Now what?

Foreword

In July, we published “Ready or Not?” This was a report that came from 25 conversations with in-house lawyers about what AI was actually doing to their work. I was surprised to learn that, though almost everyone was using it, very few were actually working less.

The response to the report was more generous than we expected, and it contained one piece of feedback I couldn’t put down. It came from Jessica Vander Ploeg, Vice President of Legal Operations at Belron:

“One observation I’d offer: the contributor list is strong on the in-house counsel experience, but the legal operations voice is missing. That’s the voice of the technology itself, the roll out, the implementation and the change management. It’s the difference between buying a tool and getting a team to use it, and it’s often where the real value is won or lost.”

She was right. We had spent hours asking lawyers what AI had done to their week, and almost no time speaking to the people who make it work.

So I went and found 28 of them. A mix of legal operations professionals, and in-house lawyers who had recently secured budget and gone through a buying process.

The difference between the two sets of conversations was starker than I expected. Across all the interviews with in-house lawyers for the first report, not one person raised change management. In the conversations for this one, almost everybody did, usually in the first ten minutes and before I’d even asked.

I expected them to talk about tools. Mostly they told me the tools had stopped being the interesting part. Ummu Fallon, the only legal operations person at Camunda, ran the most thorough evaluation anyone described to me: a wishlist whittled to ten vendors, five taken through a proof of concept. Her deciding criterion still wasn’t a feature. “I want to be sure that this is something that the company will actually use, not something that will just be brought in because it has all these amazing features.”

That is a slightly uncomfortable finding for a company that sells software. If the technology isn’t the variable, everything that decides the outcome is human, and nearly all of it happens after the purchase order.

There was a second thing they kept saying: automation is not new. Power Automate has sat inside every enterprise Microsoft licence for a decade, and a good deal of what teams are now excited about could have been built with it years ago. What AI changed is not the capability. It changed the barrier, and made automation easy enough that anybody can do it.

That is mostly a good thing, and this report is full of people doing it. But it also means a lot of building is now happening without anyone asking whether it should, or looking at it afterwards.

Successful implementation is the territory Jessica said was missing. It’s also the least documented part of legal technology.

So this is a how-to, not a review. It follows the six stages a team goes through: working out what you’re solving for, checking what you already own, choosing who to buy from, making sure it gets used, keeping it alive long enough to prove it worked, and deciding who owns it. Every stage is built from what these 28 have done, including the parts that didn’t work.

They don’t all work the same way. One ran a gradual rollout with no hard launch and no deadline. Another ran seven parallel workstreams alongside hers and tracked usage person by person. Both ended up in the same place, which is why this isn’t a playbook. What works depends on the size of your team and what your business will tolerate.

Ready or Not? ended with a ladder: contracts scattered, then gathered, then in order. Several people pointed out, kindly, that we stopped a rung early.

This report is about that rung.

The numbers first

Suzanne Jeffers has seen legal technology rollouts from both sides, as a practising solicitor and from inside a legal tech business. She once led the introduction of a case management system at a firm that had never used one. Every head of department was consulted, and the firm chose the product together. One senior partner sat the process out, expecting to have retired before it went live. “They weren’t, and when the system went live, they simply refused to use it.”

The firm had done the selection properly. “None of that guarantees adoption,” she says. She warns that organisations can end up “using the latest version of a product without extracting anything close to its full value”.

“Implementation is an event. Adoption is an ongoing process and getting the most out of technology requires a commitment to both.”

The people we interviewed told similar stories.

17 of 28 interviewees described a rollout that stalled or failed

Interviewees 17 of 28

Described a rollout that stalled or failed

Seventeen of the 28 people we interviewed described a rollout they had seen stall or fail, in their own team or in teams they work with. Most saw it happen inside their own organisation.

Three of them got their tools working in the end. None of them was quick.

77% have been through a failed legal tech rollout

77%

A larger survey puts the figure higher. Of 350 in-house lawyers and paralegals surveyed in the US and UK in 2022, 77% said they had experienced a failed technology implementation. 43% said they had been through more than one. For CLM in particular, Gartner predicted that by 2024 nearly half of initial implementations would fall short of expectations. Fast forward to today, and the numbers in this report suggest that this may have been an underestimate.

Only 24% can find all their contracts in one place

Yes 24%

“I can find all contracts in one place”

In the Ready or Not poll of 85 in-house lawyers, only 24% could find all their contracts in one place. Several of the rollouts we heard about were contract tools, from CLMs to review software.

93% use AI. Only 38% work fewer hours.

Yes 93%

“I use AI in my legal work”

Yes 38%

“Since using AI, I am working fewer hours”

The same poll found that 93% use AI in their legal work. Only 38% said they had worked fewer hours since they started. The six stages that follow are about that gap.

