Cutting edge eDiscovery tools by Epiq saves law firm time, money
By AdvocateDaily.com Staff
Cutting-edge eDiscovery tools can save an astonishing amount of time and money for law firms and their clients, says Jason Bell-Masterson, manager at the Toronto branch of the legal technology company Epiq.
The experience of one recent client perfectly illustrates how conceptual analytics tools provided by Epiq can help law firms pinpoint virtually all relevant documents in a fraction of the time it would take using traditional methods, Bell-Masterson tells AdvocateDaily.com.
Given about a week to identify the relevant documents from more than 150,000, the law firm was able to complete the job in time with the help of Epiq's software that continually refined the entire set to identify virtually all the relevant material via a targeted manual review of just 33,000 documents.
“The legal team ended up reviewing less than 25 per cent of the document set, while still identifying the relevant content to meet their obligation to the court and make the production,” Bell-Masterson says. “And they did it in a week with a team of 10, instead of taking a month.
"To meet the deadline without the use of analytics, it would have been significantly more expensive because they would have needed at least four times as many people,” he says.
Bell-Masterson says the 150,000-document set is by no means the largest his company has been involved with, but the court-ordered production deadline made it more challenging. Using an initial sampling of the documents, Epiq estimated around three per cent, or 4,500 documents, were likely to be relevant and need to be produced to the opposing side.
“It was going to be a substantial task to get through these documents using normal methods,” Bell-Masterson says.
To start the ball rolling, the law firm’s document review team pulled together a small set of 140 documents it had already identified as among the most relevant to the lawsuit, as well as a number that they identified as non-responsive.
On a Friday night, those known documents were then fed into the NexLP system by Epiq, which uses continuous active learning — a type of predictive coding — to rank the entire 150,000-document set in terms of relevance.
“Over the weekend, the NexLP software crunched the numbers and classified everything,” Bell-Masterson says.
On the following Monday, the 10-person document review team began going through the documents listed at the top end of the predicted relevancy scale. During the first day and a half of manual review, the team found 70 per cent of the material they went through met the test for production.
“Keep in mind that if you were going through these documents at random, you would only expect to find a three-per-cent relevancy rate,” Bell-Masterson says. “They were immediately finding the majority of what they were looking for, but without needing to review anywhere near as many documents as you would in a typical review.”
After some quality checking, the document review team fed their own findings back into the NexLP system, allowing it to update and further refine the likely relevance of the remaining documents.
By the end of the day Wednesday, the law firm estimated it already found around 50 per cent of the documents it needed to produce, and by Friday morning, they were up to 80 per cent.
“By that point, when you’ve got most, if not everything that you need, the next step is a random sampling of what remains,” Bell-Masterson says, adding the sampling turned up no significantly relevant content.
By the end of the week, the team had reviewed just 33,000 documents in detail out of the 150,000 total. Bell-Masterson says the team of technology-assisted review consultants at Epiq helped ensure the process was defensible in court, by using statistics to show the cost and effort involved with combing through the remaining documents by hand would not be justified, due to the statistically insignificant chance of stumbling across anything of relevance to the lawsuit.
Bell-Masterson says the system works best for text-based documents and is not ideal for cases with a high rate of Excel spreadsheets because they tend to contain more numbers than text.
“At this stage of its development, machine learning can’t yet handle the numbers and acronyms that you see mostly in those documents, he says.