e-Discovery document review: what should counsel outsource?

Earlier this week I blogged about placing the locus of control for e-discovery decisions in the right hands to ensure that the decisions made pass muster in court. To illustrate the potential impact of moving the locus of control for certain decision to an outsource partner let’s compare the document review solutions offered by H5 and Inference Data.

Gold standard counsel or expert linguists - who should take the lead? Photocredit: jeffisageek’s photostream

Gold standard counsel or Linguist - who should decide?

Both H5 and Inference enable users to improve results and potentially save vast amounts of money by teaching sophisticated software how to do document review faster and more accurately than human reviewers can. And the more the review process can be reliably automated, the more money is saved down the road because the amount of manual review is reduced. This all assumes that the software is trained correctly, of course. Which frames a locus of control question: Who’s best at training the software?

Last month I attended a webinar presented by H5. One thing that struck me as distinctive about H5 is their standard deployment of a team of linguists to improve detection of responsive documents from among the thousands or millions of documents in a document review. During the webinar I submitted a question asking what it is their linguists do that attorneys can’t do themselves. One of their people was kind enough to answer, more or less saying “These guys are more expert at this query-building process than attorneys.” Ouch.

I’ve long prided myself on my search ability (ask me about the time I deployed a boolean double-negative in a Westlaw search for Puerto Rico “RICO” cases) and I’m sure many of my fellow attorneys are equally proud. However, I know people (or engineers, anyway) who are probably better at search than I am, and I know one or two otherwise blindingly brilliant attorneys who are seriously techno-lagged. More importantly, attorneys typically have a lot on their plates, and search expertise on a nitty-gritty “get the vocabulary exactly right” level is just one of a thousand equally important things on their minds, so it’s not realistically going to be a “core competency.” So I can see the wisdom in H5’s approach, although I wonder how many attorneys are willing to admit right out loud that they are better off outsourcing this competency.

The other side of the H5 coin is represented by Inference Data, which offers a tightly designed software solution which enables attorneys themselves to become the locus of control for search. For counsel with the proper training and technical aptitude, this strikes me as a killer combination, placing the locus of control — teaching the software to find the right documents — in the hands of the attorney who is the “gold standard” subject matter expert.

I can see where, depending on a number of different factors, either solution might be better. I encourage anyone facing this choice to make an informed decision about which approach leads to the best results rather than relying on their knee-jerk reaction.

e-Discovery outsourcing 101: who makes which decisions?

Because e-discovery is complex, and the penalties for screwing it up are significant, the following choice should be considered periodically by attorneys, clients and IT people involved in e-discovery: “Do we do this piece of the project with the people we have already, or do we add people to our payroll who do this, or do we bring in an outside partner to do this?” This is when the IT people reading this post will start muttering the cliché truism “Build or Buy?” which means choosing between “do it ourselves” and finding a pre-packaged solution.

In a generalized “leadership” or “management” frame of mind the basic choice is: “Do, Delegate, or Dump.” I am fond of characterizing this choice as the assignment of the locus of control for decision-making, where an important consideration is who will do the best job of making the decisions once given that responsibility.

  • “Do” = Must I make a particular set of decisions myself – are those decisions an essential part of my role in the organization, and am I the one with the right information and motivation to make these decisions?
  • “Delegate” = Can someone else do just as well, or perhaps a better job, with making this set of decisions, especially after making these decisions became an essential part of their role?
  • “Dump”: should we even be in the business of making these decisions at all, or can we just drop that issue off of our plates somehow?

For example, one can dump having a company picnic to save money. One can’t dump bookkeeping, however, even in a very small company. But even in a very small company a leader can usually delegate or outsource primary responsibility for bookkeeping and expect to get good results while focusing on core competencies of the business such as production, customer relationships, and motivating team members.

Ultimately the choice boils down to this: Do I want to possess and maintain expertise in making certain decisions, at a certain level of granularity, as a core competency? If yes, then I must make it a core competency, which means investing the time, attention, and education it takes to do it right. If no, then I should bring in someone else who has that core competency and who is invested in doing it right.

In e-discovery, answering the question of what can be outsourced — or where to place the locus of control for decision-making — gets even more interesting since courts hold attorneys personally responsible not only for delivering high-quality document production results but for understanding and directing the process by which results are achieved. So the question becomes: Will attorneys generate better document production results when they personally control more of the process (for example, by personally, hands on the keyboard, deriving and executing search methodology)? Or, will they generate better results by collaborating more with outsourced experts, directing and supervising but delegating more of the hands-on decisions?