1. Work out what you’re solving for

Allan Smith spent twenty years at ITV and four of them running legal operations. He now advises in-house teams independently as Mint Consulting, and he says clients arrive in one of four states: looking to buy tech, fix tech, resurrect tech, or having been told to do something about AI. The fourth is increasingly the common one.

“Your CFO has told you to do something with AI because they think Legal is a good kind of training ground for that stuff,” he said.

It’s a bad place to start, but almost everyone starts there anyway.

Michael Phillips, Head of Legal at Lloyds Wealth, bought his team’s first AI tool in 2023. He went through a law firm to do it, because he was worried about the legal risk of buying direct.

“What I mainly learned was I’d thrown all of my hard earned basic lawyering skills out the window in an AI frenzy. I’d forgotten about the listening, the problem solving, the scoping, the spec’ing, and just went off and bought something and lo and behold, it didn’t work.”

He thinks it’s the same reason other GCs see tech fail: “They probably bought a tool rather than tried to understand what their pain point was. You start with the tool and then think that there’s a problem somewhere that that tool will solve. But it doesn’t. It just creates another problem.”

Quentin Peltier, who runs the legal operations consultancy Dr. Ops, doesn’t blame the vendors.

“Buying a tool feels like progress, and documenting a workflow doesn’t,” he said. “The tool becomes a way to skip the uncomfortable part. Six months later, you’ve automated the chaos, and with AI, you’ve automated it faster.”

He starts by turning the complaint into a number: “Name the pain in the users’ words, not the vendor’s. ‘Contracts are slow’ is a feeling. ‘Sales waits nine days for an NDA’ is a problem.”

Then he measures it. Volume, time, who is involved, because “without a baseline, you’ll never prove a tool worked.”

Emma Levin, Group Legal Counsel at Onebright, a mental health business, did a rougher version of the same thing before she looked at a single product. Her team is three people (two job-sharing), with no legal technology at all. She and her colleagues sat down and wrote out everything the team does. It came to two pages. Then they went down the list marking each line yes or no: could AI help with this one?

“There’s so much chatter about legal AI and legal tech on LinkedIn. But we asked: what actually of our day to day stuff could we use AI for?”

The exercise took an afternoon and produced a single number: at least half of what the team did could be done more efficiently, and that was the number she took to her CFO.

Where Emma started with a list of everything her team does, Mint Consulting’s Allan Smith starts with a single question: does the client have a list of their contract types? Most don’t, and he treats that as a readiness test. The moment you ask software to enforce anything, somebody has to decide the rules in terms a machine can apply, and a list of contract types is the smallest version of that. If a team can’t produce one, the rules underneath it don’t exist either. Which is why his first month with a new client is usually spent nowhere near a tool.

Instead it’s spent working out what they want to achieve, what any tool would have to do, what data they want out of it, and why that matters. He admits it is dull work and that nobody is excited by it, but reckons it routinely saves a purchase.

“People will say, we want to go and buy a CLM, we’ve been quoted 50 grand for it, we’re not sure what to do with it. And my answer is usually: don’t spend the 50 grand just yet.”

Richard Tapp, who co-founded NuCAS, which provides legal services and consultancy to in-house teams, would push the question back one step further. A former FTSE 100 General Counsel who built legal teams globally and first implemented legal tech 20 years ago, he says:

“Critically, before a team works out which tool to buy, it needs to decide what work Legal should be doing, and whether an issue landing on the team needs a lawyer at all. If it doesn’t need a lawyer, no legal purchase is going to fix it.

“Legal teams handle business problems which happen to have a ‘legal’ label, because lawyers are great at problem-solving, but that doesn’t mean they should be solving every issue that hits them.”

His argument is financial as much as operational. The business is likely to be paying a lawyer more than it would pay whoever ought to be handling the work; often the lawyer dealing with it feels under-valued, and eventually leaves and has to be replaced.

What he recommends in response isn’t a product but a collaboration with the business: make what Legal should be doing a shared question rather than simply the general counsel’s. He adds that before any process can be automated, Legal has to know what the process is, and the problem with that is the profession tends not to document its processes.

After his initial outsourcing failure, Lloyds Wealth’s Michael Phillips went back and did the work properly. He spent real time simply sitting with the sourcing and supplier management teams and listening. What they wanted was not sophisticated. “Binary choices, markup. So simple, simple, simple.”

Onebright’s Emma Levin agrees that legal teams tend to be basic in how they’re run. “We just want a good to-do list.” The lesson is that a lot of the time, solutions to in-house problems don’t require a budget, a vendor, or even permission.

2. Check what you already own

Before going to market, it is worth knowing what you already own. Most organisations pay for far more capability than their legal team realises, and a good deal of it will already do the job that is about to be bought.

Sarah Owens is the only legal operations person at Barratt Redrow, supporting a legal and compliance team of thirty-two. Her background is IT consultancy (the translator role between the business and the technical function), and her first move is to go and find somebody in the organisation’s own IT department.

