Current expertise has satisfied me that familiarity breeds complacency relating to responding to Antitrust Second Requests. Within the identify of the “distinctive nature” of a Second Request assessment, skilled practitioners usually preserve that solely conventional protocols will work, and unhesitatingly settle for a manufacturing set that could be upwards of fifty% nonresponsive.
Definitely, there are constraints on assessment strategies relating to responding to a Second Request. Each the DOJ and the FTC require advance written discover when “utilizing software program or know-how… to determine or get rid of paperwork, information, or data doubtlessly attentive to [the] Request.”[i] Each likewise demand particular data when the assessment course of depends on both search phrases or technology-assisted assessment. The precise reference to “seed set[s] and coaching rounds” within the Mannequin Second Requests utilized by every company displays a selective, practically exclusionary, choice for TAR 1.0. And the concomitant prohibition within the DOJ’s Predictive Coding Mannequin Settlement in opposition to any responsiveness assessment following the “predictive coding course of” nearly ensures the substantial, pointless manufacturing of nonresponsive paperwork.[ii]
However these constraints ought to by no means inhibit a continuing quest for higher strategies – strategies that make the assessment extra environment friendly or end result within the manufacturing of much less nonresponsive paperwork, or each. Very similar to the know-how titans chronicled in At all times Day One, we can’t slip into Day Two and deal with fiercely defending custom quite than inventing the longer term.[iii] As Jeff Bezos (Amazon) noticed in a 2016 letter to shareholders:[iv]
Day 2 is stasis. Adopted by irrelevance. Adopted by excruciating, painful decline. Adopted by demise. And that’s the reason it’s at all times Day 1.
As an alternative, we have to attempt to repeatedly dwell Day One, and prioritize targeted innovation over custom – notably once we cling to custom just for custom’s sake.
And, as mentioned under, there’s certainly a greater method for responding to Antitrust Second Requests, a extra environment friendly method that successfully focuses on the unique manufacturing of responsive paperwork. There’s a higher method and, frankly, no rational cause to stagnate in custom.
Custom is Stasis
Practitioners sometimes use one among two strategies when responding to Second Requests: both search phrases adopted by linear assessment, or TAR 1.0. Neither is especially environment friendly or efficient. And each present, at greatest, restricted perception into the paperwork produced to the company. Even worse, a TAR 1.0 strategy could be exceedingly dangerous as a consequence of the unavoidable manufacturing of nonresponsive paperwork.
One tangential statement earlier than shifting to the truth of each strategies: the truth that these practitioners who deal with Second Requests use antiquated eDiscovery strategies for assessment debunks the “distinctive nature” of an Antitrust Second Request assessment. It’s, in any case (assuming correct coaching by counsel), only a doc assessment – responsiveness and privilege are removed from international ideas to an skilled doc assessment crew. And the magnitude and compressed manufacturing deadlines related to Second Requests have gotten virtually commonplace touchstones amongst skilled eDiscovery distributors. One advertising and marketing piece for Second Request capabilities highlights the flexibility to course of 45 million paperwork in 106 days. Okay… however I’ve seen eDiscovery distributors able to processing upwards of 30 million paperwork (~33TB), in 30 days. There’s merely nothing actually “distinctive” about responding to a Second Request.
And if the eDiscovery realm has taught us something, we all know that search phrases adopted by linear assessment is an ineffective, inefficient doc assessment method. The Blair Maron research tells us that search phrases usually retrieve solely on the order of twenty % (20%) of the responsive paperwork from a doc assortment.[v] And whereas naysayers usually search to discredit the research – largely on the premise of advances in computerized search know-how – I’ve personally seen educated deserves counsel wrestle to search out thirty % (30%) of the responsive paperwork utilizing search phrases with even probably the most fashionable search know-how. Since 20% or 30% recall will undoubtedly be insufficient for a Second Request, practitioners might want to spend numerous hours refining the search phrases to enhance recall. And with elevated recall comes decreased precision – a direct consequence of the precision-recall tradeoff.[vi] Virtually, that signifies that each point-increase within the recall share will lower the precision within the search time period assessment set by some quantity. And, for the reason that Blair Maron research put noticed precision at roughly 80%, way more than 20% of the assessment set will likely be nonresponsive at greater recall ranges. That straight will increase assessment effort, and reduces effectivity.
Qualitatively, the search time period strategy is much more regarding. Typical Second Request doc volumes and deadlines necessitate a considerable variety of reviewers. Advertising supplies for one giant case tout the engagement of greater than 300 reviewers for a single Second Request response. That presents two sensible issues. First, extra reviewers merely imply extra inconsistency. This may be notably disconcerting when privilege calls are missed. Second, with paperwork being unfold indiscriminately amongst so many reviewers, there isn’t a alternative for gaining any actual perception into the character of the paperwork which are being produced to the company. Within the context of a fast-paced Second Request, this will imply the distinction between preparation and bare response throughout negotiations.
A TAR 1.0 strategy, alternatively, could also be extra environment friendly (by way of the variety of paperwork reviewed to attain manufacturing), however will undoubtedly be much less efficient and fewer protecting. At an affordable recall stage, TAR 1.0 isn’t notably exact. It isn’t in any respect unusual to see precision ranges at lower than 50%. And advertising and marketing supplies for one giant case recommend that precision for some collections might be lower than 30% at recall ranges of solely 75%. That signifies that upwards of half of a manufacturing to the company will likely be nonresponsive, notably since any subsequent responsiveness assessment is prohibited.