More than a few attorneys reading this might find that the choice is not as cut and dried as they think. In my next post I’ll explore this choice by applying the core competency / locus of control standard to competing document review automation solutions from Inference Data and H5.

TREC and the gold standard for document review

Ron Friedman recently blogged an excellent critique of TREC Legal Track’s effort to objectively assess eDiscovery document review practices. Like Ron, I commend TREC Legal Track while wishing to offer comments that may contribute to their success. Like me, Ron is an attorney with long experience working in the technology sector, although for comparison with his math background I can only claim four years of college courses concerning statistical methods for assessing human behavior.

Benchmarking is valuable almost everywhere.

Benchmarking is valuable almost everywhere.

I strongly recommend reading Ron’s post for the benefit of his insights, whether or not you are already familiar with TREC Legal Track. I’d also like to offer my own observations about TREC Legal Track’s finding of low consistency between document classification decisions made by subject matter experts, who are spoken of as “gold standard” reviewers, and ordinary legal document reviewers. (In TREC Legal Track’s study, ordinary reviewers were 2nd and 3rd year law students. In real life the subject matter expert role is played by in-house or outside counsel, while much of the actual review work is performed by contract or outsource attorneys.)

Generally speaking, quality control processes involve benchmarking against some standard. Mechanical processes can be meaningfully benchmarked by physically sampling output (this is the essence of Six Sigma, in particular). For example, as machine parts come off an assembly line, samples can be selected and measured and the variance between their actual size and target size monitored not only to detect defects but to flag the processes responsible for defects. Human processes can also be benchmarked in a variety of ways. (This is in part the province of ITIL, the “Information Technology Information Library,” and the basis for the idea of “service level agreements”.) For example, those responsible for a customer service center may track the number of issues handled per hour, the type of issues handled, the number of resolutions or escalations per issue, revenue gained or lost per issue, etc.

Unfortunately, “responsiveness” and “privilege” are not only somewhat subjective in document review, standards for responsiveness and privilege will vary from case to case. For this reason standards need to be developed “on the fly” for each case, and these standards will by necessity be arbitrary (aka subjective) to some degree even if consistently applied. The good news is that the latest generation of document clustering software applications incorporate tools for developing consistent document review standards on the fly. Through an iterative feedback loop, the humans educate the machines to look for documents with certain characteristics, while the machines force the humans to refine their conception of responsiveness and privilege to a degree that the machine can reliably model it. After enough iterations have passed and the machine has reached some measurable standard of consistency, the humans can step back and let the machine do the rest of the review work. The machine does it more consistently than human reviewers could themselves, and at a much lower cost.

With document review the very idea of defining a “gold standard” for classification is less useful than it sounds. For instance, even if a panel of leading legal scholars could be formed for each eDiscovery matter, the mere fact that someone legitimately may be called a leading scholar doesn’t mean that their views will be consistent with anyone else’s — just well reasoned. But a “gold standard” is not what’s important here. What’s important is that in each case the attorneys responsible for responding to a document request do everything they can to carefully define and consistently enforce reasonable document review standards. This is what the current crop of document clustering applications are intended to do. That is the current model, anyway. I don’t pretend to be able to name the vendors who can or cannot deliver on this promise, although I think this will be the number one question in eDiscovery technology before long.

UPDATE: I discuss TREC’s role in forumulating new legal procedural rules for e-discovery in a later blog post, Catch-22 for e-discovery standards?

Cloud-seeding: SaaS data classification via Panda Security’s new anti-virus offering

Panda Security recently released (in beta form) what it claims is the first cloud-based anti-virus / anti-malware solution for Windows PCs. Not only does it sound like a clever tool for data loss prevention, but it demonstrates another way in which information service providers can aggregate individual user data to develop classifications or benchmarks valuable to every user, a mechanism I’ve explored in previous blog posts.

In essence, every computer using Panda’s Cloud Antivirus is networked together through Panda’s server to form a “collective intelligence” for malware detection and prevention. Here’s how it works: users download and install Panda’s software – it’s a small application known as an “agent” because the heavy lifting is done on Panda’s server. These agents send reports back to the Panda server containing information about new files (and, I presume, related computer activity which might indicate the presence of malware). When the server receives reports about previously unknown files which resemble, according to the logic of the classification engine, files already known to be malware, these new files are classified as threats without waiting for manual review by human security experts.

Security Camera
Sampling at the right time and place allows proactive decision making.

