Machine intelligence and legal contracts: why now?

Published on
October 27, 2020
Chandar Lal
Machine intelligence and legal contracts: why now?
“It is now necessary to attempt to develop computer systems in law that can be said to embody knowledge, and even exhibit intelligence. Achievements over the last twenty years, in the branch of computer science referred to as Artificial Intelligence (A.I.), have perhaps now provided the appropriate technological framework within which the construction of such knowledge-based systems in law might be undertaken.” – Richard Susskind, 1986

Machine intelligence applied to the law is not a new concept – yet its gestation has been slow, and it has not inflected into widespread adoption. Despite the vast size of the legal services industry, only a handful of venture-scale software businesses have arisen in the last decade. Legal contracts are the atomic unit of any business interaction, yet few enterprises and law firms apply machine intelligence to them systematically today.

Perhaps this is little surprise when we consider: (a) the conservatism of end users; (b) the traditionally high cost and complexity of building sophisticated NLP products; and (c) the tension between law firms’ hourly billing structures and the incentive for efficiency.

Might things be about to change? We’re currently seeing a wave of NLP-driven legaltech startups that are translating contracts into structured data, experimenting with new use cases, and finding early product-market fit. Those who craft the right distribution strategy today will emerge as the next decade’s winners in legaltech.

Thanks to those who’ve helped us clarify our thinking – including the founders who joined our recent roundtable, as well as Mike Callahan and Joshua Walker at Stanford, and the GCs we’ve interviewed in recent weeks.

Investing in contract analysis: why now?

Legaltech has grown steadily in recent years, and certain pain points are clear. We know that natural language is difficult to parse at scale. MS Word is an ineffective way to collaborate on contracts. Monolithic file structures and email transmission create major issues with document control. But industry inertia has been strong. Now, there are new catalysts on the supply side and the demand side. On the former, advances in NLP are enabling cheaper, better products to be built. On the latter, COVID has accelerated the urge for efficiency among resource-constrained internal counsels.

The overall bull case is supported by:

  • Market size. The Western European legal services market is worth ~$150B (Thomson Reuters) – a conservative estimate that does not account for in-house counsels. Spend on legal software is small in relative terms today, but there is ample headroom for growth. Capital continues to flow in: 2019 saw a YoY doubling of global legaltech investment. In the US, the ABA’s restrictions on non-lawyers’ provision of legal services are being relaxed on a per-state basis – which may accelerate disruption of the US legal services market.
  • NLP advances as a supply-side catalyst. Pre-trained language models (see BERT, GPT-3, HuggingFace) reduce the cost of developing and deploying NLP-based products, and are lowering startups’ barriers to entry. This might create a risk of commoditisation: the winning products will build moats through the proprietary legal training data they can get from their customers.
  • COVID as a demand-side catalyst. Enterprises are scrutinising operational costs more closely, seeking cost efficiency from their in-house counsels. For many of the GCs we’ve met recently, this has created a ‘burning platform’ to drive experimentation with contract analysis tools.
  • Evolving fee models. The shift from billable hours to fixed-fee pricing, while gradual, is incentivising law firms to pursue operational efficiency in contract review.
  • Weak incumbent offerings. The first generation of contract analysis products are broadly seen as helpful for extraction, but weak at providing meaningful insight to support drafting and negotiation (e.g. market analysis of terms, comparisons with internal contract libraries). Most incumbent products tend to exhibit weak network effects, and there is potential for them to be displaced by bottom-up adoption of next-gen tools.
  • Underserved mid-market. Most existing contract analysis products are generally priced and positioned for large enterprise customers. This leaves a large mid-market customer base relatively uncontended, with resource-constrained internal functions in search of efficiency.

Where to play: customers and use cases

In B2B legaltech, there are three generic customer sets: law firms, in-house counsels, or other enterprise functions? Many startups experiment with a combination as they pursue product-market fit. What is the ideal go-to-market strategy?

Selling to law firms especially smaller, specialist firms – is a viable early strategy given the short sales cycles and potentially easy access to a library of training data. These firms can be effective partners to iterate on use cases, test product-market fit, and gain a valuable source of model training data. The key use cases tend to be big-ticket items where assisted review can have the greatest financial impact: we’ve repeatedly heard senior partners calling for assisted reviews in M&A due diligence, litigation prediction, and force majeure claims.

The key challenge is alignment with practitioners’ incentives: the goal is to build use cases for automation that improve client outcomes, reduce write-offs, or improve throughput – and not compromise the number of hours they can bill.

In the long run, we expect successful players to move in-house, upmarket, and cross-jurisdiction – each calling for careful product and go-to-market manoeuvres. Enterprise users will generally be the more attractive customer set, given the larger market size and internal incentives for operational efficiency mandated by the C-suite.

Selling to in-house counsels is rarely fun or fruitful. GCs are time-poor, often tend towards conservatism, and tend not to come with a direct business case for contract analysis solutions. Indirect, partner-led channels may be a viable route to acquiring customers in the short term, but the incentives need to be aligned carefully. Direct relationships with the customer are needed, e.g. to iterate on product, and maximise access to training data

Selling to enterprise functions is where we expect the most long-term value to be created. While in-house legal teams are often the primary buyer / sponsor of a software product, the end users are often non-lawyers in other enterprise functions. Hence, a productive go-to-market motion is likely bottom-up, via non-legal teams who engage with large volumes of contracts, and are seeking process efficiency.

Sales, Procurement, HR, Compliance, and Finance teams are viable entry points. Sales teams, as revenue centres, tend to buy faster than other departments, so are a particularly attractive target. Today, all rely heavily on in-house counsels as internal service providers, and process bottlenecks are rife. We’re excited to see simple, self-service tools that empower these teams to review and generate contracts more smartly than ever.

Evaluating early-stage startups

Here are some of the heuristics with which we’re evaluating new entrants:

  • How much will a customer spend? What is the economic incentive / ‘burning platform’ for the key business sponsor? How will RoI be measured, and how might the product up-/cross-sell in the medium term?
  • Specialist or generalist? If the former, how big is the niche and what penetration is viable? If the latter, what is the insertion point and the competitive advantage relative to incumbents?
  • How deep is the data moat? Is the product defended by access to proprietary contract libraries / datasets, a network effect between users, or simply a class-leading user experience? Do these moats self-reinforce as the user base grows?
  • Can the product expand virally between counterparties? Given the two-sided nature of contracts, there’s an opportunity to build network effects over time. DocuSign is a great case study here: while early growth was driven by partner-generated leads and internal use cases, the true flywheel was a network effect between counterparties.
  • Standalone product, or a feature of legal suites in the long run? A headline trend in recent years has been incumbents’ lateral expansion, mostly via M&A (e.g. Thomson Reuters, RELX, Big 4, DocuSign). Can we create a standalone product with a distinct enough value proposition, business case, and user group to achieve a large outcome?

These questions might help us separate good companies from great ones today. Taking a longer term view, there are plenty of questions that remain unresolved. Will users’ dependence on MS Word diminish? Will new products remain defensible amid the rise of pre-trained, cheaply available language models? Might there even be new ways to encode legal knowledge outside of natural language and static documents?

If you’re building a startup in this space, we’d love to trade thoughts on these questions – please get in touch.