should say: we didn't arrive at these four dimensions by sitting in a room and drawing a framework. We arrived at them by getting it wrong a few times. Early on, we'd do an initial conversation with a firm, decide their data quality was the main blocker, and point them toward a data cleansing programme. Six months later, they'd come back with cleaner data and the same problem - because the actual issue was that nobody in the firm trusted AI outputs enough to act on them. Cultural willingness. We'd missed it entirely. So the four dimensions are less a framework and more a list of the things we've watched firms trip over.
Data quality. Does the organisation have reliable, consistent, accessible data in the areas where you'd want AI to operate? Not perfect data - I want to be clear about that. Data that's clean enough to produce outputs that are more useful than harmful. If your client records are reasonably up to date and your document management system has some structure to it, you might be in better shape than you think. If your data lives in seventeen spreadsheets, three legacy systems, and someone's Outlook folders - well, that's a different conversation.
Systems integration. Can AI tools actually connect to the systems and data sources they need? This isn't about whether you have the most modern tech stack. It's about whether there are APIs, export options, or integration pathways that make connection possible without a nine-month infrastructure project. Some firms have surprisingly good foundations here because they invested in a decent CRM five years ago and forgot about it. Others are running platforms that are essentially sealed boxes - nothing gets in, nothing gets out.
Cultural willingness. This is the one people underestimate. Do the people who'd actually use AI tools understand what those tools do? Do they trust them enough to incorporate them into their work? And - this is the bit that really matters - do they have the skills to supervise and correct the outputs? An AI tool that generates a first draft of a client proposal is only useful if the person receiving that draft knows enough to spot when it's hallucinated something or missed the point entirely. Cultural willingness isn't about enthusiasm. It's about informed confidence. I've sat with teams who were genuinely excited about AI but treated every output as gospel. That's not readiness. That's a liability.
Governance maturity. Does the organisation have oversight structures that ensure AI outputs get reviewed before they have consequences? That errors are caught? That there's a clearly defined boundary between what's automated and what requires human judgement? For regulated firms - and most of our clients are in regulated industries - this one carries particular weight. But even for unregulated businesses, the question of "who's accountable when the AI gets it wrong?" needs an answer before you scale anything.
The relative weight of each dimension shifts depending on what you're trying to do. An internal knowledge management tool has very different governance requirements than something client-facing. A document summarisation tool needs solid data quality but minimal systems integration. The profile matters more than any single dimension, which is why "just fix your data" is usually the wrong starting point.
You don't need us - or anyone - to tell you roughly where you stand. Here's a set of questions for each dimension. Be honest with yourself. Nobody's marking this.
- Could you pull a clean, accurate list of your top 50 clients and their key contacts within an hour?
- When your team creates reports or proposals, do they generally trust the data they're pulling from, or do they routinely double-check it against other sources?
- Is your document management system structured enough that you could search for all advice given to clients in a specific sector over the past two years?
- Does your CRM talk to your email and document management systems, or do people copy information between them manually?
- When you've tried to connect two systems in the past, was it a reasonable project or a nightmare?
- Could a third-party tool access your key data sources through an API or structured export, or would it require custom development?
- Have any of your team used AI tools - ChatGPT, Copilot, anything - in their work, even informally, in the past six months?
- If you announced an AI pilot next month, would the reaction be mostly curiosity or mostly anxiety?
- Are the people who'd use AI tools understand the concept of checking and correcting outputs, or would they treat them as authoritative?
- Does your firm have a documented policy on AI use, even a basic one?
- Is there a named individual accountable for how AI tools are used in the firm?
- If an AI tool produced an incorrect output that reached a client, do you have a clear process for how that would be handled?
If you scored yourself honestly, you've probably got a rough profile. Strong in some places, weak in others. Maybe your data's decent but your governance is non-existent. Maybe your team is culturally willing but your systems are a mess. That profile is useful. It tells you where to start.
If you want a more structured version of this, we've built a self-assessment checklist that covers all four dimensions with a simple scoring methodology and a "your profile suggests" output pointing toward specific starting points. It's designed to be useful without our involvement.
I want to address three pieces of advice I hear repeated constantly - usually by people with good intentions, occasionally by people with something to sell.
"You need perfect data first."
No, you don't. You need data that's good enough for the specific application you're considering. A firm that wants to use AI for internal document summarisation needs its document management system to be reasonably structured and searchable. It doesn't need a comprehensive data cleansing programme across every system it owns. McKinsey has noted that poor data quality costs organisations an average of $15 million annually - but the answer isn't to boil the ocean. Fix the data that matters for the thing you're trying to do. Most firms I speak to have at least one area where their data is good enough to start. They just haven't been given permission to think about it that way.