Qualitatively, a TAR 1.0 strategy could be even worse than a search time period strategy. Poor precision in a TAR 1.0 assessment doesn’t enhance assessment – it will increase publicity. Nonresponsive paperwork are by no means reviewed; they’re produced on to the company. Consequently, each difficulty contained in each nonresponsive doc is straight uncovered to the company. And, since solely a small fraction of the gathering is reviewed to develop the TAR mannequin, there’s even much less perception into the substance of the manufacturing.
In the end, conventional assessment strategies don’t properly serve the spectrum of targets attendant to Second Request opinions. Neither method adequately optimizes effectivity, effectiveness and safety.
Day One, Innovating a Higher Means
To only say that there’s a higher method to answer a Second Request is an understatement. An revolutionary speedy analytic investigative assessment (RAIR) that mixes refined workflows with the adept software of analytics, and built-in and focused sampling, will collectively optimize for all three – effectivity, effectiveness and, maybe most significantly (notably given the deficiencies of conventional approaches) safety.
So, how does a RAIR assessment work?
The spine of a RAIR assessment is a small, subtle crew devoted to the investigatory, analytic, statistical evaluation of substantively comparable tranches of paperwork – sometimes lower than ten crew members. A couple of crew could also be vital, relying on the quantity and homogeneity of the gathering. However, on condition that one crew can sometimes assess just a few hundred thousand paperwork within the span of only a week, seldom will any assessment require greater than three groups.
It practically goes with out saying that the concentrated and targeted character of the RAIR crew will enhance consistency, notably over huge 300 individual opinions. And an ingrained observe of fixed communication and collaboration inside and among the many groups solely serves to additional promote not solely consistency, but in addition decision-making – drawing on the collective knowledge of the crew(s), versus the remoted particular person determinations of a single reviewer.
Every tranche of paperwork is derived utilizing all obtainable analytics, basically by aggregating units of paperwork which are substantively comparable from the angle of responsiveness. This strategy usually ends in the creation of doc units that mix 1000’s, and even tens of 1000’s, of comparable paperwork for a single choice – responsive or nonresponsive. (I’ve personally seen one state of affairs the place greater than 2 million nearly an identical paperwork have been aggregated.) This aggregation course of continues till the whole assortment has been evaluated. And the premise for aggregating every doc set is recorded to assist the defensibility of the method.
All through the aggregation course of, this targeted evaluation mechanically instills beneficial insights into the context and substance of the paperwork within the assortment, way over any single particular person would possibly garner utilizing conventional assessment strategies. This steady and well timed information of the contents of, notably, the paperwork being produced to the company could be vital to advance planning, and totally defending the consumer’s pursuits in negotiations with the company.
As doc units are aggregated, random samples of every set are generated to supply the premise for a bulk responsiveness evaluation. A minimum of one pattern having a confidence stage of 95% and a confidence interval of ±5% needs to be drawn from every set. A couple of pattern could also be drawn from bigger or extra various (much less homogenous) doc units, and small doc units could also be reviewed of their entirety.
These consultant samples are then reviewed for responsiveness. If the whole pattern is constant and proper, the whole doc set is coded accordingly. If the pattern isn’t wholly constant, the unique doc set will likely be reassessed and, if vital, additional separated into responsive and nonresponsive units. New samples of each will then be drawn and reviewed for a subsequent iteration. All through this pattern assessment course of, every doc might also be evaluated for privilege.
The statistical implications of this strategy for the responsiveness assessment are noteworthy. The responsiveness choice for each aggregated doc set is basically validated with a 95/5 pattern. In consequence, a RAIR assessment drives superior ranges of recall and precision, usually reaching better than 90% on each. That signifies that the company will get nearly all the things it could be entitled to underneath the Second Request (responsive), and nothing that doesn’t in any other case must be produced (nonresponsive).
Lastly, along with the preliminary privilege evaluation in the course of the responsiveness pattern assessment, privilege is definitely topic to a lot nearer scrutiny. Paperwork will not be merely reviewed independently; analytics are used to determine the traits (subject material, timing, individuals, and so forth.) underlying contextual privileges, and investigatory strategies are then used to search out the privileged paperwork. This strategy ensures the utmost consistency and protection, notably given the symbiotic operation of the RAIR crew and the intensive, targeted evaluation underlying the investigation for privileged paperwork.
As soon as all the aggregated doc units have been assessed, and the privilege evaluation is full, a clear concise set of paperwork virtually solely attentive to a Second Request could be produced to the company.
Abandon Custom for Innovation
The relative advantages of a RAIR assessment over conventional assessment strategies for responding to a Second Request are easy. Fewer paperwork are reviewed. Extra responsive and fewer nonresponsive paperwork are produced. Privilege safety is bigger. And a RAIR assessment gives insights into the substance of the manufacturing that will in any other case by no means be obtainable. “That’s the reason it [should always be] Day 1.”
[i] See https://www.justice.gov/atr/file/706636/obtain (DOJ Mannequin Second Request, Instruction 4) and https://www.ftc.gov/system/information/attachments/hsr-resources/model_second_request_-_final_-_october_2021.pdf (FTC Mannequin Second Request, Instruction 15).
[ii] See https://www.justice.gov/file/1096096/obtain (DOJ Predictive Coding Mannequin Settlement).
[iii] Alex Kantrowitz, At all times Day One: How the Tech Titans Plan to Keep on High Endlessly.
[iv] See https://www.vox.com/2017/4/12/15274220/jeff-bezos-amazon-shareholders-letter-day-2-disagree-and-commit.
[vi] See, e.g., https://datascience-george.medium.com/the-precision-recall-trade-off-aa295faba140