Latest / The Operator's Manual with Fexingo: Daily Decisions, Tactical Wins, and Business Execution / How One Law Firm Cut Review Time 60 Percent
Transcript
- Lucas: So there's a story out of a large international law firm that I think perfectly captures what we talk about on this show. They were drowning in document review costs on a massive merger case — think hundreds of thousands of emails, contracts, memos from both companies. Luna: That's the classic grind. Junior associates in windowless rooms, billing hours, trying to find the few key documents in a haystack. Lucas: Exactly. And the partner on the case had seen a demo at a legal tech conference — an AI tool that could flag relevant documents with a confidence score. The vendor claimed it could handle 80 percent of the initial pass, leaving humans only the high-stakes decisions. Luna: But trusting an algorithm with client work — that's a leap. Especially in a profession built on billable hours and personal liability. Lucas: Right. So they didn't jump in blindly. They ran a pilot on a smaller internal project first — about five thousand documents from an old case. The AI matched the associates' judgment on 99.2 percent of the classification decisions. The misses were all false positives — it was conservative, flagging borderline stuff that humans would also flag. Luna: Okay, that builds confidence. So then they rolled it out on the merger case? Lucas: Yes. But with a twist. They changed the workflow instead of just bolting the AI onto the old process. Previously, associates would read every document from start to finish. Now the AI would pre-sort everything into three buckets: likely relevant, possible relevance, and likely irrelevant. Luna: And the associates only had to review the first two buckets? That alone cuts volume drastically. Lucas: Right. The 'likely irrelevant' bucket — about 60 percent of the total — they spot-checked at a 5 percent sample rate. That saved an enormous amount of time. The result: total document review time dropped from an estimated twelve thousand hours to just under five thousand. That's a 58 percent reduction — we're calling it 60 percent for the headline. Luna: And the accuracy? You mentioned the pilot was good, but real-world pressure is different. Lucas: They tracked it. The final set of documents that went to opposing counsel — the ai reviewed ones — had zero errors in flagging privilege or key evidence. The partner told me the biggest surprise was that the junior associates actually caught more subtle issues because they had more time to focus on the ambiguous documents. Luna: That's the hidden benefit: when you remove the fatigue of reading ten thousand irrelevant emails, the brain works better on the ones that matter. Lucas: Exactly. And the firm also discovered something about their billing model. The client had agreed to a flat fee for the review, so the firm kept the efficiency savings. That six-thousand-hour difference — at, say, four hundred dollars an hour — that's two point four million dollars that didn't get billed to anyone. The firm earned it by being smarter. Luna: I love that framing. It's not about cutting people; it's about cutting wasted effort. And the associates got to work on higher-value tasks, like drafting motions and advising on strategy. Lucas: Look, a handful of listeners chip in monthly through buy me a coffee dot com slash fexingo, and that's literally what funds making deep dives like this possible. No ads, no sponsors — just the economics of a small group saying 'keep going.' Luna: Yeah, it's a quiet model but it works. And it means we don't have to shoehorn in commercials that break the flow. Anyway, back to the law firm — what did they do after the case? Lucas: They standardized the tool for every document review above a certain threshold. They also built a quality dashboard that shows the confidence distribution so partners can see how much of the review was automated versus human. It's become a selling point for clients: 'We use AI to reduce cost and improve accuracy.' Luna: And the junior associates? Did they resist? Lucas: Initially, some were worried about losing billable hours — which is legitimate when your bonus depends on hours. But the firm changed the compensation model for review work. Instead of paying for hours spent, they paid a flat rate plus a bonus for accuracy and speed. The top performers ended up earning more, because they finished faster and moved on to other cases. Luna: So it was a full operational redesign: process, tool, and incentives. That's the takeaway for me. Lucas: Right. And what struck me is that the one change — the AI tool — only worked because they were willing to rethink the workflow around it. If they had just handed the AI to associates and said 'use this,' they'd have seen maybe a 20 percent improvement. By changing the sorting bucket and spot-checking approach, they got the full 60 percent. Luna: That's the difference between technology adoption and process transformation. One is a new tool in an old box; the other is a new box. Lucas: Exactly. And it's a lesson that applies beyond law. Any knowledge-work process where humans scan large volumes of data — claims processing, compliance checks, medical record review — the same pattern holds. Luna: So if someone listening wants to try this in their own organization, what's the first step? Lucas: Start like they did: a small, low-risk pilot with clear metrics. Measure the current speed and error rate. Then run the AI alongside your existing process for a limited set. Prove the accuracy. Then redesign the workflow to take advantage of the new capability. Don't skip the redesign step. Luna: And be ready to change how you pay people. If incentives still reward hours, efficiency will be punished. Lucas: Yes. That's the hardest part, honestly. But the firm we looked at managed it, and now they have a competitive advantage that will last years. The AI is commodity now; the workflow is the moat. Luna: Love that line. I'm going to steal it. Lucas: Feel free. That's the kind of thing we get to say when we're not selling ad space. Luna: Alright, next episode we'll dig into a similar story from insurance claims. Let's see if the same pattern holds.