For example, imagine a new virus is released onto the net by its creators. People surfing the net, opening emails, and inserting digital media start downloading this new file, which can’t be identified as a virus by traditional anti-virus software because it hasn’t been placed in any virus definitions list yet. Computers on which the Panda agent has been installed begin sending reports about the new file back to the Panda server. After some number of reports about the file are received by Panda’s server, the server is able to determine that the new file should be treated as a virus. At this point all computers in the Panda customer network are preemptively warned about the virus, even though it has only just appeared.

According to Panda’s April 29, 2009 press release:

Utilizing Panda’s proprietary cloud computing technology called Collective Intelligence, Panda Cloud Antivirus harnesses the knowledge of Panda’s global community of millions of users to automatically identify and classify new malware strains in almost real-time. Each new file received by Collective Intelligence is automatically classified in under six minutes. Collective Intelligence servers automatically receive and classify over 50,000 new samples every day. In addition, Panda’s Collective Intelligence system correlates malware information data collected from each PC to continually improve protection for the community of users.

Because Panda’s solution is cloud-based and free to consumers, it will reside on a large number of different computers and networks worldwide. This is how Panda’s cloud solution is able to fill a dual role as both sampling and classification engine for virus activity. On the one hand Panda serves as manager of a communal knowledge pool that benefits all consumers participating in the free service. On the other hand, Panda can sell the malware detection knowledge it gains to corporate customers – wherein lies the revenue model that pays for the free service.

I have friends working in two unrelated startups, one concerning business financial data and the other Enterprise application deployment ROI, that both work along similar lines (although neither are free to consumers). Both startups offer a combination of analytics for each customer’s data plus access to benchmarks established by anonymously aggregating data across customers.

Panda’s cloud analytics, aggregation and classification mechanism is also analogous to the non-boolean document categorization software for eDiscovery discussed in previous posts in this blog, whereby unreviewed documents can be automatically (and thus inexpensively) classified for responsiveness and privilege:

Deeper, even more powerful extensions of this principle are also possible. I anticipate that we will soon see software which will automatically classify all of an organization’s documents as they are created or received, including documents residing on employees laptop and mobile devices. Using Panda-like classification logic, new documents will be classified accurately whether or not they are of an exact match with anything previously known to the classification system. This will substantially improve implementation speed and accuracy for search, access control and collaboration, document deletion and preservation, end point protection, storage tiering, and all other IT, legal and business information management policies.

The Evolution of eDiscovery Analytics Models, Part II: A Conversation with Nicholas Croce

I recently had the pleasure of speaking with Nicholas Croce, President of Inference Data, a provider of innovative analytics and review software for eDiscovery, following the company’s recent webinar, De-Mystifying Analytics. During our conversation I discovered that Nick is double-qualified as a legal technology visionary. He not only founded Inference, but has been involved with legal technologies for more than 12 years. Particularly focused on the intersection of technology and the law, Nick was directly involved in setting the standards for technology in the courtroom through working personally with the Federal Judicial Center and the Administrative Office of the US Courts.

I asked to speak with Nick because I wanted to pin him down on what I imagined I heard him say (between the words he actually spoke) during the live webinar he presented in mid-March. The hour-long interview and conversation ranged in topic, but was very specific in terms of where Nick sees the eDiscovery market going.

Sure enough, during our conversation Nick confirmed and further explained that he and his team, which includes CEO Lou Andreozzi, the former LexisNexis NA (North American Legal Markets) Chief Executive Officer, have designed Inference with not one, but two models of advanced eDiscovery analytics and legal review in mind.

As data volume explodes, choosing the right way to sift it becomes urgent
As total data volume explodes, choosing the right way to sift out responsive documents becomes urgent

Please read my previous blog post The Evolution of eDiscovery Analytics Models, Part I: Trusting Analytics if you haven’t already and want to understand more about the assertions I make in this blog post.

In a nutshell, Inference is designed not only to deliver the current model of eDiscovery software analytics, which I have dubbed “Software Queued Review,” but the next generation analytics model as well, which I am currently calling “Statistically Validated Automated Review” (Nick calls it “auto-coding”).

Bruce: In a webinar you presented recently you explained statistical validation of eDiscovery analytics and offered predictions concerning the evolution of the EDRM (“Electronic Discovery Reference Model”).

I have a few specific questions to ask, but in general what I’d like to cover is:

1) where does Inference fit within the eDiscovery ecosystem,
2) how you think statistically validated discovery will ultimately be used, and
3) how you think the left side of the EDRM diagram (which is where document identification, collection, and preservation are situated) is going to evolve?