“It’s a hill I will die on. If it’s already existing, make the most of it before you start throwing money at other things.”

What she specifically doesn’t want teams to waste budget on is workflow. “There’s an awful lot of talk at the minute about workflows, and it’s because of AI. But Power Automate turned up ten years ago.”

She is not claiming it is a pleasure to use; the older builds are clunky, and it takes work. Sarah’s argument is that clunky and already licensed will get you a working proof of concept, and a working proof of concept is what gets you the budget for something better. Most teams try to do it the other way round and stall at the business case.

A great example of bootstrapping comes from Randstad Enterprise, part of the staffing group Randstad, where an NDA self-review tool was built in-house using Gemini and Google Apps Script: not by the legal operations team, and not by a vendor, but by the General Counsel himself. Mykim Truong, who runs legal operations there alone for a team of seventeen, says work that used to be allocated to a named lawyer is now handled end to end, and only surfaces when something is wrong.

“That is stuff that used to be allocated to team members. I now don’t even have to touch those any more. They are fully automated unless they need to be escalated. That saves us procuring a tool for that. He created it himself.”

Neeharika Nakka, the only lawyer at Fibe, a thirteen-person materials startup, also had to see what she could do with Gemini. While some of the more established legal research databases and specialist subscriptions (such as LexisNexis) may not necessarily be proportionate for a legal team of one at this stage, she is exploring where AI can help streamline her workflows and where more specialist legal tools could add value.

“We are currently under a Gemini subscription for the entire organisation,” she said. “So I’ve been exploring how I can tailor it to support legal work, including creating more specialised workflows and giving it the right legal context and inputs. At the moment, it’s about making the most of the tools already available while applying my own legal judgement to the output.”

Neeharika is comfortable with the compromise for now. Specialist legal AI isn’t off the table; she is open to exploring more tailored tools where they can meaningfully streamline legal workflows and complement the tools already available across the business.

Natalia Cataife, who leads legal operations at Remote, would rather build tools herself than buy them. Her company has invested in training her on automation, Claude skills, Notion agents, and how to install a coding tool locally. She and two of her colleagues have direct access to the company’s codebase, meaning when an employment lawyer spots something needing an update, they can write the proposed fix themselves for engineers to review and merge.

“If I’m paying a hundred thousand for software that I can build in a month, I’ll do it.”

Lizzie Catterson, a Legal Operations Analyst at Omaze, agrees that building is now within reach. “I think AI has democratised the ability to build, which makes the build-versus-buy question far more nuanced than it used to be, especially for sole legal ops or leaner teams.” But she warns that the bill doesn’t stop once the tool is built.

“Buying means someone else carries the security, maintenance and support, while building means your team does, which is a cost in itself.”

Francesca East, Head of UK Legal & Commercial Services at Ensono, has a line for where a lot of legal technology sits today.

“We’ve got a lot of the A, but not a lot of the I.”

She says it about her own team, but it describes much of the market. Plenty of tools will automate a task. Far fewer deliver the insight they promise, because the data underneath isn’t ready. She sees the same gap in negotiations, where clients now send back AI-generated redlines that add noise without making anything better. “I think there’s a lot of activity that AI is generating, but not yet a lot of value.”

Her answer is to get the basics right before adding more automation, which means structuring the data and mapping the process first. That work isn’t urgent, she says, but it is important. “If you don’t put the priority of the time into it now, it will become urgent later on.” Teams that skip it end up “trying to scramble for the benefits that you can get without having done the grunt work”.

Her own team is doing that groundwork now, and she is clear the payoff comes later. “I know we will never get the dividends unless we go through the process of trying to architect our own solution and datasets in a way that enables the tool to do that.”

Even as a legal operations consultant, Mint Consulting’s Allan Smith thinks it’s no bad thing that teams sometimes have to work with what they’ve got. He compares the unnecessary use of AI to using a power drill to stir gravy.

“A lot of people are using AI for tasks that they should not be using AI for,” he said. “They could have been doing it through basic automation ten years ago, and it would be cheaper and more reliable, and you would know when it didn’t work.”

His objection is failure behaviour. An automation tool is deterministic. It follows rules, and it is either right or wrong. A model is probabilistic, doing what it calculates you probably wanted, which fails in a more subtle and more expensive way. “The endless emails to say ‘this is not working, go investigate’ are annoying. But at least you know why it hasn’t worked.”

His rule for deciding whether something needs AI or just automation is about the task, not the tool: “If you’re in drafting, quite a lot of it is deterministic. It’s putting this thing here, reading this thing, maybe somebody’s put some stuff in square brackets and you have to make a choice. You could have been doing that stuff in any kind of tool for ages… the minute you get into a little bit of judgment, things that are not black and white, like reviewing a third party paper, that’s where there is an advantage to these tools.”

