AI in the law – The industrialisation of cognition

Print This Post

By Legal Futures

1 August 2016

By Richard Tromans, founder of TromansConsulting

Tromans: AI will change the shape of the law firm model

Tromans: AI will change the shape of the law firm model

I saw the future of the legal industry in a warehouse in Shoreditch. That perhaps sounds like an unusual thing to say about a $700bn global market, but after visiting a legal tech company recently in London’s most dynamic quarter, the true scale of what could happen to the legal sector was laid bare.

What I saw is not the end for all lawyers, but instead an artificial intelligence (AI) whirlwind hitting the current world of paralegals and junior associates, whose working lives may very well be about to turn upside down.

Such a change will, or at least could, shake the economic model of law firms that has existed for decades to its very foundations. And by that I mean the end of large-scale leverage, i.e. the practice of using several junior lawyers to every equity partner and through which some law firms have been able to become incredibly profitable.

We may even one day see a return to the former partner/junior model of 1:1 leverage, something most law firms have not experienced since before World War Two. In effect the great expansion of leverage, which grew and grew since the 1980s to the present, may now be at its apex. But, let’s come back to this later on.

‘Nevermore,’ said the RAVN

The tech company I visited was RAVN Systems, which works with Berwin Leighton Paisner and Linklaters, along with other law firms, as well as banks and corporates.

(I should add that there are several other tech companies operating in and around the same niche, such as Kira and Leverton. I’m mentioning RAVN as they very kindly invited me to see a demonstration of their products and I had a chance to actually see their software in action. They are not exclusively in the legal sector, as they also work in property and fin tech.)

Its most prominent offering is a Cognitive Engine, using what RAVN calls ‘Applied AI’, which we can perhaps call AAI. This can be differentiated from ‘Theoretical AI’, which is the stuff of much conjecture and fun, but nothing in the way of actual products (yet).

In short: AAI is the real face of new legal tech. It actually works and people are using it right now.

In a nutshell, what this software and its clever proprietorial algorithms do is make ‘useable meaning’ out of ‘unstructured data’ such as written text or tables, which have not been sorted or made interactive. E.g. a printed-out lease agreement, or a PDF contract document. In other words, data where a person would have to sit down, read and ruminate for a while to make any useful sense out of it.

For example, a major property company has given a law firm 1,000 lease agreements to examine. Most of these documents only exist today in printed form, others are formatted in an array of PDFs and old Word documents. Because of a change in accounting standards, the company wants to check whether certain properties leased under old agreements may now be a liability. All need to be read, very carefully. And quickly.

This is a lot more complex than searching for key words in a standard e-discovery exercise. The software has to look for several ways that key information might be written down. This is because many law firms over many years have written these documents for the client and often the phrases and even terminology used varies. The system first ‘learns’ what are the right phrases or terms, then begins to work through the unstructured data to produce results.

In this case, the client also needs all the other relevant information connected to each lease agreement that may be a liability to be pulled out in a way that is intelligible, i.e. not just a huge data dump, or a massive and seemingly endless Google search list.

To do this, the software can look for literally dozens of additional key points of interest (KPIs), e.g. dates and deadlines and financial transaction sums, and then arrange that data as the user wants it. But, let’s say this time the client just wants a dozen KPIs extracted and compiled intelligibly all at the same time from all of the 1,000 documents.

That is no easy task for a human, i.e. a human reader is not just looking out for the variably worded landmines in the hundreds of contracts, but also all the other information connected to each landmine. In other words, we’re looking at a paralegal with a rainbow collection of highlighter pens and a six-inch stack of Post-It tabs working under fairly continuous high stress to get it right.

The AAI on the other hand, once ready, just crunches away at the data in a massive parallel process that synthesises all the extracted data in one go. And it does this very quickly. In fact, it crushes the job in terms of time.

I witnessed in real time the RAVN AAI zapping through unstructured data at speed. In one brief example, 40 pages’ worth of lease agreements were filleted of their key meaning and data, which was then presented in a readable and intelligible dashboard of information, in less than two minutes.

