Latest / The Operator's Manual with Fexingo: Daily Decisions, Tactical Wins, and Business Execution / How One Law Firm Cut Discovery Costs By 60 Percent
Transcript
- Lucas: So there's this moment in every big lawsuit that nobody outside a law firm ever thinks about. It's called discovery — the phase where both sides have to hand over every relevant email, contract, memo, and Slack message. Luna: Right, the document dump. I've heard associates describe it as a digital paper shredder, but backwards. Lucas: Exactly. And for decades, the way firms handled that dump was brutally simple: they hired an army of junior associates and temp lawyers to sit in a room and read every single document, one by one, and tag whether it was relevant or privileged. We're talking millions of pages per case. Luna: And that's where the cost lives. A single document review can run into the tens of millions for a big case. Lucas: Right. But in the early 2010s, a few firms started using a technique called technology-assisted review — TAR for short — and one firm in particular cut its discovery costs by sixty percent on a set of major cases. That's the story I want to walk through. Luna: Okay, so what was the before picture? Give me a concrete example. Lucas: Sure. Take a firm like Munger Tolles & Olson — though this applies to most Big Law firms pre-2012. In a typical securities class action, they'd collect maybe five million documents. Each document averaged about ten pages. So fifty million pages. Luna: Fifty million. And humans read every page? Lucas: Every single one. At a typical speed of about fifty pages per hour per reviewer. If you have a team of fifty reviewers working forty hours a week, that's a hundred thousand pages a week. It would take them nearly ten weeks, and the cost at a blended rate of about fifty dollars an hour — well, you do the math. Luna: That's two hundred and fifty thousand dollars a week, two and a half million for the whole review. And that's just the labor. Lucas: And it's not even the biggest cost. The real killer is that you're pulling your best junior associates off drafting and depositions to do data entry. The opportunity cost is huge. Luna: So what did TAR actually change? I hear 'predictive coding' thrown around, but I'm never sure what it means in practice. Lucas: It's simpler than it sounds. You take a small sample — maybe two or three thousand documents — and have a senior attorney hand-tag them: relevant, not relevant, privileged. Then you feed those tagged documents into a machine learning model that learns the patterns. After that, the model scores every other document in the collection on a relevance scale from zero to one hundred. Luna: So the model is saying 'this email looks like the ones the attorney tagged as relevant.' Lucas: Exactly. Then you set a threshold. Say, anything scored above seventy goes to a human for final review. Everything below goes into a discard pile. And you validate the model by sampling the discard pile to make sure you're not missing anything important. Luna: And what does that do to the volume? How many documents actually need a human to read them? Lucas: In the best cases, you cut the review set by eighty to ninety percent. Instead of fifty million pages, you're looking at five to ten million. And the cost per page drops from around two dollars and fifty cents to under a dollar. Luna: That's huge. But I've heard pushback — judges and opposing counsel sometimes argue that TAR isn't as reliable as a full human review. Is that fair? Lucas: It was a real concern early on. But there's a landmark case from 2011 — Da Silva Moore v. Publicis Group — where a federal magistrate judge explicitly approved the use of TAR, citing studies that showed it was at least as accurate as human review, and in some cases more consistent. Luna: More consistent because humans get tired and miss things? Lucas: Exactly. A human reviewer in the fifth hour of looking at contracts is going to start making errors. A machine doesn't get bored. Plus, the model can be re-trained mid-review if you find it's missing a category. Luna: So what's the catch? Why isn't every firm doing this by 2026? Lucas: Great question. The biggest barrier is cultural. Partnership models in law firms are built on billing hours. If you cut discovery time by sixty percent, you're cutting a huge revenue stream for associates and partners who rely on those hours for their compensation. Luna: So the incentive structure fights the efficiency. The firm itself profits from the waste. Lucas: Exactly. And that's the real operational lesson here: process improvement is often less about finding the better way and more about getting the organization to accept it. The firm that adopted TAR early — the one that saved sixty percent — they had a managing partner who said 'we're going to take the revenue hit on discovery and make it up on higher-value work, and we're going to win more cases because our clients save money.' Luna: And did that bet pay off? Lucas: From what I've seen, yes. They ended up winning several large mandates specifically because the clients' general counsels did the math and realized they'd save millions in e-discovery costs. It became a competitive advantage. Luna: That's a great example of how a single process change — shifting from linear review to TAR — can ripple through the whole business model. Lucas: It really is. And it's the kind of story we love to dig into on this show. You know, we don't run ads on these episodes — we keep it strictly about the operations and the decisions. If you find value in that approach, and if these conversations have moved your work forward in some small way, you can support the show at buy me a coffee dot com slash fexingo. Luna: Yeah, it's a small way to keep the thing ad-free and focused on the actual tactics. No pressure at all. Lucas: Exactly. So back to the law firm. Once they proved TAR worked on a few cases, they started building internal tools to make the process faster. They created a feedback loop where the model improved with every new case. Luna: So the model learns from every review, getting more accurate over time? Lucas: Right. The first case might require a training set of ten thousand documents. By the tenth case, maybe only two thousand. That compounds the savings. Luna: And what about the types of documents? Does TAR work as well on contracts as it does on emails? Lucas: It works better on structured documents like contracts, actually, because the language is more formulaic. Emails are messier — more slang, more typos — but still, the model can handle it. The key is having a good senior attorney do the initial tagging. If the training set is sloppy, the model will be sloppy. Luna: So garbage in, garbage out. That's true for any machine learning application. Lucas: Exactly. And that's why the firms that succeed with TAR invest heavily in the training phase. They don't just hand it to a junior associate. They have a partner or senior counsel spend two days carefully tagging a few thousand documents. Luna: That's a big upfront time investment. But if it cuts ten weeks of review down to two, it's worth it. Lucas: Absolutely. And that's the operational lesson I want to leave with: sometimes the biggest efficiency gains come from doing a small, high-quality batch of work upfront, rather than trying to scale a mediocre process. It's counterintuitive, but it's true. Luna: So for anyone listening who manages a professional services team, the takeaway might be: look for the repetitive, high-volume task that everyone hates, and ask if a small investment in training or technology can eliminate eighty percent of it. Lucas: That's exactly it. The legal industry is full of these opportunities — due diligence, contract review, patent searches. And the firms that figure out how to automate the boring parts are the ones that will thrive over the next decade. Luna: I'd love to see a follow-up on how AI is starting to handle deposition summaries. I hear that's the next frontier. Lucas: That's a great idea. Let's put that on the list. For now, I'm Lucas. Luna: And I'm Luna. Lucas: This has been The Operator's Manual with Fexingo. We'll see you next time.