Login | June 09, 2026
Predictive coding has its detractors
RICHARD WEINER
Legal News Reporter
Published: June 14, 2013
Would you allow a mathematical probability primacy against a real life evidence search?
At least two judges in the last year or so have done that—ordering a predictive coding of a mountain of evidence versus allowing the attorneys to search the evidence themselves.
Predictive coding is a computer program that uses a mathematical formula to drill down into a mass of ESI documents looking for word, phrase, or concept matches. searches the entire text, and then pulls out a random, small sample of “relevant” documents, which is supposed to, mathematically, find a viable sample of evidence, which, statistically, should be all that the court needs to look at.
Statistically, supposedly, this process has a better chance of finding a relevant document (75 percent or so relevancy, versus 60 percent when the documents are hand-read).
And some judges, in order to keep discovery costs down, are ordering that counsel engage in predictive coding of ESI instead of hand searching, say, three million documents, or using a program that searches every doc.
But the math doesn’t hold up to the realities of what is needed in litigation, said attorney Bill Speros, who ran the document search in the Bernie Madoff case and is a recognized expert in ESI.
In a lengthy post on the blog e-Discovery team, Speros ripped the reliability of PC apart.
In one of a number of reasons that he does not like PC, Speros said that the program “over-promises and under-delivers” because it does not find every relevant document, and, in fact, ignores most docs. Among the causes of ignorance are multiple meanings of a word, vagueness of search terms, and the limitations that a lawyer will place on the search altogether.
Another negative is the great potential of missing a one-off, “smoking gun” email, like the cost-benefit letter discovered in the old Ford Pinto case which led directly to a $140 million jury award. Predictive coding would never find a document like that.
Speros has numerous other examples of the inherent weaknesses of PC. But his protestations may not change anything.
There is an info war going on over document production and the various attempts to limit the out-of-control costs of ESI production, especially in massive litigation.
There will be much more to come.