<img height="1" width="1" style="display:none;" alt="" src="https://dc.ads.linkedin.com/collect/?pid=513385&amp;fmt=gif">

By Katie Fitzgerald

Published on Fri, August 5, 2016

All posts by this person

In a recent webinar, Predictive Coding – Terminology Demystified, Meribeth Banaschik of Noerr and I asked the audience what they were still concerned about when it comes to using predictive coding. Overwhelmingly, 80% of our audience answered that they are most concerned about getting their team to adapt. The root of such concerns are often found in the fear of the unknown and being daunted by new terms, processes, and ways of working.

Here are some tips for how we can remove barriers to uptake, and reap the benefits of predictive coding.

1.   Education

Understanding the technology behind predictive coding is the first step to overcoming the fear. But, I would advise against a cursory Google search, which granted will pull up a wealth of material, but will be varied, complex, and tailored to specific technologies. It is very easy to disappear down the Google rabbit hole, and resurface none the wiser to what predictive coding is or how to use it. Instead:

  • Keep it simple - Reach out to your ediscovery provider and request their educational materials. Most should have starter packs and introductions to the technology, which is a good way to begin.
  • Keep it visual– Speaking from experience, the statistics are much easier to learn if there are visual representations. I particularly like Meribeth’s article where she uses a bull’s-eye to explain recall and precision.
  • Keep it practical – Don’t get lost in the theory. Work through case studies to understand how the technology works in real life.

To get you started, here are some introductory videos on predictive coding:

2.   Baby Steps

There’s no rule which states you have to dive into a fully blown predictive coding review and abandon everything you’ve known, worked with, and used in the past. Instead, as a first step into the foray of artificial intelligence, use predictive coding as well as your traditional review methods. The two are not mutually exclusive.

There are two particular use cases where this can be very successful:

  • Prioritisation – Instead of using predictive coding to define your dataset, use it to organise your dataset. This way predictive coding is not making any substantive decisions, it is simply an aide when structuring the order in which documents are reviewed. These are documents already selected as part of the data set, say by keywords for example. Using prioritisation allows the machine’s learning to push documents most likely to be relevant to the front of the review queue. This accelerates the review process and provides a strategic advantage by being able to know what you have quicker.
  • Quality Control – Utilise the teachings of your best reviewer to quality control the work of the rest of your review team. To do this, compare the predictions made by the machine to the work product of the review team. This is a more efficient and effective way to quality control.

Remember – you can get creative! Predictive coding doesn’t simply refer to relevance – you can use the tool on particular issues or privilege categories.

3. Show Results

The proof of the pudding is in the eating! The best way to convince the sceptics is by demonstrating tangible benefits.

  • Prioritisation – To illustrate the success of prioritisation, I’d suggest measuring the responsiveness rate. This is the percentage of relevant documents categorised a particular day over the total number of documents categorised that day. In a traditional linear review we would expect to see a continuously low responsiveness rate, as document distribution is random. In predictive coding, after training has occurred, we would expect to see a spike and a high responsiveness rate. Calculate the rate for your predictive coding review and compare this to the results of a analogous linear review.
  • Quality Control – This is easier. Simply calculate the number of non-relevant documents that were incorrectly categorised, but caught as relevant by the machine’s predictions.

New technology is always daunting – especially new technology that is full of jargon and statistics! However, I have had much success using the above steps to ease teams into using predictive coding. The teams have never looked back.

If you are interested in discussing more or would like a predictive coding presentation given to your team – please reach out to me at kfitzgerald@lhediscovery.com

About the Author
Katie Fitzgerald

Katie Fitzgerald has extensive industry experience. She has worked on litigation, arbitration, fraud, and bribery investigations; regulatory investigations; and Phase II mergers. She has experience working with many regulatory bodies, including the European Commission and the FCA. Katie holds a First Class Honors degree in law from the University of Edinburgh where she specialized in Technology, as well as achieved an English Law Degree.