Re-Imagining the EDRM with AI

The EDRM has been a mainstay in e-Discovery for two decades now. The exceptionally insightful piece, created by George Socha and Tom Gelbmann, has provided the road map for practically every e-discovery litigation project since its inception. It helps provide a common language and placement for e-discovery practitioners, vendors, and professionals in describing what they are working on, where they are in the process and what needs to be done. Without it, there would be a level of anarchy within e-discovery, little in common between various providers, and very little conceptualization from a macro-level of the overall process involved. 

However, the model was created back in 2005, which in technological timetables is practically paleolithic. Think back to what the top technological trends were at the time: 

• Smartphones were still years away. The top devices were the Blackberry 8700 and Palm Treo 650. 
• YouTube was just founded in 2005, and MySpace was the biggest website. Facebook was still limited to college students at Harvard. 
• Videoconferencing/VOIP was still in its infancy. Skype (founded in 2003) and Vonage were just starting to gain traction. 
• The biggest internet provider at the time was still AOL, and broadband internet was just starting to make it into the mainstream. 
• The transition from DVD to Blu-Ray was just starting, meaning Blockbuster was still one of the nation’s biggest companies. 

When we look at what the e-Discovery landscape looked like, we’re talking about the rocky transition from banker’s boxes to the computer. This is like a walk down amnesia lane: 

• Microsoft Exchange and Outlook PST extraction was a central pain point. 
• Early versions of Law PreDiscovery and similar tools, mostly reconfigured archiving applications, were the original e-discovery platforms. 
• Web based review was still a new concept, with the most common tools being Concordance and Summation. 
• Native productions were uncommon, even largely non-existent. TIFF Images with load files were standard, with the Bates Stamping done during TIFF conversion. 
• Redactions were still primarily done through printed copy, with redaction tape. 

Which is all to say that the EDRM was designed at a time when technology is vastly outdated compared to today. Both the tech and the processes around e-discovery have evolved notably since then. With AI greatly changing how people conduct their jobs in this area, it’s perhaps a good opportunity to re-envision the EDRM incorporating more modern technology and workflows into the process. What would an AI based EDRM look like?

The model has been altered and added over the years, but the most recent version of the EDRM model came out in 2023. Here it is reproduced below: 

The key points to know about it are: 

• ESI moves through the process from left to right through the individual stages/columns of the model. 
• The “Information governance wheel” is a relatively newer addition, providing some context regarding data management within the particular organization. In eDiscovery parlance, this is often referred to as “behind the firewall.” How data is managed on the day-to-day basis within the organization has been a primary and increasing factor in how projects proceed. 
• Stacked boxes indicate processes that occur relatively simultaneously in the overall EDRM. For example, when documents in a preservation notice are identified, quite often they are preserved and collected at the same time. 
• When litigation or compliance is required, that kicks off documents through the EDRM, starting with “identification” of the documents and so forth. 
• You’ll notice the two triangles along the bottom, one labeled “volume” and the other labelled “relevance.” This signifies that from the left side of the EDRM, volume is at its highest, and over the course of the process continues to decline in size and scope. As the volume of ESI decreases, the percentage of relevant documents increases in quality and quantity. The two largely cross at the “production” marker, as that is near the point where the volume of documents you are working with equals the total relevant document set (privileged documents are withheld, which is why they don’t intersect exactly at the production marker). 

It’s a generally solid model describing the process, but is not without its criticisms: 

• It fails to put into account which tasks take up the most time and effort and which do not. For example, EDRM creator George Socha once depicted the time, effort, and cost disproportion this way: 

• Each stage is segmented often by various tasks handled separately by vendors or the organization itself. Over time, many of the vendors have since bundled these into various offerings, that don’t always conform to the model structure. Likewise, with technology, many of these tasks are now automated and with AI, require little to no human input at all. 
• Various validation steps at various points are not incorporated into the design of the EDRM and are usually based on validations within the various stages themselves. As more functions have become automated, the validations and reporting steps have become more important for defensibility. 
• Although each stage has its own share of complexities and details, of which you can dive into greater detail on the EDRM site (edrm.net), some stages are clearly more complex than others. Review, for instance, contains a very broad scope as to what is being reviewed, how, and for what. Various types of review require various tracks within just that stage that relate to all the surrounding stages in different ways. So as an example, the “Review” box representing an end-to-end AI review may resemble something like this: 

• Lastly, and perhaps most consequential with AI, behind the firewall solutions can start to handle identification, preservation, and collection of documentation as part of an overall information governance capacity. Such solutions are slowly becoming available and would be managed internally by corporate and client end users accordingly; their adoption and incorporation into corporate technological infrastructure is probably closer to reality than many think. 

When we look back at the current EDRM, and identify which stages AI is either currently providing solutions for or soon will be, we get the following result: 

In this view, the only stages untouched by AI are processing and production, processing being concerned with standardizing the various document formats as they come into a database for review, and production being the necessary bates stamping and standardization of documents going out of the review and produced for litigation and compliance. 

So, in updating and re-imaging the EDRM, those two points seem like important markers in delineating the processes going forward. If we simplify and merge overlapping stages, reconsider the various other stages using both Processing and Production as transitional stages for when documents come into review and when they are produced as discovery, we get something like this below: 

The purple stages represent areas where AI can streamline the process; because AI can handle several of these previously separate stages in a single process, those stages have now been merged into more consolidated stages. In addition, there is now an emphasis on validation and reporting requirements as well, with inputs throughout the process and the disclosures as part of the “production” stage. This helps describe the process, provides a narrative as to what was done, provides statistical results as to what eventually was produced and provides defensibility as to the overall process. We’ve also removed the background focus on decreasing volume and increasing relevance (which is intuitive) and redefined the background based on where the information resides; whether it is maintained by the client entity, in the process of an LSP, law firm or vendor in anticipation of production, or whether it has been already produced to the adverse party. 

But this only gets us to where we are today. Imagine what the EDRM will look like in the not-too-distant future. 

It really would be an end-to-end AI directed EDRM process that looks like this: 

And what a Brave New World that would be. That future is sooner than you think.