Migrated from eDJGroupInc.com. Author: Greg Buckles. Published: 2014-09-21 20:00:00Format, images and links may no longer function correctly.
Mike Simon’s recent blogs really resonate with my take-aways from all of my interviews and survey results. The market is over-saturated with promises like “Review only 5,000 items!” and “Save 75% of review cost!” Although all of my cutting edge corporate eDiscovery interviewees had used some kind of machine learning in a few cases, none felt that PC-TAR was appropriate for ALL of matters. The dominant reason for using PC-TAR was to meet impossible deadlines or overwhelming volumes of raw ESI. PC-TAR was not considered easier or cheaper when you factored in the cost of the up front analytic processing ($130-200/GB extra), technical experts to manage the process and the cost to negotiate the protocol with the opposing party. So why did they go through all that trouble if they did not anticipate real savings? Because they did not have a choice. Many cited very tight time frames in recent regulatory requests. Others talked about being ‘dump trucked’ by huge opposing productions with bare weeks before depositions. My law firm interviews revealed a struggle between conservative partners and progressive litigation support evangelists. The eDiscovery providers gave me explicit descriptions of conversations with law firm partners with quotes like, “Your ROI for this stuff comes out of my pocket!”
So here are my Top Ten Reasons Why NOT PC-TAR:
- Perception that PC-TAR costs front load the discovery cost for matters that WILL settle before trial.
- High resistance to analytic upcharges. Have to justify them on every matter, so go with path of least resistance.
- Complexity of systems and fear that counsel will not be able to defend what they do not understand.
- Customers on information overload. Marketing fatigue and growing customer indifference.
- Perception that PC-TAR reinforces known relevant selection and misses unknown/new documents.
- Rumors of SEC/DOJ in some areas fighting PC-TAR proposals.
- Realization that 95-99% recall in PC-TAR training will result in 300-500% production size. Exposure of large volumes of non-relevant ESI a serious concern for companies facing serial plaintiffs that are on fishing expeditions.
- Mature corporate customers already cull and optimize during collection or processing. If they can achieve substantial savings prioritizing/clustering review sets, why pay for actual PC-TAR analytics?
- Counsel does not want to operate PC-TAR systems. Wants Litsupport or provider to run it.
- PC-TAR takes money from the firm. Takes away associate jobs.
Do I believe that most of these market perceptions are true. No, I do not. However, they all contribute to the slow adoption rates for actual use of machine learning technology in traditional discovery review for production. Since this phase of the eDiscovery lifecycle is so frequently cited as being responsible for 75% of the cost of discovery, you can see why many providers I spoke with had become disenchanted with sales evangelism focused on PC-TAR. Providers know that they must be able to host and support some kind of PC-TAR, but with kCura’s Relativity becoming the default review platform for providers that is easy. They have backed off pushing PC-TAR in the face of direct negative feedback from some firms and resistance from corporations who expect the matters to settle. The use of these technologies for ECA, investigations, regulatory responses and analysis of opposing productions will continue to slowly eat away the fear, uncertainty and doubt that have kept the adoption rate for machine learning review at such a low level.
Greg Buckles can be reached at Greg@eDJGroupInc.com for offline comment, questions or consulting. His active research topics include analytics, mobile device discovery, the discovery impact of the cloud, Microsoft’s 2013 eDiscovery Center and multi-matter discovery. Recent consulting engagements include managing preservation during enterprise migrations, legacy tape eliminations, retention enablement and many more.