Sarah Owens’ advice to start with your own IT department is sound, but how it plays out depends on the organisation. For Hugh Ndhlovu, legal counsel at Corona Energy in a legal team of two, IT is a partner with its own long list of priorities. Protecting the business means access to external AI models is restricted, and Hugh currently works with a more limited version of Copilot while the case for an upgrade is built.

“Legal, IT, and the business all have competing demands on the same budget,” he said. “Legal doesn’t generate revenue, so when we ask for new software or licences, the value isn’t always obvious to the rest of the business. It can look like a cost with no return. That isn’t a failing on IT’s part. It’s on us to show the benefit in terms the whole business recognises.”

So, he has been making the most of the tools already in place. He found that Microsoft Loop offered more workflow capability than the standard Copilot setup, and has used the two together to build a Risk and Regulation workspace. At its top sits a governing constitution, a set of core rules that every part of the workspace must follow. Beneath it is a hierarchy of role-based system prompts (standing instructions that tell the AI what role to play and how to approach a task), each designed around a problem the business needs solved: regulatory and licence conditions, vendor onboarding, complaints handling, and data privacy among them. Separately, he built a tracker for data subject access requests in a spreadsheet: reference numbers, a counter, and a countdown to the statutory deadline. Neither is a legal technology product, but both are working well.

That doesn’t mean he has given up on more advanced tools. His plan for the licence is to move from making the argument to demonstrating the value (more on that later, since it’s the subject of the fifth stage).

“My colleagues from big tech bought 70 licences… And they use two of them.”

3. Choose a partner, not a product

Every demo you will ever sit through has been engineered to succeed. That’s something worth remembering when assessing if a tool is likely to deliver real value, says Belron’s Jessica Vander Ploeg.

“Everything looks shiny. And in theory, demos should go perfect. So if everything’s shiny and all of them are perfect, how do you actually choose?”

Teams do get stuck there, and Sophie Jackson, Head of Legal Technology Consulting at Addleshaw Goddard, has a name for it: “Some teams come to us caught in pilot purgatory. They’re hands-on with different tools but tell us the features all feel very similar. Without the right structure, it’s hard to compare them and make a confident choice.”

Xavier Torres, Senior Paralegal at NuCAS, learned what was behind the shine in a previous job, where his team had bought on the strength of a whizzy demo.

As a managing paralegal he ran an NDA review, and the pitch he saw was a recorded call in which a contract went in one end and a clean, correct markup came out the other. The playbook was pre-loaded and the configuration was already done.

“They would say it was oven ready,” he said. “So they say, here’s one we prepared earlier; you hit start and here it is. But they weren’t upfront with the gritty elements of actually getting to that stage beforehand.”

Retrospectively, he estimates that roughly six hours of preparation sat behind that demo, and his own team would have needed about eighteen to reach the same starting line. His metaphor for the whole experience is a car dealership that sells you a fantastic car for getting from A to B and then, once you’ve agreed, mentions you’ll need to put the wheels on yourself.

Out of that comes what might be the single most useful piece of advice in this report: refuse the demo they’ve prepared.

“If I was going back on that call for the first time, I would say, okay, can you actually not do the pre-approved demo? Here is our playbook, and here is an NDA that another side has sent us.”

Make the document hard on purpose, says Xavier. The people running the demo are generally not lawyers, so their unaided attempt tells you what the product does on its own, stripped of the configuration you weren’t shown. Then ask the second question almost nobody asks: how long would it take us to get to that output?

Xavier’s overall read on the market is not cynical, although he remains to be impressed. “I am absolutely sure that there exists a product out there that can do as advertised. But none of them can do so as easily as they say.”

Claire Eldridge, Director of Legal & Compliance Operations at Quantinuum, blames the ubiquitous phrase “plug and play”:

“Plug and play does not exist, and they should stop saying it. They should say: we have an implementation plan.”

She puts most failed implementations down to a misunderstanding of how complex they are, and to “buying into the vendor zeitgeist of it’s plug and play”. What’s often missing is “getting that strategic plan first, before you even engage the vendor.”

Claire’s complaint has a structural cause, and Neil Willson, Director of Legal Operations at Cambridge University Press & Assessment, has thought about it harder than most. He contributed to the first report, where he put legal technology’s adoption problems down to a missing services wrapper around the products. Asked about it again, he put a number on it.

“All of the vendors out there have started on the point of view that their revenue model is licence based, it’s not services based. And actually the truth is, to get adoption, it’s probably more like a 50/50 revenue model. But nobody’s doing that. So if you haven’t then got the in-house staff in legal ops, and you haven’t got the services from the vendor, well, how do you get adoption?”

A team without legal operations is sold a licence and left to do the implementation itself, in time it does not have. So the challenge to put to a vendor isn’t only what the product does. It’s who is going to do the work of making it do that, and whose time that comes out of.

