Eleven weeks ago, we launched our predictive coding functionality. We know we’re not the first—or even the twentieth—vendor to offer this type of machine learning capability. Being first out of the gate isn’t generally our priority. Our focus has been on doing it better and on being responsive to user needs.
Since the release, we have been seeking feedback from our ediscovery users. Repeatedly, we have heard that the platform effectively finds relevant documents and sifts out those with less value. In the coming weeks, it will get even better at this, as we launch an algorithm improvement.
However, we have also heard that this feature could do better when it comes to workflow. Teams have very specific workflow needs: ensuring confidence in their results, preventing double work, efficiently completing review, and being able to explain how work was done. To address these needs, we will release several upgrades to our predictive coding in the coming weeks. Here’s what you’ll soon be able to do:
1) Mark Documents for Responsiveness, Privilege, or Anything Else
Previously, you coded documents as ‘Hot,’ ‘Warm,’ or ‘Cold,’ and predictive coding returned other documents likely to be ‘Hot.’ You will still be able to do that, of course. However, you’ll also be able to code documents on other scales and get predictive results on those scales. For example, you could set up the codes ‘privileged’ or ‘not privileged.’ After your team rates some documents on that scale, predictive coding will find other documents likely to be privileged. You could have many separate scales: one for responsiveness, one for privilege, and so on. This way, you not only code documents on multiple scales, as recommended by Mark Lyon, but you also get predictions to help find documents that belong in those categories. This ensures that the predictive power of this tool isn’t just applied for one task, but for many – saving time.
2) Exclude Documents from the Training Set
You can already ignore predictive coding results that aren’t meaningful to you. For example, your team can choose not to review documents predicted to be relevant, but which are from a certain e-mail domain or time period that you’ve otherwise deemed unimportant. Moving forward, you will also be able to exclude certain documents – whether by custodian, file type, or anything else you decide – from the predictive coding analysis entirely. You can set these documents aside for linear review and rest assured that they aren’t impacting the predictions made by predictive coding, saving time.
These are just two of the predictive coding workflow updates coming soon. But you can also help us prioritize future upgrades. How could we make the predictive coding workflow more useful to you? What is your team already doing that our platform does not yet support? Let us know in the comments!