A clouded view of Records and Auto-Classification
This morning, I was reading Lawrence’s blog titled “Does Records Management Give Content Management a Bad Name?”, which picks on one of the points in Cheryl’s article “It’s a Digital-First World: Five Trends Reshaping Records Management As You Know It”, with some very insightful comments added by Christian. I started leaving a comment under Lawrence’s blog (which I will still do, pointing back to this) but there are too many points I wanted to add to the debate and it was becoming too long…
So, here is my take:
First of all, I want to move away from the myth that RM is a single requirement. Organisations look to RM tools as the digital equivalent to a Swiss Army Knife, to address multiple requirements:
- Classification – Often, the RM repository is the only definitive Information Management taxonomy managed by the organisation. Ironically, it mostly reflects the taxonomy needed by retention management, not by the operational side of the business. Trying to design a taxonomy that serves both masters, leads to the huge granularity issues that Lawrence refers to.
- Declaration – A conscious decision to determine what is a business record and what is not. This is where both the workflow integration and the auto-classification have a role to play, and where in an ideal world we should try to remove the onus of that decision from the hands of the end-user. More on that point later…
- Retention management – This is the information governance side of the house. The need to preserve the records for the duration that they must legally be retained, move them to the most cost-effective storage medium based on their business value, and actively dispose of them when there is no regulatory or legal reason to retain them any longer.
- Security & auditability – RM systems are expected to be a “safe pair of hands”. In the old world of paper records management, once you entrusted your important and valuable documents to the records department, you knew that they were safe. They would be preserved and looked after until you ask for them. Digital RM is no different: It needs to provide a safe-haven for important information, guaranteeing its integrity, security, authenticity and availability. Supported by a full audit trail that can withstand legal scrutiny.
Auto-categorisation or auto-classification, relates to both the first and the second of these requirements: Classification (using linguistic, lexical and semantical analysis to identify what type of document it is, and where it should fit into the taxonomy) and Declaration (deciding if this is a business document worthy of declaration as a record). Auto-classification is not new, it’s been available both as a standalone product and integrated within email and records capture systems for several years. But its adoption has been slow, not for technological reasons, but because culturally both compliance and legal departments are reluctant to accept that a machine can be good enough to be allowed to make this type of decisions. And even thought numerous studies have proven that machine-based classification can be far more accurate and consistent than a room full of paralegals reading each document, it will take a while before the cultural barriers are lifted. Ironically, much of the recent resurgence and acceptance of auto-classification is coming from the legal field itself, where the “assisted review” or “predictive coding” (just a form of auto-classification to you and me) wars between eDiscovery vendors, have brought the technology to the fore, with judges finally endorsing its credibility [Magistrate Judge Peck in Moore v. Publicis Groupe & MSL Group, 287 F.R.D. 182 (S.D.N.Y.2012), approving use of predictive coding in a case involving over 3 million e-mails.].
The point that Christian Walker is making in his comments however is very important: Auto-classification can help but it is not the only, or even the primary, mechanism available for Auto-Declaration. They are not the same thing. To take the records declaration process away from the end-user, requires more than understanding the type of document and its place in a hierarchical taxonomy. It needs the business context around the document, and that comes from the process. A simple example to illustrate this would be a document with a pricing quotation. Auto-classification can identify what it is, but not if it has been sent to a client or formed part of a contract negotiation. It’s that latter contextual fact that makes it a business record. Auto-Declaration from within a line-of-business application, or a process management system is easy: You already know what the document is (whether it has been received externally, or created as part of the process), you know who it relates to (client id, case, process) and you know what stage in its lifecycle it is at (draft, approved, negotiated, signed, etc.). These give enough definitive context to be able to accurately identify and declare a record, without the need to involve the users or resort to auto-classification or any other heuristic decision. That’s assuming, of course, that there is an integration between the LoB/process and the RM system, to allow that declaration to take place automatically.
The next point I want to pick up is the issue of Cloud. I think cloud is a red herring to this conversation. Cloud should be an architecture/infrastructure and procurement/licensing decision, not a functional one. Most large ECM/RM vendors can offer similar functionality hosted on- and off-premises, and offer SaaS payment terms rather than perpetual licensing. The cloud conversation around RM however, comes to its own sticky mess where you start looking at guaranteeing location-specific storage (critical issue for a lot of European data protection and privacy regulation) and when you start looking at the integration between on-premise and off-premise systems (as in the examples of auto-declaration above). I don’t believe that auto-classification is a significant factor in the cloud decision making process.
Finally, I wanted to bring another element to this discussion. There is another RM disruptive trend that is not explicit in Cheryl’s article (but it fits under point #1) and it addresses the third RM requirement above: “In-place” Retention Management. If you extract the retention schedule management from the RM tool and architect it at a higher logical level, then retention and disposition can be orchestrated across multiple RM repositories, applications, collaboration environments and even file systems, without the need to relocate the content into a dedicated traditional RM environment. It’s early days (and probably a step too far, culturally, for most RM practitioners) but the huge volumes of currently unmanaged information are becoming a key driver for this approach. We had some interesting discussions at the IRMS conference this year (triggered partly because of IBM’s recent acquisition of StoredIQ, into their Information Lifecycle Governance portfolio) and James Lappin (@JamesLappin) covered the concept in his recent blog here: The Mechanics on Manage-In-Place Records Management Tools. Well worth a read…
So to summarise my points: RM is a composite requirement; Auto-Categorisation is useful and is starting to become legitimate. But even though it can participate, it should not be confused with Auto-Declaration of records; “Cloud” is not a functional decision, it’s an architectural and commercial one.