Back to the future: The science and fiction of legal tech
Our success at Konexo in optimizing legal operational processes has been regarded by some as close to science fiction. This is our mission though; we want our clients to enjoy increased productivity at reduced costs and lower levels of operational risk. A critical component to us delivering on this mission is unlocking lawyers from administrative and non- legal work and enabling them to focus on clever, high value stuff. We achieve this ‘unlocking’ by doing all we can to extend the effective scope of technology deep into traditional legal operational territory.
Now, there is already powerful Legaltech on the market, having a profound impact across legal operating models. Nowhere though have these ‘contract’ vs ‘technology’ engagements been more extensively played out than in the data extraction context. We are barely at the end of the beginning of this journey and we already know there are regions of most major contract estates that will always require lawyer time to unpick. Our observation though is that these “regions” tend to be smaller than first thought, creating great opportunity for the ‘unlocking’ we referred to earlier. Any assumed limits about the scope for technology are not Machiavellian – it’s simply that technology driven change is unsettling and is not something that can be done to clients or on clients; its got to be done in partnership with clients.
To restate our ‘mission’ slightly then with this in mind, we at Konexo want to ‘unlock’ lawyers from administrative and non-legal work and in doing so, by necessity, push the boundaries of where effective technology begins and ends. In the early days this means working collaboratively with our clients, making sure we bring nuanced insights to the table, often quietly starting important digital transformation journeys with small steps, relentlessly striving to generate trust and buy-in. Let’s step through a few fundamental considerations in a typical data extraction project, which, if engaged with carefully and empathetically, develop trust and earn us the right to partner with clients.
Understand that the technology zeitgeist is always evolving. There is of course the ubiquitous “Artificial Intelligence” (AI) designator. Seldom has a term been so over used or misapplied. Once you peer through the murk though, there are a small number of players with undoubtedly powerful solutions. The ongoing issue with AI though, no matter how powerful, is that, in the final analysis, it is making extraction (or any other legal operational activity) decisions based on a probabilistic methodology. This does tend to trigger risk controllers into demanding significant quality assurance, which drives up effort and cost. Your other options are “Machine Learning” (ML) and “Matching Engines” (ME). What both of these have in common is an inevitable and significant “training” requirement in their maturity cycle before they are effective. Simply put, ML transmutes its training period into a complex set of rules that govern future extraction behavior. ME use training to progressively populate a clause library, against which, new incoming documents and language are matched.
Genuine AI can be extremely powerful, almost magic, but is always expensive. In regards to ML, we often see that, over time, the bank of ML rules becomes extremely complex and starts generating significant performance and data quality concerns. Therefore, our preference is usually to deploy mature ME technology as it is effectively ‘plug and play’ out the box and because in ‘matching’ extraction candidates versus an existing clause library, not using a probabilistic ‘guess’, you no longer have the quality assurance requirements. Human intervention on ME extraction projects is only required where the technology encounters language it does not have in its clause library, which needs to be set up for the first time. Thereafter the technology is smarter and won’t ‘fall over’ again.
These are simple considerations when you understand the vocabulary and landscape. The history of data extraction projects though is littered with poor zeitgeist recognition and the consequent fall-out in trust.
Box clever in the early days. Another fraught moment in data extraction project delivery is the very act of starting. Often, if there is no established culture of digital transformation, which is common in legal areas, the first steps can become over imbued with significance, potentially at the expense of delivery sensibilities. As an example, for data extraction projects, you need to know enough about your contract estate to set up the technology and set it off running but not a whole lot more. The fastest way to do this is to explore permutation not volume analysis of the document estate.
This nuance is worth dwelling on for a bit; this is about looking at contractual behavior patterns (permutations) operating across the contract estate and using this knowledge to calibrate the technology. We are not talking about looking at multiple examples (volume) of the same behavior patterns and driving an atomic understanding at the contract level before even allowing technology onto the pitch. It is natural to find comfort in, and reach for, volume analysis but still a mis-step and generates an additional quantum of effort and time, the cost of which often kills proof of concept work before it gets started. Existing ME technology can support permutation analysis by quickly ingesting the contract estate and then immediately bucketing together contracts with similar features (permutations). It is really that quick and easy. No more than one or two contracts per permutation need to be reviewed and this typically takes no more than a few days.
Be prepared to contemplate the unknown. Infamously, a powerful AI tool doing a run of the mill data extraction on retail customer contracts picked up a recurring pattern, that, on investigation, turned out to be indicators of a nascent mis-selling scandal. As this was unexpected, legal privilege and information flows had not been considered. It was tricky for all concerned, even though, overall, the detection of this behavior was a positive outcome. This is obviously an extreme example to make a point, however, many data extraction exercises do in-fact unearth uncomfortable things like: contracts that are not properly executed; clauses that are outside of policy; incorrect deal data; and yes, positions where conduct risk is at the limit of risk appetite.
It is important to accept that these discoveries are inevitable, especially on large historical contract estates. Simply set- up, pragmatic governance will avoid anxiety on low/intermediate risk findings and map out a defined escalation path for high risk items. With all eventualities covered, there is no need to delay or shut-down a project, no matter what emerges.
This is hopefully a flavor of the early project considerations that we try to knock out the park, always hopeful that these experiences with us drive a different technology transformation experience for our clients. For if we can consistently convince with respect to these concerns, we can unleash the power of the technology and unlock lawyers to focus on tasks, which genuinely require their specialist expertize.