Five Things Legal AI Leaders Are Thinking About

Recently I've been talking with legal leaders about where they are with AI. Some are just getting started and others are well into their rollouts, and the same themes keep coming up. None was about which tool  to buy. All were about people, pace, measuring ROI, data, and judgment.  

Many teams already have the tool in place. In a survey of more than 1,500 professionals in legal, tax, risk and government, 40% said their organization uses generative AI, up from 22% a year earlier. The questions around the tool are a different kind of problem. A purchase decision doesn't answer them, and I don't see any of them fully settled. 

I lead legal at Execo, and I'm still working out parts of this myself, so I'm not offering answers here. This is what leaders are thinking about, set next to what the research says, with my own view (where I have one). 

1. Internal champions make a real difference.  

Before anyone asks whether the tool is good enough, leaders ask who will pick it up and keep using it. Some described appetite as uneven, even inside one team: a few people take to it quickly, while others use it for little beyond the basics. Some preferred the person leading the effort to be someone who is eager to learn the technology, rather than someone who was simply assigned.

The research is broader than legal, but it points the same way. BCG's 2026 AI at Work survey found that 72% of employees say the skills expected of them have shifted, while only 36% feel they have had sufficient training. Microsoft's 2026 Work Trend Index found its most advanced AI users were far more likely to be on teams that refine their processes together, 63% against 32%, though it surveyed only people who already use AI. None of these studies looked at legal teams specifically.

Leaders further along in AI adoption said internal champions who encourage others to use the tool make a big difference. 

2. Phased rollouts with regular check-ins are easier for people to absorb.

Some leaders described rolling out in stages, one group before the next, as easier for people to absorb than giving everyone the tool and the training at once. Some said training delivered all at the start can overwhelm. My own view is that a tool people find hard to learn becomes one more thing on the to-do list that gets pushed further down. Staging is a way to make sure the right group learns the tool first (e.g. an AI task force or committee) and has a chance to ask questions and give feedback before it reaches the wider business (e.g. legal, procurement, sales). Lawyers face a second limit on pace: what can go into the tool. Confidentiality and privilege can restrict how freely a team can pilot on live matter data, so the question of what stays out comes up before the training does.

I haven't found a study comparing staged and all-at-once rollouts in legal, so this rests on what leaders told me and on that adjacent research. 

3. Measuring AI tool effectiveness is still a work in progress.

Leaders want to show the tool is working, and some weren't sure their measure would hold up. I didn't hear a settled method. I heard several metrics: time saved, work avoided, less spend on outside counsel. I don't know if there's an exact science to it.

The research shows how common that uncertainty is. In the Thomson Reuters Institute's 2026 AI in Professional Services Report, a survey of more than 1,500 respondents across 27 countries, only 18% of professionals said their organization tracks AI ROI, and another 40% said they don't know if it's measured at all. McKinsey's 2026 State of AI survey found 44% of organizations scaling AI across the enterprise, but only 37% attributing any EBIT impact to it, unchanged from a year earlier.

Two things make this harder in legal. If a team doesn't record time, there's no ready-made "before" to compare against, so a baseline has to be built. And some of legal's value is a dispute or a loss that never happened, which cost-and-usage measures don't capture. That second point is my view.

4. Clean data is key for a smoother AI transformation.

Some leaders talked about what they want AI to do next, and every aim sat on top of data. They included searching across every contract a team holds, applying written standards the same way on every review, and building an AI strategy on the company's own data. Leaders described these as aims, not as where their teams are today.

One leader I spoke with described treating the data as the first job, before any AI. They started by deciding which data the business actually needs to run, then worked on making that data reliable, and only then built AI on top of it. In their view, an AI strategy without that groundwork has nothing solid to stand on.

Leaders in regulated industries said the hardest part was, and still is, moving or consolidating data across the business, because each unit guards its own data and regulators set different limits on what can be transferred. 

What these aims share is that a tool can only work with what a team has captured. If the standards live in people's heads, or the contracts sit in folders nobody can search, the tool has little to draw on. "Clean" here means complete, consistent and easy to find. In my observation, that work rarely gets a line in the rollout plan, and it certainly is not a weekend job.

The research points the same way, though it isn't specific to legal. Deloitte's State of AI in the Enterprise 2026 found that only 40% of leaders say their organization's data management is highly prepared for broad AI adoption. 

5. The final call stays with the lawyer.

On this one, leaders were in agreement. AI tools keep getting more capable, but review and the strategic calls stay with the lawyer. 

For a lawyer, keeping the final call is also a professional duty. The ABA's Formal Opinion 512 expects lawyers to understand a tool's capabilities and limitations and not to rely on its output without independent review, though the level of review depends on the task, and the lawyer remains ultimately responsible.

The research shows both sides of this. In the Thomson Reuters Institute's 2026 report, about two-thirds of professionals said generative AI should be applied to their work in some way, yet accuracy and reliability remain the main reasons they hesitate. The survey covers professionals across several fields, not only lawyers.

My view: Reviewing final outputs is second nature for seasoned lawyers, even with advancing AI tools. The challenge is to keep up with such tools, and to understand where AI is needed and where AI is not.

Where your team goes from here

If one or two of these are on your team's mind, you're asking what other legal leaders are asking. They came from leaders at very different points in their rollouts, so asking these questions isn't a sign you're behind.

People, pace, measuring ROI, data and judgment — choosing a tool doesn't settle any of them. The upside is that you don't have to wait for a tool decision to start, and the work carries over to whichever tool you choose.

So which of the five would you put first for your team, and who owns the answer today? If you can't yet name an owner, that's where I'd start.