Never Second Guess A Second Request: Leveraging Predictive Coding For Reviewing Documents In Antitrust Matters

A unique challenge for today’s antitrust attorney is how to best utilize analytics such as predictive coding—often referred to as Technology Assisted Review or TAR ¹Note: The term “Technology Assisted Review” or “TAR” can also be interrupted to encompass document review analytics technologies beyond predictive coding, such as search analytics, email threading, and concept clustering, etc.—to comply with a Second Request in an efficient and timely manner.

With a 25 percent year-over-year increase in merger filings in 2014 (1,663 reported transactions—the highest number since 2008) and an 8.5 percent increase in the number of Second Requests (51 Second Requests in 2104 versus 47 Second Requests in 2013) ²See Kroll Ontrack, Top 10 Takeaways from the 2014 HSR Annual Report, available at http://www.ediscovery.com/cms/hsr/index.html., now is the time for legal professionals to understand new and innovative methods to drive the best outcome when seeking regulatory approval of a merger.

What is a Second Request?

Under the Hart-Scott-Rodino (HSR) Antitrust Improvements Act of 1976, most corporate mergers that meet a threshold financial requirement are subject to review by federal antitrust agencies—the Department of Justice (DOJ) or the Federal Trade Commission (FTC). After the parties file a premerger notice with both of the agencies, a 15 to 30 day waiting period prohibits consummation of the merger transaction. ³See 15 U.S.C. 18a(b) (2000). During the waiting period, the FTC or DOJ reviews the initial information provided by the parties to determine if further information is required to assess whether there may be a violation of federal antitrust laws. If more information is needed, the DOJ or FTC can issue requests for “additional information and documentary material relevant to the proposed acquisition,” generally known as Second Requests. ⁴See 15 U.S.C. §18a(e)(2).

The process is typically a burdensome task for merging organizations, as it calls for companies to review, analyze and produce massive volumes of data within a very short time frame. Merging companies must meet the “substantial compliance” requirement of the agencies. ⁵See 15 U.S.C. 18a(e) (2000).

While the merging companies are able to decide when to respond to a Second Request, most are highly motivated to act quickly and work cooperatively with the agency to meet its requests due to contractual and market factors. The longer the wait, the longer it will take for a decision on whether the merger will move forward, and there are added pressures from stockholders or the board of directors to get the deal done.

Furthermore, due to the legal thresholds applied by the agencies for issuing a request, Second Requests tend to be larger in scale than most civil litigation matters. ⁶See Tracy Greer, Technology-Assisted Review and Other Discovery Initiatives at the Antitrust Division, Dept. of Justice, Antitrust Div. 1, available at http://www.justice.gov/sites/default/files/atr/legacy/2014/03/27/304722.pdf. While the parties are motivated to set short time frames, they also know they are likely to be dealing with extremely broad requests for information and must allocate enough resources and time to collect, process and produce the relevant documents and data.

What is Predictive Coding?

Predictive coding is one of the most effective tools under the umbrella of TAR. Predictive coding is a software application that applies mathematical computations (i.e., an algorithm) against a document set in order to categorize it based on similarity of characteristics. The computer “learns” from human input and then applies this information to the entire document population.

Common document classifications such as “responsive,” “not responsive” or “privileged” allow review teams to prioritize and channel documents to specific human subject matter experts as needed for further calibration.

When done well, this technology finds key documents significantly faster, with a much lower number of human reviewers, thereby saving hours, days and potentially weeks of document review.

It also allows legal teams to find the significant documents much earlier in the review process, which can positively impact the case strategy and aid the negotiation process regarding production volumes, timing and scheduling.

The use of predictive coding in large document reviews in civil cases is becoming commonplace, with the judiciary stating, “[i]t is now black letter law that where the producing party wants to utilize TAR for document review, courts will permit it.” ⁷See Rio Tinto PLC v. Vale S.A., 2015 BL 54331 (S.D.N.Y. Mar. 2, 2015). Likewise, Second Requests are particularly ripe for the use of predictive coding.

What Do The Agencies Require?

