"Predictive Coding" in E-Discovery

Blawgletter gets calls each work day from people who want to sell something. It happens to you, too. When you pick up the phone, you hear dead air for a second or two before a person comes on the line. The "agent" often asks if he can speak with "[your name here]".

The tell-tale lag means that the caller forms part of what Blawgletter knows as "outbound telemarketing" -- and that the agent used a "predictive dialer" to reach our line. Our source for all facts whose accuracy doesn't matter -- Wikipedia -- tells us:

Predictive dialers use statistical algorithms to minimize the time that agents spend waiting between conversations, while minimizing the occurrence of someone answering when no agent is available.

The dialer raises the time when an agent speaks with a callee from about 40 minutes per hour to around 57.

What, you demand to know, does that have to do with e-discovery. Just this: A growing number of fancy-pants e-discovery vendors claim that their software programs vastly enhance the efficiency of looking for stuff -- in this case, hot documents instead of hot prospects for buying Veg-a-Matics, credit repair services, and hair-weaving devices.

But does the hype match the True Facts? Because if it does a lot of lawyers will spend a lot less time searching through terabytes of electronic documents to find things relevant to lawsuits. And clients will spend a lot less money paying for all that doc-review time.

Call us slightly more on the side of the hypers. Slightly.

Using algorithms to locate hot docs makes a ton of sense for that first cut through the Great Mass of E-matter that you image from your client's gargantuan server. The process does better than having lawyers do "keyword" searches, not least because keyword searches can go on and on and on. The big gain from using the software should come from having more faith that the program will catch the right documents. If you do it over and over again, though, you haven't done much more than find a new way to waste time.

And yet you just might do that. Per the ABA Journal:

Essentially, the software works by delivering results based on a barrage of keyword inputs, which are tweaked by a seasoned attorney who then continually refeeds the best resulting documents back into the system as examples until the software “learns” what the attorney is really looking for.

So you still need someone who can teach the software what types of docs to sort into the human-review pile. But one person instead of dozens. Which should improve consistency but makes the value of the whole effort depend on how hard and well the one person does the job.

Who should play the role? We say someone who has Tried a Case. Because we don't care how smart a lawyer you have culling through the docs that make the first cut. If that person doesn't have a good idea of what you can do with the hot docs in front of a judge or jury, you'll likely get a big pile of docs that will never see an exhibit sticker.

So, we like the tool. Sort of. But like any tool it will save time only if you use it with planning and discipline.

Comments

[Max Kennerly]

Some of my wife's law school loans are through Citibank. Each month, when Citibank inevitably fails to auto-deduct or to follow through with the payment I set up, we start getting multiple calls every day, almost all of which result in 10 seconds of dead air followed by the automated caller hanging up on us.

On the rare occasion that a person actually picks up after 5-7 seconds of dead air, I explain to that person that Citibank's own online system says my payment is "processing," and then ask why a bank's own transaction system takes 3-5 days to process its own payment. They blame me for the problem and then hang up.

Then an hour later another automated call comes up and there's another 10 seconds of dead air followed by it hanging up.

Consider me unimpressed.