When I mentioned to one of the RAVN team that filleting and making sense of so much data in two minutes was impressive, they added the system could operate far faster than that. And I’m sure it can. The test example I was shown was only operating off a laptop. One imagines that when operating off a top-spec, mega-MHz system in a law firm or bank, then processing speeds would be far higher. (I don’t have the exact speed data, but it’s clear no human could keep up with it.)

In comparison, let’s consider how a paralegal would do. It takes between 90 seconds to two minutes to read one page of a well-written novel if you are at the ‘medium to fast reader’ end of the spectrum. But this work is a dry legal agreement with many clauses and sub-clauses. It’s also vital no information is missed. So, perhaps let’s say three minutes per page to read, highlighting as well, which seems generous.

Naturally, a law firm would use a number of paralegals, maybe even a dozen or more on a large matter. But, the work rates would still be slow, taking several days to complete the job rather than hours when compared to an AAI. Plus, the more paralegals/associates on the job one has, the more they need to be managed, which eats up partner time. Not to mention the actual logistics of assembling the team, getting them in one place, which further slows the job down.

Also, different individual paralegals will approach the task in different ways and that creates variance between their performance, i.e. the more hands involved in making a product, the more chaos creeps into the system.

Clearly there is no competition in human vs AAI cognitive mechanics, at least not on this type of job. It’s like comparing a wheelbarrow (or group of wheelbarrows) to a jet-powered monster truck. The client gets so much more shifted, more quickly.

RAVN and its clients have also found that their AAI is more accurate than human readers. This is not surprising, given what monotony does to the brain. Boredom = mistakes. It’s a basic rule of human nature. Algorithms don’t get bored. AAIs don’t daydream, at least not yet.

So: quicker and more reliable. But, even better, new raw data can be poured into the system as and when needed. It is also very scaleable. Processing 1,000 documents is no different to 100,000. It takes a little more time, but that’s all. For a law firm, assembling and running a team to manually process 100,000 documents would be a significant task.

And, perhaps finally, once the AAI has learned the parameters and variations in a certain set of documents, it can be set to work on similar matters with most of its ‘learning’ already done – that ability is not lost, it’s ‘remembered’. The AAI’s capabilities therefore, at least in theory, can aggregate over time. Hence its productivity actually could increase the longer a law firm uses it.

What does this mean for the law firm model?

Now comes the strategic part of the story. How does this impact the business model of law firms? There’s a number of elements to consider, but here I’ll just give a ‘big picture’ view.

Much will depend upon how law firms respond to these changes. First, let’s look at a positive scenario.

The positive scenario

Law firm ‘A’ realises that in the long term, using large amounts of leverage is over. It invests heavily in AAI (and ‘smart contracts’), either making its own or working with legal tech companies to handle this need.

It informs the client base that fees for process work will be significantly lower than before (even compared to the process centre it built a few years ago in Belfast/Manchester/Glasgow). It may charge a fee per page processed, or agree a fixed fee that takes into account the added value of speed and accuracy. But it certainly can’t charge on a time basis, or a manpower basis. Some very radical firms may even offer a freemium model and only charge for senior lawyer input, though I would imagine the ‘value added’ fixed-fee approach may be the favoured option.

This produces an ‘AI halo’ around the law firm. Clients are happy because they get a better service, one that’s faster and more accurate, plus cheaper (at least for the process work). But the firm can also now take on more clients at the same time because so much of the process work is done by the AAI’s far greater capability.

Although there is a cost to the law firm in developing or buying in the AAI, its staff costs are removed from the equation. Margins may not have to be eroded, if managed – and sold – well.

Meanwhile, workflow actually increases at the firm even though leverage has gone down – we have the paradox of less total staff, but more work. This is simply down to far higher productivity at the process ‘bottlenecks’ inside the law firm.