That gives a buyer something concrete to ask for. A provider that guarantees implementation, like Nomio does, has priced the hard part in. One that doesn’t has left it on your desk, and the licence will look cheaper than it is.

But while a contract can tell you what’s included, it can’t tell you whether they’ll be any good at delivering it. Jessica Vander Ploeg’s answer is to stop assessing the product and start assessing the humans behind it, using a framework she likens to dating.

Belron is currently choosing a matter management system. Jessica put together a task force of ten people covering every region, including deputies from several of them. It was the first time the company had let people outside legal operations into a global technology decision. Five providers were invited to demo.

She offered all five of them half an hour to an hour with her beforehand, to prepare. She told them: “I’ll give you everything you need to know. I’ll tell you everyone that’s in the room. I’ll tell you everything that’s gone well. I’ll tell you all of our skeletons in the closet, all of it.”

Only three took advantage of it, and the demos made it clear who had and who hadn’t.

“Even something as simple as that shows whether you’re willing to put in the time before a demo to be as prepared as possible.”

The framework isn’t only about preparation. Her green flags are behavioural, and all observable before you sign anything:

“How long are you going to sit and talk before you actually show me the technology? People pay attention to that. Are you willing to give me a pilot? People pay attention to that. Does the pilot have a cost? Can I choose how many people go into it?”

Her sharpest red flag is a vendor who reached the top five of a contract management tender a few years ago and then declined to fill the tender document in, having been burned by buyers who were never really buying. She has some sympathy for the position, but none for the way it was handled.

“None of that context was given. It was just: we’re not going to do your RFP. That’s an immediate red flag. I have no intention of marrying you because our dating experience has gone poorly.”

It isn’t only a metaphor. She thinks the vetting stage is where the outcome of most implementations is decided:

“So many people are willing to look at shiny technology, ignore the red flags from the team, buy the technology, not be prepared for change management, and then be confused as to why they’ve had an implementation absolutely fail. And a lot of it is, you’re not prepared. You don’t actually know who you’re buying from. You’ve not vetted them, and you’ve ignored all of the red flags they’ve shown throughout this dating experience.”

Michael Phillips from Lloyds Wealth reached the same conclusion the hard way, by buying badly the first time: “Most of the vendors had hired specialist salespeople who were very committed to selling you what they’d already built and showcasing how brilliant it was, which is probably why so many general counsel have bought the wrong thing. They bought the sales.”

He chose his second supplier because they spent the meeting listening rather than presenting.

One final piece of advice on finding the right technology partner is to eschew the cold emails they send you in favour of recommendations from peers actually using them. Emma Levin’s entire market research was one message into a WhatsApp group of small in-house teams, which drew about a hundred replies.

Katy Crane at GWI has recently been looking at tools for data subject access requests, and the few she has seen all came through recommendations. “It is good to know someone who has actually used it.”

Claire Eldridge would add one refinement: don’t accept the references the vendor hands you.

“Rarely trust a vendor’s reference,” she said. “Try and find an independent one, because they’re only ever going to put one forward that’s successful. They’re never going to say, oh yeah, speak to them, because they found it really difficult.”

She’d rather hear from the ones it went wrong for: “It’s as valuable hearing people whose processes failed as the ones that rave about it.”

4. Get people to actually use it

Barratt Redrow’s Sarah Owens has a test she applies before any rollout begins, and it disqualifies more projects than it approves.

“If nobody wants it; if there’s literally nobody in the department who’s interested in using this thing, put it down. Don’t start,” she said.

This doesn’t only apply to lawyers. Katy Crane is the only legal operations person in a legal team of eight at GWI, a human insights company. Her team’s first use of AI was deliberately small. The legal team was getting “inundated” with tickets from vendor onboarding questionnaires, asking for things like the company number. So the team fed the documents to an AI tool that answers those questions in Slack, because “our business sits in Slack most of the time”. It is still in use today, and has saved Katy and the legal team an enormous amount of time.

“We got that going first, and got the business used to using it,” she said. Only then did the team pick its next project. “When you say change management, I think it’s a lot to do with avoiding trying to do everything all in one go.”

Lucy Nixon, Chief Executive and co-founder of NuCAS, thinks a lot of this is settled by who is in the room when a rollout is designed, and that both of the obvious options fail.

“If you only include busy lawyers it might not happen,” she said. “If you only include legal operations and paralegals, the lawyers don’t see the benefits firsthand, and this hampers wider adoption.”

It is an argument for the sort of task force Jessica Vander Ploeg assembled at Belron, which deliberately reached outside legal operations.

Lawyers are the harder half of that mix to engage, and Sarah Owens says their resistance has a particular source: “I’ll be blunt. I think lawyers are really hard to enforce change upon. In any corporation there are certain people that are expected to just do as they’re told, and certain people that are supposed to bring value add… and when you take a resource like that, you can’t force them to change.”