The FTC highly encourages merging parties to produce information early—either with or after the HSR filing. Providing information ahead of schedule may assist the agency with either narrowly tailoring the Second Request, which would reduce the burden on the merging parties, or may obviate the need for a Second Request altogether. The FTC for example has provided a list of key information that parties should consider submitting during the initial waiting period. ⁸See Best Practices, supra note 4, at 1–2 (including strategic and marketing plans for the previous two years, a list of all currently manufactured, marketed or sold products, and products in development, a list of top 10 customers with contact information for customers of overlapping products, a list of competitors with contact information for competitors who provide overlapping products, market share information for overlapping products, a list of the types of reports the company prepares on a regular basis, an organization chart, and a data map of the types of data the party creates and stores). Further, due to the short time frame common among Second Requests, the FTC points to the effective use of a “pull and refile” to reset the HSR timing provisions. ⁹See id. at 2. This process allows an acquiring party, upon written notice, to withdraw its HSR filing and refile once without paying an additional filing fee, and the FTC has suggested that using this procedure reduces the likelihood that a Second Request will issue—or that a Second Request, if issued, will be narrower in scope. ¹⁰See id.

Agency Expectations.

The negotiating posture of the merging parties and the antitrust agencies can be quite different from the civil litigation context. The FTC acknowledges its duty to negotiate Second Requests “in good faith” to reduce the burden on the parties while still allowing the agency to obtain the necessary information to evaluate the proposed merger. ¹¹See id. at 2–4. Moreover, the FTC specifically addresses the use of predictive coding and search terms during the Second Request process.

The FTC will not agree that a particular list of search terms is sufficient for the company to certify compliance with a Second Request. Instead, the agency requires the merging parties to provide a list of search terms in advance of producing documents so it can comment on the proposed terms. ¹²See id. Even after reaching an agreement on the specific searches to be run, the results of the searches are likely to yield a high volume of documents, with only a small portion likely to be responsive to the request.

While the antitrust agencies have addressed predictive coding in the past, ¹³See Greer, supra note 5, at 2–6. the FTC recently published a new Model Second Request that includes guidelines regarding the recommended methodology to employ, including specific metrics and validation steps to evaluate the quality of its production stemming from the use of predictive coding. ¹⁴See Model Request, supra note 4, at 15–16 (explaining the description of collection methodology should include (a) how the software was utilized to identify responsive documents; (b) the process the Company utilized to identify and validate the seed set documents subject to manual review; (c) the total number of documents reviewed manually; (d) the total number of documents determined nonresponsive without manual review; (e) the process the Company used to determine and validate the accuracy of the automatic determinations of responsiveness and nonresponsiveness; (f) how the Company handled exceptions (“uncategorized documents”); and (g) if the Company’s documents include foreign language documents, whether reviewed manually or by some technology-assisted method and that companies should provide all statistical analyses utilized or generated related to the precision, recall, accuracy, validation, or quality of the document production in response to the Second Request). Incorporating these practices into the workflow during Second Request document reviews that leverage predictive coding is essential to complying with the antitrust agencies’ guidelines and quickly meeting the substantial compliance requirement. In addition, by designing a predictive coding workflow that meets or exceeds the agencies’ requirements, the risk that additional collections or supplemental productions will be needed may be reduced.

Taking its predictive coding recommendations one step further, the antitrust agencies have actually embraced the use of predictive coding. A speech from the Deputy Assistant Attorney General discussed its use in Second Requests: “When it works well, predictive coding reduces the document review and production burden on parties while still providing the [DOJ] with the documents it needs to fairly and fully analyze transactions and conduct. Of course, for predictive coding to work for the [DOJ], we require a high degree of cooperation and transparency about the implementation and structure of the predictive coding process.” ¹⁵See Deputy Assistant Attorney General Renata B. Hesse, IP, Antitrust and Looking Back on the Last Four Years, Speech at Global Competition Review (Feb. 8, 2013), available at http://www.justice.gov/atr/file/518361/download.

The antitrust agencies see increased value in the use of predictive coding for Second Requests. The FTC includes specific directions in its latest Model Second Request outline and has published Best Practices for Merger Investigations. ¹⁶See Federal Trade Commission, Model Request for Additional Information and Documentary Material (Second Request) 15–16 (Aug. 2015) [ Model Request], available at https://www.ftc.gov/system/files/attachments/merger-review/guide3.pdf; and Federal Trade Commission, Best Practices for Merger Investigations (Aug. 2015) [Best Practices], available at https://www.ftc.gov/system/files/attachments/merger-review/best_practices_for_ merger_investigations_august_2015.pdf.

