AI Governance for Legal Ops: 5 Lessons from Franklin Templeton’s Patty Corey
Patty Corey, Legal Operations Manager at Franklin Templeton, is building the company’s first global legal ops function. And she is doing this with attorneys at different stages of AI adoption. Some are just learning how to prompt; others are already building skills in Claude and ChatGPT.
With so much in flux, Patty is faced with a unique challenge; how does she provide Franklin Templeton’s legal team with an AI governance framework that embraces innovation, maintains security, and is clearly understood?
In a recent interview, she shared her approach to supporting AI experimentation while still keeping guardrails in place. According to Patty, it comes down to: training, open communication, and a simple intake process. She calls this behavioral governance, which means shaping how in-house legal teams use AI through training, culture, and shared use cases, so good habits exist before a formal policy does.
Here are five lessons from her playbook that legal ops can act on today.
1. Take a Seat at the AI Governance Table
As AI becomes part of everyday legal work, in-house teams are asking how to make sure they can trust it. For Patty, it comes down to “trying to strike the balance between how do we encourage experimentation and innovation but still have guardrails in place and a policy that governs how we use it in a responsible and healthy way.”
When she asked a mentor of hers how legal ops can help governance keep pace with AI, the answer was simple: “AI governance is a team sport, ultimately. You don’t need to be at the head of the table. You just need to sit at the table.”
And legal ops has good reason to be there. Because legal works with so many teams in the business, Patty says, it gets a view into how the broader company is using AI, including how those teams use it when they work with legal. Legal workflows also differ from the rest of the company’s, so someone has to make sure they are captured in the enterprise policy.
To keep things friction-free, Patty’s advice is to talk early and often because open communication with your AI governance team is the key. If you’re drafting a use policy for a specific legal tool, ask them to review it and flag any duplicated effort. She also suggests helping the team see that AI governance isn’t purely a technical problem. There’s a behavioral piece that has to be built in too.
Sharing what you’re seeing helps just as much. Patty says keeping a dial on new AI regulations is a big part of influencing the framework. When something new comes out, she passes it along to the governance team and invites a conversation about it. The point is to start a discussion, instead of raising a flag.
2. Treat Training as Governance
“I think that training is really a form of governance,” Patty said.
She starts with the basics and meets her attorneys where they are. That means showing them how to set up their AI tool with the right context and custom instructions, so the output is something they can build on instead of rework again and again. It also means teaching them to verify results: check citations, and read the source documents before and after they go into the tool.
In addition to teaching skills, training also builds the culture around policy. The goal is a team where getting the best output from AI is simply how people work.
3. Build a Problem Pipeline, Not Another Approval Step
At one point, Patty considered a policy requiring any custom GPT shared across the legal department to go through legal ops. Then she thought it through. “No one is going to build if they have to do extra work to go through legal ops to do it.” The policy had good intentions, but it would have stifled innovation, so she dropped the idea. She didn’t want legal ops to be seen as the department of no.
Her alternative is a legal ops intake form she calls a problem pipeline. Attorneys share their problems, legal ops turns them into use cases, and the two teams build the solution together before sharing it with the rest of the department. Patty described it as “building governance without making it a stringent policy.”
Patty starts with the problem, not the proposed use case, because the two can be very different. “I very much start with the problem, and I try to back into the solution.” And sometimes the fix isn’t AI at all. Because she’s capturing every problem, she can see where automation or process redesign makes more sense before AI enters the picture.
So what about the attorneys who just start building on their own? Patty is fine with that. She wants people to experiment on their own and asks only that they share their use cases. The goal is visibility, not a gate.
The pipeline works alongside ongoing conversation. In every AI training, Patty asks attendees what they’re building, what they’ve used AI for, and what their AI wins are. Then, those wins get featured in a legal ops newsletter, which works as an incentive to use AI, since a feature could help in a performance review.
4. Battle Test Before You Share
The attorney who builds an AI tool isn’t always the best person to test it. Patty points out that building and testing are different skill sets. Testing is where she steps in, drawing on years of evaluating tools.
AI is nondeterministic, which means the same prompt can produce different answers. Patty’s advice for navigating this is to “think of AI really as a human.” Ask a person the same question twice, and you might not get an identical answer both times.
That’s why Patty is big on battle testing any AI tool. It means running it through the scenarios people will actually use it for and checking whether the output is reliable and cites sources appropriately. Then you iterate.
She also sets a practical bar. “If you’re 90% of the way there, I think that’s good enough, and we have to accept that this isn’t going to be perfect,” Patty said. That can feel unsettling for attorneys, but she finds it comforting. “Sometimes perfection can be the enemy of the good, and innovation in many ways,” she said. “I’m comfortable living in that grace space because I trust my judgment.” She wants attorneys to see their own expertise the same way. “That extra 10% is where you use that judgment and that nuance.”
5. Start Small If You’re Starting from Zero
For legal ops professionals with no AI governance in place, Patty’s advice is to start with training. If you are still learning AI yourself, host AI office hours. Getting people talking surfaces the problems they’re running to, and those problems tell you what to train on.
Then capture them. “You need a problem pipeline. I can’t say it enough.”
Hers was a simple build. Patty set up a Microsoft Form, synced it to an Excel workbook, and then used the MCP connector from Excel to create an artifact in Claude that generates a dashboard for her team. Each submission can be assigned to a teammate, and she can add notes. She noted that it works like a Jira intake system, and it took her about two hours to build, probably less.
Final Takeaway: Legal Ops Can Lead on AI Governance Now
Here’s the good news for legal ops: you don’t need a finished policy to start. Grab a seat at the table, train your attorneys, collect their problems, and test what they build. Each of those steps shapes how your department uses AI, and none of them has to wait.
Patty sees this as a shared effort. “We’re all in this together. It’s just a crazy time in the legal profession, but it’s also really exciting.”
Listen to the full In-House Outliers episode for the whole conversation.