In fact, as workflow rises and the demand for more partner-level experience rises to handle all the extra work, the firm starts growing at the middle and top of its career pyramid. In effect, the law firm model pyramid inverts, or perhaps morphs into a mushroom shape – wide at the top, with a little stem of junior staff below.

Those law firms that move fast and embrace AAI get ahead of the curve, take market share and are ‘future proof’. Also, because they have built what clients need, the client base feels well served and adapts to the new offering with relish. AAI becomes a true competitive advantage and slower firms get left behind.

The nightmare scenario

Then there is the other way. In this version of the future, law firms turn their back on AAI, seeing it as a threat to their margins and structure. They do a little bit of AAI work to appear to be keeping pace, but in reality they are rather like oil companies talking about how much they love wind and solar power… i.e. it ain’t gonna happen.

But tech companies still exist. Clients with large procurement teams and smart general counsel still exist. Like water finding its own level, eventually AAI finds its place in the hands of in-house legal teams, which find they really don’t need external law firms as much as before as they build their own AAI legal capability and slash their costs massively in the process.

How much a loss of income to the commercial legal market this pathway into the future would be is unknown. But one could estimate that large commercial firms could see between a quarter to a third of their revenue taken in-house by AAI. Or perhaps another legal provider, possibly even one of the Big Four accountants, comes in and works with the client base to give them what the law firms won’t.

In this scenario, the recalcitrant law firms resist change and see a steady fossilisation of their business model. They survive by making access to the equity pool harder and harder to prop up profits per equity partner, while cutting process costs using cheaper and cheaper human labour, which is surely the most basic approach possible to solving the tech challenge.

That is to say, using cheaper and cheaper legal labour/business support staff to compete with AAI is like a farmer feeding his plough horse fewer and fewer oats to reduce costs so he can compete with the farmers who have embraced tractors powered by the internal combustion engine.

Eventually the old model becomes unsustainable (or to complete the metaphor: the horse dies). In the case of the $700bn global legal market, it would be forced into a systemic and messy meltdown, which could easily be avoided.

The future

What AAI and other legal tech developments are taking over is ‘cognitive’ work, or ‘reading stuff’ to you and me. It may sound a simple thing, but lawyers, especially junior ones, spend a lot of time reading. In effect, this is the industrialisation of cognition. This process will start at the ground level with paralegals and work its way upward.

And, it has already started.

To conclude: this is just a small snapshot of what is happening with AAI and what it means for the legal market. While it can sometimes be hard to navigate between the ‘end of all lawyers’ at one extreme and the ‘nothing will ever change’ hold-outs at the other, we have no choice but to pick a way through to the legal tech future we want to build.

Let’s just hope that law firms opt for the first, more positive scenario. But, sadly, as to the paralegals, to misquote Douglas Adams: ‘So long, and thanks for all the reading.’

Richard also blogs at Artificial Lawyer

2 Responses to “AI in the law – The industrialisation of cognition”

  1. Sir,

    I would appreciate your calculations or citation on how you came to the $700 billion dollar figure for the global legal market. Thank you.

  2. D Kent on August 1st, 2016 at 5:48 pm
  3. Well worth the read, thank you for sharing.

    It will inevitably be forced by clients demanding more for less, once again those with the vision will see the huge opportunity this offers.

  4. Jonathan Maskew on August 2nd, 2016 at 11:21 am

Leave a comment

* Denotes required field

All comments will be moderated before posting. Please see our Terms and Conditions

Legal Futures Blog

The importance of being expert

Steve Rowley 3

I recently sat on a panel debate in Manchester, with the debate entitled – ATE insurers and sub-£250k claims. Whilst the title of the debate was probably written ahead of the government’s consultation paper to introducing fixed recoverable costs in lower-value clinical negligence claims, where £25,000 rather than £250,000 is being recommended, it nevertheless raised an interesting point on how after-the-event insurers can make premiums proportionate to damages, especially for cases worth less than £25,000.

April 26th, 2017