企業內部法律主管如何看待 AI：他們的親身説法

From left: Carlo Salizzo, Dentons; David Kirton, Dentons; and Ann O'Dea, Silicon Republic at a report preview event on 22 October. Image: Dentons

Multiple surveys count how many lawyers are using AI. A new report set out to ask what happens to their workload when they do.

AI isn’t replacing in-house lawyers, but the capacity it frees up is being absorbed by new work, much of it drafted by AI itself.

That is just one of the pictures to emerge from interviews with legal and compliance leaders at companies that included eir, Flipdish, Gilead Sciences, IMRO, Mastercard, Uniphar and Wayflyer for a new report, ‘From AI Adoption to AI Advantage’, produced by Dentons Ireland and SiliconRepublic.com.

Rather than creating another survey full of statistics, we set out to ask what responsible and effective use actually looks like inside a legal function today, and the results often belied the received wisdom.

Augmentation, not replacement

Almost nobody we interviewed reported AI replacing a role or a function outright. Where it is paying off, it is mostly in repetitive work such as invoice review, contract checking and customer screening, rather than in the research, drafting and judgement calls at the heart of the job.

Sergey Kinchin, senior counsel at Uniphar, said his own thinking had changed completely in a year. “Last year I would have thought it would probably reduce headcount. Now I’m seeing it more as unlocking unmet demand,” he said. “I think the complexity of the work that’s left after AI will increase. The bottom might shift up, but the overall pyramid of work will increase and grow.”

It was a sentiment echoed by many of our interviewees. Máire Neary, who was VP of people and legal at Flipdish when she was interviewed, described treating AI “as a junior colleague, something that’s additive, that can challenge me, but you have to remember you’re the lawyer in charge at the end of the day”.

The catch? More to read, more to answer

The catch is that the same tools are now in everyone else’s hands. Producing a long, polished complaint, subject access request or grievance is close to cost-free today, and the report says AI-assisted drafting now seems to touch the majority of incoming disputes, subject access requests, grievances and customer complaints.

Sorcha Nic Mhurchú, general counsel at eir, said the cost lies less in reading that material than in sorting through it. “You end up going back and having to figure out, line by line, what is relevant”, she said. “So that adds a lot of volume and time to any response, and that’s a reality for everybody, not just us.”

At IMRO, chief commercial officer Seán Donegan is glad he saw it coming. “Two years ago, I sat down with our licensing team and I said, guys, I’m going to make a prediction now,” he said. “The volume and tenacity of queries and complaints is going to increase exponentially with these AI tools.”

And it’s not just incoming correspondence. Uniphar’s Kinchin described a call with a part of the business that had gone all-in on AI “We joined a call and we had five people bring six AI-generated opinions to the call, on a question that touched a regulatory issue,” he said. “In trying to reconcile it, the AI had essentially confused itself.”

Deirdre McCarthy, chief legal and people officer at Wayflyer, added the issue is quality as well as quantity. “AI can produce so much content that sometimes looks good, but may not always be good,” she said. “So you can end up with people spending a lot of time just trading red lines because they’re putting it into their AI tool.”

Many of the interviewees remarked that this makes legal expertise all the more valuable. “In a world increasingly full of AI-generated legal content, genuine human judgement may become more valuable rather than less,” said Carlo Salizzo, head of technology and digital regulation at Dentons Ireland, speaking at a preview of the report last week.

The ROI question

What AI costs is easier to count than what it delivers. Flipdish’s Neary was clear about the risk of experimenting for its own sake. “I think we’ve seen a lot of AI for AI’s sake, but every lawyer, every operator, needs to ask what the return on investment is,” she said. “It’s great if your team is enjoying building and experimenting and having fun, but what have you got to show for it, particularly once you’re looking at a high LLM [Large Language Model] bill, or the cost of maintaining it down the line?”

None of the interviewees were confident they could demonstrate a return beyond obvious gains such as efficiency and risk reduction. Rebekah Dobson, head of international legal operations at Gilead Sciences, said the premise of the question itself is flawed.

“No one has proved ROI on AI as of yet because companies were not tracking the metrics prior to AI to have a starting point for setting KPIs and measuring impact of AI,” she said. “I did read an interesting article the other day that said ‘you don’t make Excel prove its ROI, it’s just there’, and I think that’s a really valuable comparison.”

Legal teams have a particular problem here, according to David Kirton, partner, technology and digital regulation at Dentons Ireland. “We tend to record the work lawyers do. We are much less good at recording the work they no longer have to do,” he said.

Tokenomics

Cost, meanwhile, has so far only moved one way. As AI providers shift from flat licence or per-seat fees to consumption-based, token-metered pricing, bills become much harder to forecast.

Nic Mhurchú said eir feared being on a treadmill. “The licence fees and the token fees are so huge that we fear that if we start adopting and using and buying licences, we’re on a treadmill of just paying more fees, because these companies have to justify the high investment that they’ve been putting into these tools,” she said.

Jessica Williams, vice president of legal tech strategy at Mastercard, says she wants to see vendors offer more solutions and controls. “Budgets aren’t infinite and I’m going to have three or four people who are AI geniuses and they’re going to burn through tokens”.

“Then I’m going to have a bunch of other people who just do one or two things here and there … How do I balance all of that and create a predictable budget line for the company?”

Cornelius Grossmann, Dentons’ global head of solutions development, expects AI costs to rise two- to fivefold within two years and warned of a coming ‘cost cliff’. “I think the whole token economics will mean we’ll probably hire people to monitor AI cost in a big law firm,” he said. “And others may need to do that too, because it could bankrupt you if you don’t pay attention.”

Wayflyer’s McCarthy expects token spend to become the next big management question. “I think this is the next interesting wave, what we spend on AI and who gets to spend what, based on the tasks at hand?” she said.

It was a sentiment echoed by Dentons’ Kirton. “Who gets access? Which tasks justify the cost? How do you allocate consumption across a business? And how do you prevent successful adoption from producing an unexpectedly large technology bill?” he asked, adding that many legal leaders are still working out how to manage that exposure without throttling hard-won AI adoption.

Overall, interviewees described something more demanding of both sides of the client and law firm relationship: a renegotiation of who does what, who is trained for what and who pays for what, conducted in real time while the technology changes apace.

“It is probably fair to say that the legal function has not been transformed by technology to quite the same extent as areas such as finance, sales or procurement,” said Dentons’ Salizzo. “AI may be the catalyst that changes that.”

The full report is now live and can be downloaded here.

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.