Her point is that you cannot simply instruct a room of lawyers to try something, because trying things carries risk and they know it better than you do. So how do you get them to adopt a new tool?

Erin Ferguson runs legal operations at the Financial Times, across a legal, compliance and company secretarial team of around thirty. She isn’t a lawyer; she spent fifteen years elsewhere in the business, much of it close to technology, and arrived with what she describes as a keen interest in change management. Her rollout of Wordsmith had no hard launch, no mandate and no deadline.

“It was a gradual implementation. It is more of an evolution than an implementation,” she says, and it’s still evolving.

Everyone in her department was in the pilot, and the team stayed in Wordsmith while the contract was signed, so by the time it formally began, a handful of people were already using it heavily and everyone else dabbled. The FT has historically embraced different ways of working across its teams and business areas, so a tool flexible enough to accommodate them, rather than one that required a single mandated approach, was a selection criterion.

The rollout was personalised. Erin spent her time in one-to-ones asking two questions: what’s one thing you hate doing, and what’s one thing you really struggle with. Each week a few more people started using it regularly. Within three months, 90 per cent of the team were active users. It was a different approach, and she has been pleasantly surprised at how well it worked.

To bring it all together, the team holds a fortnightly AI session where anyone can show the room something they’ve worked out, including things in ChatGPT or Gemini, not only in the tool the FT had just paid for.

She calls it “a show and tell” and it works as a kind of social proof.

“Other people get FOMO, because they’re like, oh, they’re doing that, and I really wanted to do that. So we didn’t force anyone to use it. We didn’t mandate that anybody had to use it. But over time, everybody’s using it. Everybody’s found value in it.”

One of her colleagues, who has been at the FT for many years amassing a huge amount of institutional knowledge, now has several agents that she built herself: filling in due diligence questionnaires that arrive from vendors, helping with sanctions checks, handling admin she still reviews.

The knowledge that was previously stuck in her head is now in something other people can use. It’s something Erin calls “revolutionary”.

At GWI, Katy Crane’s legal team recently held an AI hackathon, and worked through their pain points spending a whole day building. Katy knows the lawyers are busy with their actual legal work. “It’s really good to dedicate time as a team and get everyone’s input.”

Belron rolled out the same tool. Like the FT, it kept pilot users in the tool while the contract was being signed, so that 60 per cent of the eventual user base was already familiar with it before day one. But where Erin’s rollout was personalised, Jessica Vander Ploeg’s was engineered. Seven separate initiatives ran alongside the launch, most aimed at power users: differentiated training, power users paired with beginners, a competition to submit and build use cases, and a learning series.

Rather than being dumped on people, communication was rationed and repeated, because “most people won’t read your emails the first, second, or even sometimes the third time.”

Eight months on, she reckons the team has not gone a fortnight without hearing from her about it. Monthly usage sits at around 98 per cent.

Christopher Allen, Assistant General Counsel at Safelite, Belron’s US business, saw two rollouts in quick succession. The first was a contract repository (Nomio), and everyone could already see the current state was bad.

“It was nearly impossible to find what you wanted,” he said. “So the pitch to the business team and executives was super simple. We can take this mess and make it not a mess. Everybody was instantly already on board. I didn’t have to emotionally or intellectually carry them along the journey.”

“With respect to the Nomio onboarding, it was white glove. They said, we don’t care how you get your documents to us. We’ll find them for you. We’ll go into your Docusign. They really did a fantastic job, holding our hand and doing all of the lifting.”

The second was the same Wordsmith rollout at Belron, where Christopher led the legal operations working group supporting Jessica, and the response was entirely different: “Wordsmith is how you do your work, and if you adopt it fully, it’s how you do everything. That change management piece carries a lot of emotional baggage. It carries a skill set that a lot of people just don’t have. It carries fear. Fear of job loss, fear of becoming obsolete, fear of ‘I don’t know what I’m doing’. Also fear of missing out.”

Nathan Corr, a legal engineer at Wordsmith, has seen the same barrier across the profession, at the Law Society of Scotland in a working group of partners, heads of innovation and legal engineers writing barriers to adoption onto sticky notes and grouping them on a wall.

“One of the biggest ones was fear. Everyone said it. Scared of change, scared of being obsolete. What shifts perspective is people seeing the tool take the work they don’t want to do, not their job.”

Christopher’s response to the mistrust across Belron’s global legal team, which Safelite’s lawyers are part of, was to stop trying to persuade anyone. With seventy people and sixty-five licences, he couldn’t, and neither could the legal operations team. So he built a menu and let people choose their own level: training sessions at fixed times, open to anyone; office hours you could drop into with a question and leave with the thing built; and a steady drip of messages that paired a problem with a solution rather than announcing a feature, each with a prompt you could copy and paste. He surveyed the team for what they wanted to be able to do, built it, and handed it back.