Implications of Noncompliance

The Hart-Scott-Rodino Act allows for parties to provide a statement of reasons for noncompliance in responding to Second Requests. ¹⁷See 16 C.F.R.§ 803.3 (2000). While the standard is substantial—as opposed to perfect—compliance, the agencies do require parties to justify why an answer is not provided or is incomplete.

Parties seeking a speedy resolution of the merger review process should proceed carefully, as hurriedly producing documents without full compliance can lead to substantial civil penalties, ¹⁸See 15 U.S.C. §18a(g) (2000) the potential rejection of the merger transaction, ¹⁹See, e.g., FTC v. McCormick & Co., 1988-1 Trade Cas. (CCH) para. 67,976 (D.D.C. 1988) (ordering McCormick to comply with the issued Second Request and prohibiting the consummation of the transaction until the request had been met). or the loss of valuable time as the parties seek merger approval.

Predictive Coding Done Right: A Case Study

While best practices for employing predictive coding in the civil litigation context may closely align with agency guidance, the FTC’s new guidelines should be followed closely to ensure that standards for constructing the predictive coding population for sampling, seeding, training and validation, meet or exceed agency standards. As an eDiscovery service provider with extensive experience with Second Request matters, our company has partnered with merging parties and their counsel to successfully utilize predictive coding in producing documents to the agencies. The following case study demonstrates that while the agency standards may seem stringent and burdensome, in fact they are practical components of a reliable and defensible predictive coding workflow.

A recent technology sector merger utilized our predictive coding technology and review platform to respond to an FTC Second Request. The case serves as a prototype for implementing predictive coding best practices. Specifically, the merger presented a number of challenges, including: a large data set, multiple jurisdictions, complicated review guidance based on the document requests and a shorter than originally planned production deadline.

To help meet the short time frame, the parties decided to leverage predictive coding to ensure that their document review costs were kept to a minimum while remaining compliant with the Second Request deadlines. Regarding their use of predictive coding, the companies submitted to the FTC information relating to the “statistical analyses utilized or generated related to the precision, recall, accuracy, validation or quality of the document production.” ²⁰See Model Request, supra note 10, part (g).

The first step in setting up the workflow involved creating a random document sample to provide a baseline for predictive coding.

From the approximately 600,000 searchable documents, a random sample of approximately 2,300 documents was generated, sampled at a 95 percent confidence level and at 2 percent margin of error. The sample population document coding provided an estimate for the number of responsive documents in the matter.

In addition to the 2,300 random documents, the document review team trained approximately an additional 6,000 documents to help guide the predictive coding system in making suggestions. A variety of methods were used to identify further documents for training, including using e-mail threading, concept searching and keyword searching, analyzing previous document coding/suggestions and leveraging the most highly responsive documents.

Next, the categorization process began by designing a prioritization review based on the documents trained. The review leveraged the predictive coding rankings based on what the system “learned” about the likelihood of being responsive from the trained documents.

The highest ranked documents were directed to the review team for review, and learning sessions were run automatically on a periodic basis allowing for adjustment to the system learning process.

Finally, the trainers analyzed the system-generated effectiveness measures and sampled the low-scoring documents for exclusion. Once the results were acceptable, the producing party moved ahead with production.

The predictive coding workflow successfully ranked documents effectively and the team reviewed the most relevant documents first and determined which documents did not need to be reviewed; which aided the merging companies in meeting the FTC’s deadline and reducing overall costs of review.

Conclusion

For merging parties, Second Requests and large scale antitrust eDiscovery projects are synonymous with complexity, aggressive deadlines and enormous data volumes. Predictive coding technology can alleviate some of these burdens by dramatically reducing document review time and allowing for measurably lower document review costs while raising accuracy beyond traditional linear review techniques.

As such, predictive coding and other review technologies can play a vital role in delivering efficient, accurate and defensible Second Request productions to the government antitrust agencies. Working with an eDiscovery provider with significant experience with technology assisted review can certainly benefit companies looking for the best outcome when seeking regulatory approval of a merger.