Let me save you eighteen months.
That's roughly how long the average mid-market law firm spends between "we should probably do something with AI" and actually doing anything. Eighteen months of conference panels, partner away-day discussions, vendor demos that all blur into one, and a working group that meets three times before quietly dissolving. I've watched it happen at enough firms that I could set my watch by it.
I was on a call a few months back with the managing partner of a 280-person firm in the Midlands. Sharp guy, genuinely curious about AI, clearly frustrated. He'd been "exploring the space" for about a year. Had a shortlist of vendors. Had a steering group. Was waiting for the regulatory picture to settle before committing to anything. Meanwhile, a competitor firm - smaller, less well-resourced - had quietly deployed an AI-assisted document review tool, run it for two quarters, and was already on their third use case. The managing partner didn't know this yet. He found out three weeks later when a client mentioned it in passing.
And while you're designing your AI strategy, appointing your AI committee, and debating whether you need a Chief AI Officer (you don't, by the way), the firms that started small six months ago are already on their second and third use cases. They didn't wait for perfect clarity. They picked something specific, ran it for a month, measured what happened, and made a decision.
But AI in legal is complex. The regulatory picture isn't settled. We can't afford to get this wrong.
I hear you. And you're right that the stakes are real - this isn't a sector where you can move fast and break things. But the firms using regulatory uncertainty as a reason to wait are confusing caution with inaction. They're different things. Caution means starting with well-scoped, low-risk use cases and building governance around them. Inaction means doing nothing and hoping the world slows down to match your pace. It won't.
So here are three specific things your firm can do this quarter. Not theoretical. Three use cases that fit the technical capability and data maturity of a typical 300-person law firm right now, with realistic timelines, honest resource requirements, and clear ways to measure whether they're working. I've written companion pieces covering the same exercise for financial services firms and management consultancies - the approach is the same, but the use cases are sector-specific.
Start here if you've never done anything with AI before. It's the most mature, the most evidenced, and the one where the risk-to-reward ratio is most obviously in your favour.
What it does: automated first-pass review of document sets for relevant clauses, defined terms, risks, and issues. Instead of a junior associate spending three days reading through a data room, the AI does the initial sweep and flags what needs human attention. The associate then reviews the flagged items, validates the AI's work, and focuses their expertise where it actually adds value - on the judgement calls, not the page-turning.
Published research from firms using these tools suggests time savings in the range of 40-70% on initial document review, depending on document volume and complexity. That range is wide, I know - the honest version is that it varies significantly by document type and how well the system has been calibrated. At one firm we worked with, initial review time on a mid-sized due diligence exercise dropped by around 55% after the first six weeks. That's not a small number. On a deal with meaningful document volume, you're potentially giving fee earners back days of their week.
The tools in this space - Luminance, Kira, Harvey, and others - have been around long enough that there's a genuine track record. This market moves quickly, and by the time you read this the specific options may have shifted. Check what's current. But the category is well-established and the underlying capability is proven.
What it requires: a representative sample of documents for initial calibration, a fee earner willing to supervise and validate the first ten to fifteen outputs, and a clear quality assurance protocol before any AI-reviewed output leaves the firm. That last point isn't optional. Your professional indemnity insurer will want to know about it, and frankly, so should you.
One thing I'd add from experience: don't let the calibration phase be treated as admin. We ran a pilot for a corporate team last year where the supervising associate was brilliant technically but kept getting pulled onto live matters during the calibration window. The pilot dragged on for three months instead of six weeks, the data was patchy, and the results were inconclusive. The technology was fine. The time protection wasn't. Get that agreed upfront.
Realistic timeline: proof of concept within four to six weeks. Expanded use within a quarter.
What to measure: time saved per document review cycle, accuracy rate versus manual review on a validation sample, and - this one's often overlooked - fee earner confidence score after four weeks of use. If the people using it don't trust it, adoption will stall regardless of what the numbers say. I've written separately about how to build confidence in AI without overpromising, which covers exactly this dynamic.
How do fee earners at your firm currently find a relevant precedent?
If the honest answer is "they ask Sarah in corporate because she's been here nineteen years and knows where everything is" - and at most firms, that is the honest answer - then this use case is for you.
Every law firm I've worked with has the same problem. Decades of excellent work product sitting in a document management system that's functionally a digital filing cabinet with a terrible search function. Fee earners know the knowledge is in there somewhere, but finding it takes so long that they either recreate work from scratch or walk down the corridor and ask someone who might remember.
AI-powered search changes this. It sits on top of your existing document management system and enables fee earners to find relevant precedents, prior work, and internal guidance using natural language queries rather than navigating folder structures. "Find me an indemnity clause from a SPA we did for a manufacturing client in the last two years" instead of clicking through seventeen folders.
If your firm is in the Microsoft ecosystem - and most mid-market law firms are - Microsoft Copilot for SharePoint and Teams is the obvious starting point. If you're on iManage or NetDocuments, both have AI-powered search capabilities worth evaluating. As with document review tools, check what's current at the point you're reading this.
Now, here's where I need to be straight with you. This use case has a dependency that might slow you down: the quality of your document management. If your DMS is well-organised and reasonably well-tagged, you can have a pilot running in one practice area within three to four weeks. If your DMS is - and I'm being diplomatic here - a bit of a mess, you'll need to do some groundwork first. That doesn't mean you can't do it this quarter. It means you should be realistic about which practice area to pilot in. Pick the one with the best-organised documents, not the one with the loudest partner.