"You need a Chief AI Officer or a dedicated AI team."
For a 150-person consulting firm? Probably not. What you need is one person with clear accountability for AI governance - someone who owns the question of "are we using this responsibly and effectively?" That can absolutely be an existing role with expanded scope. Your CTO, your head of innovation, your COO - whoever has the combination of seniority and curiosity. The moment you create a dedicated AI team in a mid-market firm, you've created a silo. And silos are where good AI initiatives go to die, because the people who understand the work and the people building the tools end up in different rooms having different conversations that never quite connect. Gartner has flagged that by 2026, organisations that operationalise AI transparency and governance will see their AI models achieve 50% better adoption - but governance doesn't require a new department. It requires clear ownership.
"Start with your biggest opportunity."
This one I feel strongly about. Start with the safest, not the biggest. The application that carries the lowest risk if it goes wrong, while your organisation is still developing judgement about when to trust AI outputs and when to override them.
But we want to show real impact quickly - that's how we get buy-in.
I get that. But if your biggest opportunity is client-facing - automated compliance advice, say, or AI-generated client communications - and something goes wrong during week three of your first pilot, the damage isn't just operational. It's reputational. And it'll set back AI adoption in your firm by years, because the partners who were already sceptical will have their evidence. I've watched this happen. A firm I spoke to last year went straight for a client-facing document generation tool. The output was wrong in a way that was subtle enough to get past a tired associate on a Friday afternoon. The client noticed. The partners noticed. The AI programme was quietly shelved for eight months.
Start with internal tools. Document summarisation. Research synthesis. Proposal drafting. Things where a human reviews every output before it goes anywhere. Build confidence, build competence, build the governance muscle - then expand.
I'll be direct here, because if you're reading this far you're probably evaluating whether to do something, not just reading for interest.
Honestly, most of these take us about two to three weeks from kick-off to recommendations. Sometimes less if the firm's got its act together. It's not a six-month consulting engagement. It's not an 80-page report that sits on a shelf - I've seen enough of those to know what happens to them.
The core of it is structured conversations with the people who own data, systems, culture, and governance - your CTO or IT lead, your COO or operations director, a cross-section of practitioners who'd actually use the tools, and whoever's responsible for risk or compliance. We look at existing documentation and data samples - not to conduct a full data audit, but to check whether what people tell us in interviews matches what we can actually see. They don't always match. We benchmark against comparable firms where we can, because context matters - a firm that thinks its data quality is poor might actually be in the top quartile for its sector.
What comes out the other end is a prioritised readiness profile across the four dimensions, a set of "start now" recommendations for the areas that are ready enough, a set of "prepare here" recommendations for the areas that need development, and an honest view of which AI applications are within reach in the next twelve months and which need more groundwork first.
A readiness assessment doesn't produce an AI strategy. It produces the foundation on which an AI strategy can be built. Our guide on agentic AI at work goes deeper into the implementation sequencing if you want to understand what comes after.
I sat with a CTO last month - 300-person firm, professional services, good data in their CRM, decent document management system, team that was genuinely curious about AI. They'd been told six months earlier by another consultancy that they needed to "develop a comprehensive data strategy" before pursuing any AI initiatives. Six months later, the data strategy wasn't finished, no AI work had started, and three of their competitors had launched internal AI tools - one of which had already cut proposal turnaround time by about 60%.
That firm wasn't unready. They were ready in two of the four dimensions and could have been running a useful, low-risk pilot within 90 days. The advice they received was technically defensible - yes, a data strategy is good to have - but practically counterproductive.
We're not ready yet is the most expensive sentence in professional services right now. Not because every firm should rush into AI. But because treating readiness as a single, binary threshold keeps good firms on the sidelines while their competitors learn by doing.
Partial readiness enables partial adoption. You don't need all four dimensions at full strength to start. You need enough strength in the right dimensions for the specific application you're pursuing. And the act of starting - running a real pilot, with real data and real people - is itself the fastest way to develop the dimensions where you're weaker. You learn more in 90 days of a contained pilot than in six months of strategy documents.
If you want a structured way to assess where your firm sits across the four dimensions before deciding whether a formal assessment makes sense, download the self-assessment checklist. And if you'd rather talk through your situation directly, book a readiness conversation - we'll give you a direct view of where your firm is and what the realistic starting points are, without a proposal.