Recommind and Predictive Coding
March 2, 2010
I received a flood of “news” from vendors chasing the legal market. Now law firms have fallen on hard times. One quip making the rounds in Kentucky is that a law degree is as valuable as a degree in Harry Potter studies from Frostburg State. I did not know one could get a degree in Harry Potter, so this may be some cheap jibe at the expense of attorneys.
The real action for legal licensing is in the enterprise. In the lousy financial climate, its seems that lawyering should be done in doors and back at the ranch. The demand for software and services that can chop discovery down to a management hunk of work have been selling. I prepared a legal market briefing for a couple of clients last year, and I was surprised at how much churn was underway in the segment. Even storage vendor Seagate poked its nose into the eDiscovery market.
I was delighted to receive a file from a reader that had the title “An Interview with Craig Carpenter of Recommind: A Discussion on Predictive Coding.” My recollection was that Recommind’s Mr. Carpenter, a polymath and attorney, was working on Recommind marketing. He is also a vice president of Recommind and teaches at the University of San Francisco. His focus in his class work is high technology marketing, content management, and digital rights management. Heady stuff.
You can get a copy of this document from JD Supra, whose tag line is “Give Content. Get Noticed.” I had not heard of this service previously.
Several points in the interview struck me as interesting. Let me highlight these and offer some of the ideas flapping around my goose brain.
First, Recommind won an award as the best product in the Knowledge Management Systems category. I think that is a good marketing angle, but I do not know what “knowledge management” means. Mr. Carpenter explained Recommind’s “knowledge management” product this way:
MindServer Search is our flagship enterprise search product. It provides highly accurate and relevant search results through a simple, intuitive interface. It uses proprietary, machine learning technology to automatically create concept models based on the information within the enterprise. That gives it the unique ability to accurately identify and rank relevant information for each user without the need for additional input from the user. For our legal customers, we also offer a popular Matters and Expertise module for MindServer Search, which enables them to find all relevant matter information and expertise within the firm. The module’s Expertise Location feature automatically updates areas of expertise based on work product, projects, clients, etc., which makes it simple to find attorneys with relevant experience on a particular topic, as well as the documents and matters associated with them.
Well and good, but I think this is search, retrieval, and social graph functions. That means that I understand Recommind’s definition of “knowledge management.”
Second, Recommind offers a description of its “knowledge management” system. The elements are CORE (context optimized relevance engine), which I believe is a probability based method somewhat akin to Autonomy’s approach. But the interesting statement, in my opinion, was:
There’s no doubt Predictive Coding is accurate enough – this has been proven in many cases. A number of AmLaw 30 firms have proven it by using Predictive Coding and comparing it to the results from contract attorney review (and partner review as well) on the same data. The results in every case were that they achieved better accuracy with Predictive Coding, and in the process saved 50-80% of what they would have spent on traditional review because contract attorneys were either not needed or were able to work far more efficiently (or both). This is what we mean when we talk about revolutionizing the economics of eDiscovery; no one else is doing this.
Third, this system’s automated methods can be used in legal matters. I found this statement interesting:
Judges care about getting to a just result as efficiently as possible; they care far less about the means used to get to that result – so long as the means do not undermine the pursuit of justice. So judges are not in the business of ?validating? any particular technology or process. That said, given the broken economics of today’s eDiscovery judges have definitely been expressing a fervent desire for a better approach to document review, and prominent judges like Judge Facciola, Grimm and Peck have indicated that technology can and should be brought to bear on the problem, because it can really help. It’s important to look at how the top litigation firms have responded now that they have a mandate from judges to change the economics of eDiscovery. And if you look at the top firms in the world — WilmerHale, Morgan Lewis and Fulbright & Jaworski, just to name three — they have made a commitment to Predictive Coding as the future. That’s a very, very strong endorsement.
Endorsements are good marketing, but in my limited experience with the legal system, what’s okay and what’s not okay can be variable.
Fourth, the role of humans remains important. I found this statement interesting:
There will always be a need for human review in eDiscovery. But bear in mind that the traditional eDiscovery process relies on an outdated, paper-based model that requires attorneys to sit in a room and review terabytes of ESI, one at a time. That’s a textbook example of work that should be assisted by intelligent automation. With the continuing rise of eDiscovery, there will always be plenty of work for attorneys. Some firms, and some clients, will always want to have an attorney’s eye on every document – which does not at all preclude the use of Predictive Coding. Even in such a case, they can perform that task much faster and more consistently using Predictive Coding.
Fifth, this comment about the cost of the system was instructive. This is the relevant passage:
We have certainly added more choices to our price list to accommodate the overwhelming demand we’ve seen, but if you are asking if we have had to lower our prices the answer is not at all. It’s definitely the case that much of the eDiscovery process, including culling, processing, hosting and forensic imaging, has been commoditized; older vendors trying to maintain market share and the rather simplistic appliance offerings and vendors have pushed this trend. But where we play and what our products are capable of doing for clients – Predictive Coding being perhaps the best example – is nowhere near becoming commoditized. The basic problem with eDiscovery is that it still uses the paper-based, linear review model, even though 99% of information these days is digital. Most EDD products try to alleviate the symptoms of that problem rather than address the problem itself, e.g. ?better linear review, a simple culling appliance, etc. Those technologies are commoditized now or will soon be. But we attack the fundamental problems of eDiscovery, the illness rather than its symptoms. Predictive Coding doesn’t just streamline document review for human reviewers-though it delivers that too-it actually automates the majority of the process using intelligent technology and defensible workflow. That’s something no other company or technology can deliver – period. And because it truly is game-changing technology, law firms and clients alike are more than willing to pay a premium. After all, it will save them a tremendous amount of time and money so the investment is easy to justify. Because this is so unique and such a difficult problem, in spite of a noisy market there’s no danger those capabilities will be commoditized any time soon.
If true, it suggests that the statistical methods used by other vendors such as Autonomy and Google, for example, should perform in a similar manner.
My view on this automation and prediction angle is that Recommind’s approach works well. If we accept that statement, what will happen if Autonomy or Google offers a lower-cost service. Might that shift some customers toward the lower-cost service. Numbers are numbers.
In a price war, Google—if it decides to push into the legal sector—might have an advantage over Autonomy with nearly $800 million in annual revenues. Recommind’s argument sets the stage for an interesting dynamic if larger firms go after this sector offering more value per dollar.
Excitement ahead in the fiercely contested and tumultuous legal market I “predict”.
Stephen E Arnold, February 27, 2010
No one paid me to write this article. Since I mentioned legal activity, I will report a no fee write up to the DOJ, an organization which cares about the law.
Comments
One Response to “Recommind and Predictive Coding”
-
- Subscribe to Beyond Search
Feature archive
News archive
-
[…] from the coding and auto-processes the remainder of the production. According to at least one source predictive coding may be as accurate, if not more accurate, than current doc […]