The holy grail of records management is automated classification – taking people out of the record retention process and having computers automatically decide which electronic documents are which record types. Quite frankly, in the past, I’ve had my doubts on auto-classification of records, but maybe it’s time to take a new look.
There are problems with traditional record retention approaches for electronic information. One traditional strategy is to have all employees manually classify all electronic records, but this does not work very well. Electronic information is growing about 30 percent to 60 percent per year. Deluged with electronic documents, it takes a lot of time for employees to classify them and many employees tend to put it off. In assessments across a number of companies, we have found these employee-driven, manually-oriented retention programs experience fairly low program compliance, especially for electronic documents. This is a big problem as electronic records typically constitute 90 percent of an organization’s records.
The record management discipline can benefit from the inroads being created by predictive coding in e-discovery. Predictive coding technologies allow first pass review of documents, doing a pretty good job of determining which documents are relevant. Even with predictive coding, people today still need to re-review the computer’s work, but it does make the job easier. The same search and identification technologies currently being used in e-discovery are making their way into record retention products.
Questions arise whether auto-classification is compliant. There is very little case law validating predictive coding, and even less around the sufficiency of automatic classification for records management. Nevertheless, I think we are approaching the point where these auto-classification technologies do an imperfect, but better job of records classification than traditional methods. As record compliance is measured by how well you actually follow your policy, if auto-classification produces better end results than manual classification, one could argue that it is more compliant.