That handholding is important because what nobody likes to admit about a rollout is that, for a while, new tools are slower than the thing they replace. NuCAS’s Xavier Torres thinks it’s the reason so many collapse in the first month. His estimate for the tool his team was given in a previous role was three to four months before a paralegal would turn a document round faster with it than without it.

“We accept that it’s going to take initially far longer than if you weren’t using AI, but I promise it’s worth the time investment,” is what he thinks businesses should say.

Sometimes a delay in implementation can be caused simply by a lack of confidence. Quantinuum’s Claire Eldridge had a legal AI tool in place for eighteen months before she realised how cautiously her team was using it. It was because nobody had ever established what could safely go into it. So she went back to information security and asked for a full risk assessment, with one condition attached.

“I said I need a definitive yes or no. So if you’re not certain, it’s a no, and if you are certain, it’s a yes. And they came back with a definitive yes. And that changed our use of that tool overnight.”

Nothing about the product changed. What changed was that people finally knew where the line was, and stopped treating it like a public chatbot.

5. Keep it alive, and prove it worked

Mykim Truong is about to hire a contract manager. The reason is a weekly report that nobody owns (embarrassingly, it’s one of ours).

Randstad Enterprise has used Nomio’s contract repository for several years and part of what it does is send the legal team a weekly summary of contracts coming up for renewal.

“We currently don’t have an owner who takes action on that,” she said. “So that person will be fully dedicated to reaching out to the teams and saying: this needs to be renewed. Are we going to do something.”

Nothing failed here. The purchase was right, the rollout worked, the business likes it, the data is good. The tool simply produced an output that was nobody’s job, and an output that is nobody’s job is indistinguishable from no output at all.

Sophie Best, who advises legal teams and trains lawyers on AI through The Best Advisory, hears the same thing every week: teams that have bought the big-name tools and still feel they’re drowning.

“Everyone’s buying the Ferrari. Most of us haven’t learnt to drive it.”

“The hard part of legal AI starts after the contract is signed. And it starts with people,” she says. “I see two reactions in almost every team. Some trust the output too much and stop checking. Others don’t trust it at all and quietly go back to doing it manually. Neither gets you anywhere. What you need is people in the middle. Curious enough to try it on real work. Realistic enough to know where it falls down. That mindset doesn’t come with the licence. It has to be built.”

Camunda’s Ummu Fallon believes that legal technology failure is often caused by no one being responsible for ensuring it’s being used to its full potential. It’s the absence of a customer success manager, if you will.

“I find that people just say, ‘Here’s the technology, use it, go ahead.’ However, you’re not checking in with people on a regular basis. You’re not seeing how they’re actually using the platform, and there aren’t enough opportunities to help users get the most value from it.”

The other half of the problem comes from the vendor, and the FT’s Erin Ferguson has feedback about hers that she knows sounds ungrateful: new features arriving too quickly.

“My feedback has been that it’s a bit fast.”

What she wanted was time for her team to absorb what was already in front of them: “Making sure that the opportunities and the features and functions that are there are really well used and really well understood, before the new shiny disco ball comes out.” Wordsmith took the feedback on and made changes that let her manage the pace.

It’s an interesting mismatch: vendors compete on how fast they ship, while buyers are trying to build habits, which needs the opposite.

To ensure that habit-building is happening, once a quarter, every user at Belron completes a survey. It takes fifteen to twenty minutes, runs to ten or twelve questions, and asks what costs were avoided, what time was saved, what was done with that time, what problems came up, and what people built.

“None of them are anonymised,” said Jessica. “They all have to be willing to give full feedback with their name, so we can go back to them.”

Jessica is candid that this only works because she has leverage. The tool is funded centrally, so business units get their licences free. Completing the survey is the price. She’s equally candid about why she bothers: “If we don’t do it, it’s easily forgotten, and then people kind of slip out of the tool.”

Belron’s legal department has a formal target, with every person expected to automate 10 per cent of their work using AI in 2026. But Jessica now thinks it was the wrong metric.

“I would choose a different metric going forward, because I think a percentage of work is hard to measure without a starting point.”

Despite not being able to be precise, she has a rough idea of what they’ve achieved. She told her CFO: “Back of the napkin, we’re at 16 per cent.”

That is the most successful implementation in this report; however, its own owner cannot prove the return. None of which helps if you need budget.

As Sarah Owens from Barratt Redrow says, businesses run on money, so bring money. Quantify your own overtime, pick one pain point that eats time and returns little (NDAs, usually) and fix that. Carving out the time to do it is the hard part, as she acknowledges: “You will have to be brutal. You’ll have to find it.”

Then you document what it saved, and you spend that evidence on the bigger ask.

Corona Energy’s Hugh Ndhlovu has worked out the cheapest version of the same idea. Every contract his team has ever handled has an email trail behind it, timestamped from the moment it arrived to the moment it was signed. That is a historic baseline, free, sitting in the mailbox.

