I've Seen the Same AI Rollout Mistake at Dozens of Plaintiff Firms

A firm buys an AI tool and everyone gets a login. Six months later the managing partner is staring at the same numbers from the day the contract was signed: margins, settlement values, conversion rates, all exactly where they started. The tool is everywhere and the results are nowhere.

I've seen this at dozens of plaintiff firms, and the software is almost never the problem. The problem is that nobody ever decided how the firm would use it.

Buying the tool felt like the decision when it was only the purchase.

The purchase is a project with a finish line. Adoption is a people problem, and people problems don't have finish lines. The firms that solve it anyway tend to run one of two plays.

Hand the tool to your toughest critic first.

At Duffy & Duffy, a medical malpractice and nursing home negligence firm, CTO Jessica Lockhart built her pilot around the firm's harshest skeptic on purpose. An expert nurse reviewer had told her flatly that no AI tool could do what she did. Jessica gave her the tool anyway. "I knew she'd push it to its limits, so we'd learn exactly what it was capable of," Jessica says.

The first time it caught something in a set of records the reviewer had already been through herself, she became the loudest advocate in the building.

I like this play because it runs opposite to instinct. Most firms pilot with their enthusiasts, and enthusiasts convince nobody. The person with the deepest expertise and the least patience for hype is the only endorsement a skeptical building will accept. If the tool survives her, everyone else believes the results.

Build for the bottleneck people already hate.

The second play comes from the Law Offices of James Scott Farrin, one of the largest plaintiff firms in the Mid-Atlantic with close to 300 employees. Before anyone there saw a demo, the implementation team interviewed every department and found the task each one complained about most. Then they built a single button that solved it. Workers' comp paralegals got a tool that summarized the last 30 days of a case and pre-filled their call prep before every client check-in.

"When my paralegals saw it for the first time, they absolutely had that wow factor," says Ryan Bliss, the firm's Director of Legal Technology. People started emailing him asking when they could have it, before anyone had asked them to use anything at all.

Neither tactic depends on finding people who love technology. Both hand the whole firm a result nobody can argue with.

If your name is on it, you read it.

None of this works without guardrails, and there are two I won't bend on.

The first is mandatory human review on everything that leaves the firm. The tools make review faster and more accurate than ever, not optional. And review means you, not a layer of comfort inserted between the output and the person responsible for the case: an outsourced reviewer, a QA team, a second tool checking the first. That looks like caution. It is usually distance dressed up as diligence. The attorney whose name is on the filing is the one whose judgment has to catch the error.

The second is a standard everyone at the firm understands: whatever has your name on it is your responsibility. Assume the first pass is imperfect and treat your job as finding what is wrong with it. If what goes out the door is exactly what the machine handed you, you didn't do your job. Good tooling makes that review real, checking whether a quote actually appears, verbatim, in the file and pointing you straight to the source. What it will never tell you is whether a paraphrase is fair or a legal conclusion is sound. That judgment stays with you.

The reason I won't bend is only partly liability. The person on the other end of the case does not get a second draft of their own life. A missed detail in a demand costs the client whatever that case was supposed to fix for them.

The part you control.

None of the firms above succeeded because their staff were more open-minded than yours. Someone chose who saw the tool first, or started with the task where the win was obvious, and let the result argue for itself. Then they held the line on review once adoption took.

There is a version of this piece that is just me telling you to buy more AI, and I'll be candid: heavier usage is good for my business. But no firm above got better by using more of any particular product. Adoption was a decision somebody made and then enforced. The tool is table stakes now. Everyone has it. Whether it changes how your firm works is still up to you.