How do sellers of legal services differentiate themselves? How do buyers of legal services choose between them?
Much has been written in the past about these paired questions, and traditionally, the answers can be distilled into a blend of words like expertise, experience, creativity, trust, prestige, recognition, size, or presence. Yet regardless of how specifically firms have positioned themselves, the presence of some competitive advantage has clearly been critical for the longitudinal financial success of the world’s largest firms.
These competitive advantages were typically built over decades, with firms becoming known for developing unique deal structures, representing clients in bet-the-company cases, scaling to global mega-mergers, or navigating the most complex regulatory environments. While recognition became a proxy for success, the success itself was typically based on the durable assets that firms held, like knowledge, expertise, and precedent - all forms of information - and their ability to scale that information through timekeeper effort-hours.
What, then, happens when the ability to access information becomes more accessible? What happens when you don’t need scale to scale?
LLMs Change the Game
While ChatGPT has captured the public’s interest, API access to large language models and viable open source foundation models are already changing the game inside of many organizations.
For low-level tasks, such as those that occur in diligence and discovery, there are clear economic benefits to using LLMs to automate these tasks. While prior generations of natural language processing and machine learning had some uses, the most recent generation of large language models have shown a significant reduction in human effort hours required to meet the same quality work product. Even small firms are able to recognize this benefit, as it allows them to manage tight timelines or large projects without needing to hire for their highwater mark.
But more than just low-level tasks are on the table now. As we have demonstrated with our research on the Bar exam and CPA Exam, these models are becoming increasingly capable. Even in the hardest assessments, like zero-shot prompting on exams that many humans fail, models like GPT-4 have demonstrated broad knowledge and remarkable progress on reasoning capabilities.
When models like GPT-4 are combined with knowledge graphs, the LLMs are capable of performing even more complex annotations and reasoning, as often occurs in contract drafting, negotiation, or interpretation and strategic planning. For example, tools like our open-sourced SALI Search system allows users to easily search or tag with SALI’s Legal Matter Standard Specification (LMSS), reducing the risk of hallucination and increasing domain-specific performance past zero-shot GPT.
With the rise of these technologies, it’s become increasingly possible for smaller organizations to perform work that was previously impossible for them. And as these trends continue, many of the historical strategies for competitive advantage will be rendered obsolete.
While we’ve previously discussed the importance of future-proofing your AI strategy per se, it’s clear that firms need to align this planning with their existential competitive strategy.
When everyone has access to the same models and data, what will competitive advantage look like?
Bespoke Models are the New Competitive Advantage
The simplest response to this strategic hypothetical is to reject the premise. Don’t use the same models and data as everyone else. But how practical is it to build or maintain your own bespoke model?
While we discuss this topic in more detail in other posts, the answer is that it is more practical by the day. Competition between commercial vendors, advances in academic research, and collaboration in open source projects has resulted in an exponential decrease in cost over just the last six months. As further investment and interest bears fruit, it is likely that training and maintaining organization-specific models will be common in the near future.
The next question, of course, is what firms will put into these organization-specific models or how they will differently optimize them. Will some firms specialize in models related to litigation or regulatory lobbying? Will others prioritize training on internal precedent and deal docs from their own DMS? How will these choices interact with a firm’s broader AI risk management process?
Productizing Your Expertise
While so-called productization of legal services has existed for consumer-focused legal needs, productization is still relatively rare for larger and more complex legal work. Much of firm strategy and structure today is still based on the idea that internal knowledge and human capital can be billed under cost-plus models to willing corporate payors. And while rumblings of the death of the billable hour and rise of alternative fee arrangements are not new, it’s possible that change is truly coming now.
On the flip side, law firm profitability has historically been limited by the cost-plus model. No one could escape the gravity of timekeeper rates, headcounts, and utilization. If firms really could scale their checklists, workflows, processes, templates, or advice and counsel through productization or technology-enablement, could they be even more financially successful?
Going forward, the most difficult part of productizing will likely be translating the internal knowledge into a product or technology-enabled service that is safe and reliable. Luckily, offerings like our Kelvin Legal Data OS and related services are lowering the cost for firms to assemble their own solutions.
Moving Forward, Safely, Quickly
Developing a competitive strategy in such a fast-moving environment is challenging, but it’s clear that these changes will provide opportunities for firms of all sizes to capitalize - if they so choose to do so. Making decisions and making progress, safely and on competitive timescales, can seem daunting, but with the right help, there’s never been a better time to build a better firm.