The change management piece matters too. Fee earners have spent years developing their own workarounds for finding information. Asking Sarah is fast, familiar, and comes with context. You're asking people to change a habit, which requires more than sending a "we've launched a new search tool" email. Budget time for training, for visible senior adoption, and for someone to be the go-to person when the system returns an unexpected result.
I had a conversation with a head of knowledge management at a top-50 firm a while back. She'd spent six months getting a knowledge search tool technically right - genuinely impressive setup, well-integrated, fast. Adoption after three months was 11%. The problem wasn't the tool. Fee earners didn't know it existed, the one training session had been optional, and no partner had been seen using it publicly. She had to essentially relaunch it with a proper change programme. The second time, adoption hit 60% within eight weeks. Same technology. Completely different result.
What to measure: search query volume, time from query to found document, and adoption rate among fee earners in the pilot group. If adoption is below 30% after four weeks, the problem is almost certainly change management, not technology.
This is the one that makes managing partners nervous. And honestly, it should - a bit. Any use case that touches client communication in a law firm needs more careful design than the first two. But dismissing it entirely means leaving significant efficiency on the table while your clients wait longer than they need to for answers to straightforward questions.
Think about the enquiries that consume disproportionate fee earner time but deliver minimal value. "Where are we with my matter?" "Can you send me the latest draft?" "What does this term mean in the context of our agreement?" Legitimate questions, all of them. But they don't require a senior associate to spend twenty minutes composing a response from scratch each time.
What this looks like in practice: a supervised AI response system for defined enquiry types, with clear escalation to a human for anything outside set parameters. And I want to emphasise the word supervised. This is not a chatbot left to its own devices. Every response should be reviewed by a fee earner or PA before sending for at least the first three months. Full audit trail. Clear escalation triggers. No ambiguity about where the AI's role ends and the human's begins.
The SRA's current position on AI-assisted communication is something you need to review before implementing this - check the SRA Technology and Innovation page and any published guidance notes directly, because this area moves quickly and my summary of it is no substitute for reading the actual document. The direction of travel has been pragmatic rather than prohibitive. They're not saying "don't use AI." They're saying "use it responsibly and maintain professional standards." But involve your COLP in the design from day one, not as a sign-off at the end.
What it requires: a defined list of enquiry types the system will handle - start narrow, five to ten types maximum - a response template library for initial training, and a named person responsible for review and escalation. The compliance design is the critical part. Get that right and this becomes a genuine differentiator. Get it wrong and you've got a problem that no amount of technology can fix.
Realistic timeline: this one takes longer to set up properly. Allow six to eight weeks for design and compliance review before any client-facing output. But once it's running, the time savings per enquiry are meaningful, and client satisfaction often goes up because response times drop dramatically. We saw this at a financial services client running a similar setup - not a law firm, so the compliance context differs, but the pattern holds. Clients don't care that a human didn't draft the initial response. They care that they got a clear, accurate answer in two hours instead of two days.
What to measure: time saved per enquiry, client satisfaction scores on a sample, and escalation rate as a proxy for system accuracy. If more than 25% of enquiries are escalating, your parameters are too narrow or your training data needs work.
Let me be blunt about the investment. Software licensing for tools in this space typically runs from a few hundred to a few thousand pounds per user per year depending on the platform and the use case. Integration work varies significantly based on your existing setup - a firm already in the Microsoft ecosystem with a well-maintained DMS will spend considerably less than one running fragmented legacy systems. A short advisory engagement to help with design and governance adds to that. For a 300-person firm running all three pilots, you're probably looking at somewhere between £80k and £150k across the first year, though that range shifts depending on your starting point. The honest answer is that a proper scoping conversation will tell you more than any figure I can give you here.
But the money isn't the hard constraint. The hard constraint is people. Each use case requires at least one fee earner or operations lead willing to commit real time to calibration, supervision, and honest reporting. "Real time" means four to six hours a week during the pilot phase, not "have a look at it when you get a chance." If you can't identify those people - and more importantly, if you can't protect their time - the pilots will drift, the data will be inconclusive, and you'll end up back in the "AI doesn't really work for us" conversation twelve months from now.
The firms that make this work treat it like any other client matter: named lead, defined scope, agreed timeline, regular reporting. The firms that don't treat it like a side project and get side-project results.
Don't try to run all three simultaneously. Rank them against your firm's current readiness. Which one requires the least data preparation and the least change management for your specific situation? Start there.
Run a four-week pilot with success criteria agreed before the pilot begins. This is the bit that almost everyone skips, and it's the bit that matters most. If you haven't defined what success looks like before you start, you'll spend the review meeting arguing about whether the results were good enough rather than deciding what to do next. I've sat in that meeting. It's not a good use of a Tuesday afternoon.
Report the results honestly to the management team - positive, negative, and inconclusive. Especially inconclusive, because "we don't know yet" is useful information that tells you what to test next. I've written about what using AI in your business actually means for readers who want the foundational context before diving into specific use cases.
If you want to understand which of these three use cases is the best fit for your firm's current data quality, technical capability, and risk appetite - and what a four-week pilot design would look like - book an AI opportunity assessment. We've also built an AI readiness scorecard for legal services that maps each use case against four readiness criteria: data quality, technical environment, fee earner readiness, and compliance posture. It takes about ten minutes and gives you something concrete to share with your partnership team before the conversation about investment.
And if you want to build the internal case for approving these pilots and governing them through to a result, there's a governance framework that gives your steering group the structure to move quickly without losing oversight.
The window for "we'll get to it eventually" is closing. The firms that start this quarter - even imperfectly, even with just one use case - will be meaningfully ahead of the ones still designing their AI strategy in twelve months' time. The first step is a lot less dramatic than most managing partners expect. Pick one. Run it for a month. See what happens.