9 Reasons Why Lawyers Don’t Use Machine Learning on Review
Okay, so 9 isn’t a nice round number. I originally started with 5 which didn't fully convey the message and although I didn’t quite make it to 10, I hope on reading this blog some of your hang ups around predictive coding ( or machine learning/technology assisted review/artificial intelligence, whatever you want to call it) are allayed.
One thing I am sure of is that many lawyers are missing a trick by failing to make use of some of the advanced forms of legal technology available today. We’ve been harping on, as an industry, about this subject for years and I feel like we, as providers, need to take the majority of the blame for not making it easy for clients to appreciate the value of adopting a new approach to legal review.
After all, lawyers’ clients are becoming more demanding in relation to fees, and information governance doesn’t seem to be keeping pace with the volumes of data generated so unless you want to spend extortionate amounts trawling through irrelevant data you need, at some point, to seek a different way.
- My client has to pay extra for this service
It is true that sometimes you have to pay for efficiency. However, that doesn’t mean that those prices must break the bank. We, along with a handful of other providers, have developed our own software with our own machine learning capabilities and these are all available within the usual processing charges. This technology therefore doesn’t have to cost any extra but it does depend on the review platform selected.
- I’m expected to hold back on a full scale review while a reviewer codes a sample set
I suspect this notion came about through hesitancy on the part of providers to trust the technology themselves and this went on to pervade the industry. Our recent experience with machine learning reinforces the opinion that you don’t have to set up your review team in a certain way to realise the benefits from the technology. Although ideally we’d like to apply the machine learning to a specific individual, that doesn’t preclude other reviewers from categorising documents concurrently while the system finds its feet.
- I don’t feel confident defending this review strategy under scrutiny
Confidence is born out of experience and while I concede that it’s more difficult to become comfortable with ‘black box’ technology than keyword filtering, for example, that’s a journey we’re here to help with. I should also make it clear that machine learning can be used in conjunction with any other filters you apply. Most clients who are converts to machine learning began this journey by simply using the system to prioritise documents. As the system learns from the decisions being made it will award percentage relevance figures to each document and will push documents up through batching to ensure reviewers are seeing the most relevant documents sooner. On the vast majority of the projects we work on the statistics (we’d update you regularly during a review) support the fact that relevancy rates increase rapidly after the first couple of days of coding and fall away gradually a few days into the review and will usually look something like this…
- The computer intelligence may miss a hot document
I suppose this is an extension of the reservation above. Machine learning doesn’t need to be used to cull documents although this is often a consideration in the latter days of a review when you’re presented with review progress such as that above. There does not need to be any risk attached to using this approach. The more regularly you use prioritisation the more comfortable you’ll feel with the technology and it can then be trusted to run effective quality checks either of human reviewers or the remaining pool which you’re considering casting aside.
- I need to review all my documents in any case
But wouldn’t you prefer to strike upon those which are relevant earlier in the process?
- It’s too complex and I don’t trust the technology
It’s not necessary to understand the algorithms behind the technology, although we do have people who can go into this level of detail if needs be. From my perspective, I understand how the different machine learning offerings out there differ but I’d be more concerned about the results of the review as these should speak for themselves.
- I won’t be able to charge my client as many billable hours
Okay, I admit, a lawyer has never used this as an objection (at least without a wink of an eye and a wry smile) but there’s no harm in stirring a little controversy! Truthfully, in my experience, lawyers genuinely want to do what’s best for their client and it’s the lack of trust in the technology that prevents them from taking advantage of the cost savings which this technology can bring about.
- There’s no precedent for using the technology
First thing to say is that if you’re just using machine learning for prioritisation then a precedent isn’t needed. You’re still reviewing all the documents so you’re unimpeachable. The most recent decision relating to culling documents using the technology has taken place in the Irish Courts in the Irish Bank v Sean Quinn case. This use of predictive coding technology is a creeping inevitability and those early adopters will not only benefit from the experience but will also stand out through the lens of their client as being technology savvy and will attract more business as a result.
- I don’t have enough documents to justify its use
In line with the above reasoning, we routinely recommend and use machine learning to bring forward relevant documents on all projects above 10,000 documents. Although the power of the technology won’t be as acutely harnessed as it would be on a larger data set, you would still recognise benefits even with these relatively small document numbers.