His plan is to build a small parallel process somewhere it doesn’t matter much, run it, and compare: “If I can start to show that actually this is the difference: look how much quicker this has been this way.”

All of it brings us back to where the last report finished. Ready or Not? ended with a ladder: contracts gathered, then validated, then finally in a state where AI could answer questions about them reliably. That turns out not to be the top.

Randstad Enterprise has had reliable answers arriving weekly for two or three years, and until somebody’s job description changes, nothing happens as a result.

The top of the ladder isn’t information. It’s ownership.

6. Decide who owns it

There’s a question in legal teams that didn’t exist two years ago: do we need a legal engineer?

Quentin Peltier from Dr. Ops answers it with a scenario. A team hires its first one, convinced it has bought the next-generation version of legal operations. Six months later the automations are brilliant, and the budget, the vendor renewals, and the board reporting are all still sitting on the general counsel’s desk.

His point is that these are different jobs for different employers. Legal engineer is mostly a vendor title; legal operations is a client one. A vendor needs someone to build and deploy its product. An in-house team needs someone to run a function, and running a function includes the building.

“Legal ops includes legal engineering. The reverse isn’t true. If your legal ops role doesn’t include the engineering part, it’s not a legal ops role. It’s a coordination role with a new title.”

Nathan Corr, who holds the title, says Wordsmith splits its Legal Engineers into three groups: one works with the development team to shape the product, one with customers on implementation and adoption, and one with the sales team to ensure the product meets the customer’s needs. All three are vendor-side: their job is to make the customer’s legal function work better.

Jasmine Oddy, Head of Legal Operations and Business Strategy at Rebel Legal, thinks the argument is newer than the work. Any in-house lawyer, she reckons, has been doing legal operations for years and has probably dabbled in the engineering. Her division of labour is the cleanest anyone described: legal operations asks who is doing what and whether it could be quicker, and legal engineering builds the answer. “One does the thinking, one can actually go and build it.” On a small team, that’s the same person on different days.

She also doesn’t have a law degree, and thinks it helps: “Often what I’m doing is translating the legal jargon or the risk into how it actually stands operationally for sales or procurement. I often say to our lawyers: if I don’t understand, they’re not going to understand. This isn’t their day job.”

Whatever it ends up being called, it is moving fast. Remote’s Natalia Cataife has been doing the job for four years and can’t describe it the same way two years running.

“If you asked me two years ago what legal ops was, maybe I don’t have the same answer today. I was not doing what I’m doing today, two years ago.”

And the ground could be about to shift again. Allan Smith expects AI to get more expensive. The pricing has been subsidised, he says, and the losses behind it won’t be carried by investors forever. When the bill arrives, Legal will have to account for what it is running, what that costs, and whether anyone is using it. Those answers will decide what gets renewed.

Somebody has to be doing that work, whatever their job title says. Which is Quentin Peltier’s whole point: “The risk isn’t choosing the wrong title. It’s confusing a role to hire with a function to cover.”

Conclusion

Every stage in this report describes work that somebody has to do, and almost none of it is legal work. It is change management: getting busy, risk-aware people to work differently, and keeping them at it.

That is what legal operations is for, which is why the legal ops people raised it unprompted when not one lawyer in the first report thought to mention it at all.

Awkwardly, most legal teams don’t have anybody dedicated to legal ops. Only some of the people in this report hold the job formally. The others are lawyers doing it in the gaps, or consultants doing it for other people, or, in some cases, the only lawyer in the building doing everything.

So the honest question we arrive at is: what do you do if the legal ops person isn’t going to be you, and isn’t going to be hired?

Rob Jarrett works as a fractional counsel to high-growth tech companies through Size Seven Law, and sees what this does to a lean team. They would need time to understand the problem, find a vendor, trial it, choose it, implement it, and get it adopted.

“They need to offload work in the first place to free up the time and headspace to properly think about and action this stuff, and that’s never going to happen without some form of flexible support or external guidance.”

The best answer to this problem came from the FT’s Erin Ferguson:

“My advice for the small teams who have no dedicated legal ops team is to get into a network and borrow someone else’s legal operations and knowledge.”

This works better than it sounds, because the people you would be borrowing from want to be asked. Almost everyone we spoke to volunteered time to the wider community, whether through roundtables, WhatsApp groups, conference corridors, or offering unpaid advice to strangers. Legal operations is a small profession full of people who like solving problems.

The paid version exists too. Ummu Fallon at Camunda would bring in outside legal ops support, interim or fractional, rather than wait for a permanent hire. Jessica Vander Ploeg puts that option a distant third, on the grounds that most consultancies don’t know the market as well as legal ops people do.

What everyone agrees on is that none of the problems described in this report were solved by the product alone. All of them were solved by somebody deciding that was their job.

If you take one thing from this report, take that. Then go and find someone who has already done it, and ask them how.