Apr 11, 2012
Litigants worry that producing electronic documents relevant to a civil lawsuit is consuming a growing share of litigation costs. A RAND report examines this problem and recommends options for addressing it. According to the study, the review process accounts for 73 percent of the total costs of production, and predictive coding, in which computers do the bulk of the review now performed by attorneys, is a promising option for cutting costs without compromising the quality of the process. The study also calls for rule changes to address concerns about the scope and process of preserving data in anticipation of litigation.
The cost of producing electronic documents in a civil lawsuit has become a policy concern in recent years. Pretrial electronic discovery, or e-discovery, consists of essentially three activities: (1) collecting potential sources of information in response to an opposing party's demand, (2) processing that information to reduce its volume and organize it for review, and (3) reviewing processed digital information to identify relevant and responsive documents to produce and privileged documents or confidential information to withhold. With the explosive growth of digital information, many believe that the cost of these activities—particularly review—accounts for an increasing proportion of total litigation costs.
A RAND study assesses e-discovery costs and describes how they can be reduced without compromising the quality of the discovery process. It also examines what litigants perceive to be the key challenges of preserving electronic information for the purposes of litigation.
The researchers worked with eight very large companies in diverse industries that were willing, with assurances of confidentiality, to provide information about their discovery expenses in 57 large-volume cases. They also interviewed company representatives about their preservation duties and concerns.
The authors found that the review process is by far the most expensive task in producing electronic documents, as shown in the figure. At least for the large-volume cases included in the study, review accounts for 73 percent of the total cost of producing documents, while processing the information accounts for 19 percent. The work itself is largely performed by other parties: outside counsel (typically for review) and vendors (typically for processing and collection). As the figure shows, only 4 percent is spent on the litigant's own in-house counsel, information technology staff, and other employees.
Because it makes up the greatest percentage of e-discovery costs, the researchers targeted the review process in their search for ways to reduce costs. They set an ambitious goal of reducing review costs by three-quarters to bring them in line with processing costs, assuming that doing so would reduce many of the complaints now voiced about the expense of e-discovery. The authors also summarize comparisons of the quality of traditional review practices that require eyes-on review of documents by attorneys with the quality of new approaches in which computers perform the bulk of the review process.
Companies are trying a variety of alternatives to the traditional use of outside law firms to reduce the cost of review, such as using lower-cost attorneys, increasing the rate of review, and grouping documents to streamline review. The study finds limitations in all of these approaches.
Litigants and their counsel are hiring temporary attorneys or using legal process outsourcing (LPO) companies with stables of contract attorneys. But the rates paid to such project attorneys for large-scale reviews in the United States may well have bottomed out, with further reductions of any significant size unlikely. Some companies have turned to LPOs with access to lawyers in other countries, such as India and the Philippines, but this practice may raise concerns about information security, oversight, maintaining attorney-client privilege, and logistics.
The highest estimates of review speed are about 100 documents per hour, and this number assumes that reviewers have the strongest motivations and experience and are examining documents simple enough that a decision on relevance, responsiveness, privilege, or confidentiality could be made in an average of 36 seconds. Given the trade-off between reading speed and comprehension, especially in light of the complexity of documents subject to discovery in large-scale litigation, it is unrealistic to expect much room for improvement in the rates of unassisted human review.
Litigants are also using computerized techniques to organize similar documents together:
In many cases, such techniques organize material rather than reducing the number of documents needing review. Unless they can be grouped to allow bulk coding—in which a decision on one is applied to the others in the same group—these techniques cannot make sufficiently dramatic improvements in review speed needed to minimize review costs as a source of concern.
Existing research suggests that reviewers often disagree with one another when they review the same set of documents for relevance and responsiveness in large-scale reviews. In one case, for example, seven teams of attorneys, all trained in a similar manner and given the same instructions, examined 28,000 documents organized into 12,000 clusters to judge whether the clusters were responsive to the facts of the case. Results showed that the teams agreed only 23 to 54 percent of the time, depending on the pair of teams being compared.
These high levels of disagreement, confirmed in other studies, are caused by human error in applying the criteria for inclusion, rather than ambiguity in the documents or the scope of the production demand.
Predictive coding is a type of computer-categorized review process that classifies documents in terms of how well they match the concepts and terms in sample documents. With predictive coding, attorneys initially examine samples of documents from the review set and make determinations about whether they are relevant, responsive, or privileged. The application then uses those decisions to assign scores to other documents in the review set. When attorneys assess samples of the computer's decisions, the application uses that input to increase the reliability of its decisions. The results of this iterative process are eventually stabilized, at which point the software's decisions should closely match those of human reviewers.
Because this is nascent technology, there are only a few studies that examine its accuracy or its costs compared with those of human review. Studies of its accuracy show that predictive coding identifies at least as many documents of interest as traditional eyes-on review, with about the same level of inconsistency as—and, in some cases, greater consistency than—human review.
Because evidence is scant and vendor pricing models are still in development, it is too early to know the exact level of cost savings that could be achieved with predictive coding. But early studies suggest that the savings could be considerable. One such study examined a document set that had been reviewed by attorneys in the traditional way and reported that predictive coding would have saved about 80 percent in attorney review hours. A reduction of hours to this degree would likely meet the savings target of 75 percent in total review costs.
Another serious challenge presented by the growth in digital information is how much of it to preserve for the purposes of future litigation. Interviews with corporate counsel from the companies in the study highlighted several concerns:
The authors propose three recommendations to address complaints of excessive costs of review and uncertainty about legal responsibilities for preservation.
Adopt computer categorization to reduce the costs of review in large-scale electronic production. Despite its promise, predictive coding has not yet been adopted by litigants, partly because there is a dearth of judicial decisions at the trial court level that clearly address the use of the technology. The authors propose that the best way to bring the technology into the mainstream is for public-spirited litigants to use it—and proclaim its use—for large-scale e-discovery efforts. If litigants were able to win judicial approvals in several jurisdictions, then use of these techniques would spread, the technology would improve, and long-term cost savings would result for the civil justice system as a whole.
Improve tracking of the costs of production and preservation. Without such data, companies cannot develop strategies for dealing with massive data volumes, such as investing in automated legal-hold-compliance systems or advanced analytic software for early case assessment. The rule-making process has been hampered by a lack of reliable cost information related to e-discovery.
Bring certainty to legal authority concerning preservation. Steps must be taken soon to address litigant concerns about complying with preservation duties. The absence of clear, unambiguous legal authority that applies across jurisdictions appears to be thwarting thoughtful preservation efforts and may be leading to overpreservation at considerable cost. Modifying the rules of court seems to be the most practical way to bring certainty to this area.