Some of the enthusiasm surrounding predictive coding and its reputation as a cost-saving measure took a beating in the courtroom of US district Judge Anthony Battaglia, in San Diego, earlier this month. At the end of a four-year patent case, he awarded $2.8 million to prevailing defendant Qualcomm for costs it incurred to review documents by means of a proprietary computer-assisted method.
The bill is part of a $12.4 million award of attorneys’ fees, which included $392,000 to pay contract attorneys to review documents that had been identified as responsive by the technology. The case is believed to be one of the first reported instances in which a judge has ordered losing parties to pay the costs incurred by an adversary to use computer-assisted review.
Plaintiffs Gabriel Technologies and Trace Technologies have appealed the ruling. They argue the computer-assisted review “obviously did not work well.” Judge Battaglia found that their suit for patent infringement and misappropriation of trade secrets against Qualcomm and two other defendants, Snaptrack and Norman Krasner, was frivolous.
Battaglia granted summary judgment in favor of the defendants in September 2012.
The judge said the $2.8 million expense item is due to “a complex computer generated algorithm.” The system was designed and implemented by H5, a San Francisco-based service provider that obtained patents for some of its review process last year.
Though Battaglia’s February 1 ruling says the defendants used a computer-assisted method to review all 12 million documents collected in the case, court documents suggest the $2.8 million fee covered review of a much smaller subset of records.
$2.8 million to review one million documents?
A bill of costs and corresponding invoices filed in October show Qualcomm first paid ProSearch Strategies $840,000 to cull the 12 million documents down to about a million for review concerning their responsiveness.
In its motion for attorneys’ fees last October, Qualcomm said:
“[We] collected almost 12,000,000 [electronic] records…. Rather than review this entire volume, the parties negotiated and agreed to a set of search terms… to cull irrelevant documents from the review population. [We] applied those terms across all the ESI… collected for this case. Rather than manually reviewing the huge volume of resultant records, [we] paid H5 to employ its proprietary technology to sort these records into responsive and non-responsive documents. The H5 algorithm made initial responsiveness determinations for more than one million documents (sic).”
Next came the humans. Having already spent $3.6 million to identify responsive documents, the defendants deployed an army of contract attorneys from the firm Black Letter Discovery to review the rest of the records for privilege and relevance. Black Letter billed Qualcomm for 7,000 hours at rates of $55 to $67 per hour.
In all, the defendants paid more than $4 million to three e-discovery service providers. The court ruled it could recover $3.2 million of that. Qualcomm sought payment of fees from the plaintiffs’ lead attorneys at Hughes Hubbard & Reed, as well. The parties settled this dispute out of court for an undisclosed amount.
Judge Battaglia’s displeasure with the plaintiffs’ case also was evident in the sanction he imposed on plaintiffs’ local counsel, Wang Hartman Gibbs & Cauley, for $64,000 for “failing to undertake a reasonable investigation into the merits” of the case.
Ronald Abramson, an attorney for the plaintiffs at Hughes Hubbard, in New York, did not respond to a request for comment. Latham Watkins partner, Alan Kraus, attorney for Hughes Hubbard and Reed, declined comment.
Jeffrey Karr, attorney for Qualcomm at Cooley in Palo Alto, did not respond to a request for comment.
Costs soar despite predictive coding
The large computer assisted review fee is a high-profile counterpoint to mounting evidence that the technology outperforms traditional methods in price and efficiency. The plaintiffs blame the high costs on “failure to conduct a targeted collection,” a review platform “that obviously didn’t work well,” and excessive billing for vendor-attorney communications.
Qualcomm says if it had employed contract attorneys to review one million documents, “the result would have been far more time-consuming and no less expensive.”
Attorney Jay Yurkiw, chair of Porter Wright’s E-Discovery Practice Group, in Columbus, says court documents don’t always tell the whole story.
“It seems there’s something going on here that we’re not seeing,” he told ACEDS. “The costs seem high, but I would not jump to the conclusion that computer-assisted review is ineffective. That’s not the case.”
Yurkiw said H5 may have developed a custom algorithm for the complex case, and “run it a few times” to get sound results.
H5 says technology cuts costs by 50 to 75 percent
H5 Executive Managing Director, Jeff Kangas, told ACEDS the large volume of documents in the case contributed to the high costs. He said he cannot talk about details of the case, but believes H5 handled “many more” than a million documents for Qualcomm.
“The cost of any matter will correspond to the volume of documents it produces,” he said. “Computer-assisted review eliminates the downstream costs of attorney review, which can cost $2 to $4 per document, if done manually.”
“We typically see cost savings of 50 to 75 percent,” he adds.
Invoices show that H5 billed Qualcomm between $132,000 and $264,000 for “services rendered” on 14 separate occasions between June 30, 2009 and July 31, 2010. H5 delivered about 137,000 documents to Qualcomm between April and June 2011, records show, but it is not clear from the invoices whether that figure represents the total production.
H5’s Kangas cautions that the costs should be viewed in the context of a case now more than four years old. Computer-assisted review methods have evolved and improved during that time, he says.
Attorney gives best practices for cutting review costs
The large sums generated by the case undermine success stories to which computer-assisted review proponents point. The case shows how widely e-discovery costs can fluctuate irrespective of review method.
Matthew Nelson, an e-discovery attorney at Symantec in San Francisco, says many first generation computer-assisted review tools can produce unexpected costs.
“Machine learning has been around for decades and is only now being applied in the legal sphere,” he told ACEDS. “We’re still in the early stages of evolution.”
Nelson said some platforms are not fully automated, which introduces the risk of human error. Others are not transparent and require vendor personnel to validate statistics and perform other tasks at great costs. Most service providers also charge a premium for proprietary predictive coding, he said.
Court awards some but not all e-discovery expenses
The Qualcomm matter is unique because it is one of the first reported cases where a court awarded fees to a prevailing party for expenses incurred in computer-assisted review.
“My one takeaway is that a court was willing to reimburse a party for technology,” says Yurkiw, the Porter Wright attorney.
The framing of these e-discovery expenses — in terms of fees rather than costs — appears to have been a factor in the court’s decision to award them.
“Because [we] would seek compensation for that traditional, attorney-driven responsiveness review, it is reasonable for the Court to award the fees charged to Defendants by H5 for developing and employing a first-level document review algorithm,” Qualcomm said in requesting the award of fees.
Judge Battaglia found this method to be “reasonable under the circumstances.”
However, the court denied in Qualcomm’s bill of costs $929,000 in “fees for exemplification and copies,” including $840,000 in e-discovery costs for “analytical review tasks [performed by ProSearch] such as search term analysis, culling and deduplication.”
The defendants did not sufficiently describe the costs and may not recover them, said Samuel Hamrick, Clerk of the US District Court, in San Diego, in a December 18 cost taxation order.
The plaintiffs had argued that these e-discovery costs are not recoverable under the federal Taxation of Costs law at Title 28, US Code Section 1920(4). Notably, the court awarded computer-assisted review costs under two statutes that allow for recovery of attorneys’ fees, the US Patent Act and a California Statute.