Nick: To first give some perspective on the genesis of Inference, it’s important to understand the environment in which it was developed. Prior to founding Inference I was President of DOAR Litigation Consulting. When I started at DOAR in 1997, the company was really more of a hardware company than anything else. I was privileged to be involved in the conversion of courtroom technology from wooden benches to the efficient digital displays of evidence  we see today,  Within a few years we became the predominant provider of courtroom technology, and it was amazing to see the legal system change and directly benefit from the introduction of technology. As people saw the dramatic benefits, and started saying “how do we use it?” we created a consulting arm around eDiscovery which provided the insight to see that this same type of evolution was needed within the discovery process.

This began around 2004-2005 when we started to see an avalanche of ESI (“Electronically Stored Information”) coming, and George Socha became a much needed voice in the field of eDiscovery. As a businessman I was reading about what was happening, and asking questions, and it seemed black and white to me – it had become impossible to review everything because of the tremendous volume of ESI with existing technology. As a result I started developing new technology for it, to not only manage the discovery of large data collections, but to improve and bring a new level of sophistication to the entire legal discovery process.

Inference was developed to help clients intelligently mine and review data, organize case workflow and strategy, and streamline and accelerate review. It’s the total process. But, today I still have to fight “the short term fix mentality” – lawyers who just care about “how do I get through this stuff faster”, which is the approach of some other providers, and which also relates to the transition I see in the EDRM model – I want to see the whole thing change.

Review is the highest dollar amount, the biggest pain, 70% of a corporation’s legal costs are within eDiscovery. People want to, and need to, speed up review. However, we also need to add intelligence back into the process.

Bruce: What differentiates Inference, where does it fit in?

Nick: I, and Inference, went further than just accelerating linear review and said: it has to be dynamic, not just coding documents as responsive / non-responsive.  I know this is going to sound cheesy I guess, but – you have to put “discovery” back into Discovery. You need to be able to quickly find documents during a deposition when a deponent says something like “I never saw a document from Larry about our financial statements”, and not just search for “responsive: yes/no”, “privileged: yes/no.”

Inference was, and is, designed to be dynamic – providing suggestions to reviewers, opportunities to see relationships between documents and document sets not previously perceived, helping to guide attorneys – intuitively. Inference follows standard, accepted methodologies, including Boolean keyword search, field and parametric search, and incorporates all of the tools required for review – redaction, subjective coding, production, etc.

In addition to that overriding principle, we wanted the ability to get data in from anyone, anywhere and at any time. Regulators are requiring incredibly aggressive production timelines; serial litigants re-use the same data set over and over; CIOs are trying to get control over searching data more effectively, including video and audio. Inference is designed to take ownership of data once it leaves the corporation, whether it is structured, semi-structured or unstructured data.

Inside the firewall, the steps on the left side of the EDRM model are being combined.  Autonomy, EMC, Clearwell, StoredIQ — the crawling technologies – these companies are within inches of extracting metadata during the crawling process, and may be there already. This is where Inference comes in since we can ingest this data directly. I call it the disintermediation of processing because at that point there is no more additional costs for processing.

In the past someone would use EnCase for preservation, then Applied Discovery for processing (using date ranges and Boolean search terms), and at some cost per custodian, and per drive, you’d then pay for processing. It used to be over $2,500 per gig, now it’s more like $600 to $1500 per gig, depending on multi-language use and such.

But once corporations automate the process with crawling and indexing solutions, all of the information goes right into Inference without the intermediary steps, which puts intelligence back in the process. You can ask the system to guide you whenever there’s a particular case, or an issue. If I know the issue is a conversation between Jeff and Michele during a certain date range, I can prime the system with that information, start finding stuff, and start looking at settlement of the dispute. But without automation it can take months to do, at much higher costs.

Inference also offers quality control aspects not previously available: after, say, one month, you can use the software to check review quality, find rogue reviewers, and fix the process. You can also ultimately do auto-coding.

Bruce: I think this is a good opening for segue to the next question: how will analytics ultimately be used in eDiscovery?

Nick: The two most basic components of review are “responsive” and “privileged.” I learned from the public testimony of Verizon’s director of eDiscovery, Patrick Oot, some very strong statistics from a major action they were involved in. The first level document review expense was astounding even before the issues were identified. The total cost of responsive and privileged review was something like $13.6 million.

The truth is that companies only do so many things. Pharma companies aren’t generally talking about real-estate transactions or baseball contracts.

Which brings us to auto-coding… sometimes I try to avoid calling it “auto-coding” versus “computer aided” or “computer recommended” coding. When someone says “the computer did it” attorneys tend to shut down, but if someone says “the computer recommended it” then they pay attention.

Basically auto-coding is applying issue tags to the population based on a sampling of documents. The way we do it is very accurate because it is iterative. It uses statistically sound sampling, recurrent models. It uses the same technology as concept clustering, but you cluster a much smaller percentage.  Let’s say you create 10 clusters, tag those, then have the computer tag other documents consistently with the same concepts. Essentially, the computer makes recommendations which are then confirmed by attorney, and then repeated until the necessary accuracy level has been achieved. This enables you to only look at a small percentage of the total document population.

Bruce: I spoke with one of the statistical sampling gurus at Navigant Consulting last month, who suggests that software validated by statistical sampling can be more accurate than human reviewers, with fewer errors, for analyzing large quantities of documents.

Nick: It makes sense. Document review is very labor intensive and redundant. Think about the type of documents you’re tagging for issues – it doesn’t even need to be conscious: it is an extremely rote activity on many levels which just lends itself to human error.

Bruce: So let’s talk about what needs to happen before auto-coding becomes accepted, and becomes the rule rather than the exception. In your webinar presentation you danced around this a bit, saying, in effect, that we’re waiting for the right alignment of law firms, cases, and a judge’s decision. In my experience as an attorney, including some background in civil rights cases, the way to go about this is by deliberately seeking out best-case-scenario disputes that will become “test cases.” A party who has done its homework stands up and insists on using statistically validated auto-coding in an influential court, here we probably want the DC Circuit, the Second Circuit, or the Ninth Circuit, I suppose. When those disputes result in a ruling on statistical validity, the law will change and everything else will follow. Do you know of any companies in a position to do this, to set up test cases, and have you discussed it with anyone?

Nick: Test cases: who is going to commit to this — the general counsel? Who do they have to convince? Their outside counsel, who, ultimately, has to be comfortable with the potential outcome. But lawyers are trained to mitigate risk, and for now they see auto-coding or statistical sampling as a risk. I am working with a couple of counsel with scientific and/or mathematic backgrounds who “get” Bayesian methods- and the benefits of using them. Once the precedents are set, and determine the use of statistical analysis as reasonable, it will be a risk not to use these technologies. As with legal research, online research tools were initially considered a risk. Now, it can be considered malpractice not to use them.

It can be frustrating for technologists to wait, but that’s how it is. Sometimes when we are following up with new installs of Inference we find that 6 weeks later they’ve gone back to using simple search instead of the advanced analytics tools. But even for those people, after using the advanced features for a few months they finally discover they can no longer live without it.

Bruce: Would you care to offer a prediction as to when these precedents will be set?

Nick: I really believe it will happen, there’s no ambiguity.  I just don’t know if it’s 6 months or a year. But general counsel are taking a more active role, because of the cost of litigation, because of the economy, and looking at expenses more closely. At some point there will be that GC and an outside counsel combination that will make it happen.

Bruce: After hearing my statistician friend from Navigant deliver a presentation on statistical sampling at LegalTech last month I found myself wondering why parties requesting documents wouldn’t want to insist that statistically validated coding be used by parties producing documents for the simple reason that this improves accuracy. What do you think?

Nick: Requesting parties are never going to say “I trust you.”

Bruce: But like they do now, the parties will still have to be able to discuss and will be expected to reach agreement about the search methods being used, right?

Nick: You can agree to the rules, but the producing party can choose a strategy that will be used to manage their own workflow – for example today they can do it linearly, offshore, or using analytics. The requesting party will leave the burden on the producing party.

Bruce: If they are only concerned with jointly defining responsiveness, in order to get a better-culled set of documents — that helps both sides?

Nick: That would be down the road… at that point my vision gets very cloudy, maybe opposing counsel gets access to concept searches – and they can negotiate over the concepts to be produced.

Bruce: There are many approaches to eDiscovery analytics. Will there have to be separate precedents set for each mathematical method used by analytics vendors, or even for each vendor-provided analytical solution?

Nick: I’d love to have Inference be the first case. But I don’t know how important the specific algorithm or methodology is going to be – that is a judicial issue. Right now we’re waiting for the perfect judge and the perfect case – so I’ll hope it’s Inference, rather then “generic” as to which analytics are used. I hope there’s a vendor shakeout – for example, ontology based analytics systems demo nicely, but “raptor” renders “birds” which is non-responsive, while “Raptor” is a critical responsive term in the Enron case.

Bruce: Perhaps vendors and other major stakeholders in the use of analytics in eDiscovery, for example, the National Archives, should be tracking ongoing discovery disputes and be prepared to file amicus briefs when possible to help support the development of good precedents.

Nick: Perhaps they should.

PREVIOUS POST: The Evolution of eDiscovery Analytics Models, Part I: Trusting Analytics
NEXT POST:
Tape Indexing Breathes Life Into Tape Storage