From: redpointai
The legal industry is undergoing a significant transformation due to the integration of Artificial Intelligence (AI). Lorra, led by CEO and co-founder Max Unistron, is a company at the forefront of applying AI to the law industry, working with many top law firms globally and having raised over $100 million [00:00:25].
The Evolution of AI in Law
In 2020, when Lorra began, AI models like early BERT from Google were not performing well, especially in non-English languages like Swedish [00:01:40]. The arrival of GPT-3.5 marked a “paradigm shift” that enabled real progress [00:01:52]. The industry has moved from pure experimentation to actually implementing AI solutions that handle end-to-end work deliverables [00:01:59].
For instance, due diligence, which previously involved physical data room visits or complex manual processes, can now be handled by placing documents into Lorra, specifying what needs to be found, and then generating a report based on the findings [00:02:08]. This signifies a shift from simple queries against data sets to having Large Language Models (LLMs) follow a defined process using accessible tools to produce usable work deliverables [00:02:29].
The Future Landscape
While AI models continuously improve, the most significant leverage comes from surrounding frameworks like function calling, tool calling, and MCP [00:02:53]. The legal software space, historically fragmented with separate tools for translations, document comparisons, searching, and reviewing, is now seeing these functionalities integrated through AI [00:03:07].
Legal work can be viewed on a spectrum of complexity:
- Bottom Quartile: Simple tasks like data extraction [00:03:26]. Lorra has already fully automated much of this segment [00:03:39].
- Top Quartile: Complex tasks like drafting a share purchase agreement [00:03:33]. The goal is to gradually move up this spectrum [00:03:45].
Law firms and legal professionals will need to identify where their specific expertise, context, and instruction for models add the most value, and where off-the-shelf LLMs are sufficient [00:03:50].
Why Law is Uniquely Suited for AI
The legal industry has historically seen limited software development compared to other fields [00:04:32]. Traditional industry incentives sometimes do not align with efficiency or software adoption [00:04:42].
Key characteristics making law suitable for AI:
- Repetitive Tasks: In-house counsels frequently handle the same types of documents, such as NDA reviews and MSAs, focusing on controlling business risk [00:05:00].
- Categorizable Work: Legal work broadly falls into reviewing, reading, drafting, writing, or researching [00:05:20].
- Cross-functional Capability: Unlike older software focused on niche problems (e.g., contract drafting or review), AI is capable of performing tasks across this entire stack, enabling platforms like Lorra to serve a “wall-to-wall” range of needs [00:05:36].
The Hourly Billing Dilemma
The traditional hourly billing model has been seen as a barrier to software adoption, as increased efficiency would reduce billable hours [00:07:03]. However, client pressure is changing this dynamic [00:06:09]. Clients, especially large private equity firms, are increasingly using AI tools internally and are reluctant to pay for previously expensive, time-consuming tasks like due diligence or contract review [00:07:31]. This puts pressure on law firms, as competitors adopting AI force others to adjust to remain competitive and avoid being seen as inefficient [00:07:59].
Lorra’s Strategic Advantages and Product Philosophy
Lorra’s journey, starting in the Nordics, allowed it to grow from a “small fish” to a dominant player in a smaller market before expanding globally [00:08:58]. Europe’s fragmented market forced Lorra to develop a multi-market, enterprise-ready product from the outset [00:09:18].
Key strategic choices:
- Fast Second Mover: Instead of focusing on training their own LLMs like early players, Lorra, with less initial funding, chose to focus on building an application layer that people would be excited to use [00:09:43].
- Application-Centric: The philosophy is to build an application and leverage existing AI labs as platforms, avoiding building functionalities that these labs will eventually offer for free via APIs [00:14:24].
- Example: Lorra initially built its own citation feature, but if LLM providers offer it for free, their code will be deprecated [00:17:01].
- User-Centric Design: Coming from a non-legal background, Lorra’s founders were humble and attentive to client feedback and the evolving relationship between law firms and their clients [00:10:19].
- Broad Functionality: Unlike many niche legal tech companies, Lorra aimed to service every lawyer, providing broad capabilities across the entire workflow [00:11:14].
- Enterprise-First: Lorra started by serving large enterprise clients from day one, which helped them develop a product and processes suitable for high expectations, including significant investment in SOC and ISO certifications [00:36:41].
Challenges and Adaptations in AI Adoption
User Adoption and Training
Teaching lawyers to use new AI tools requires significant effort [00:12:44]. Traditional software rollouts in law firms might see 5-10% adoption, but Lorra is achieving 70-80% adoption rates, with lawyers actively seeking access to the tools [00:13:07]. This shift is due to the inherent value and client pressure [00:13:17].
Managing user expectations is crucial [00:38:05]. Users have varying levels of understanding, from savvy associates who set up workflows and prompt libraries to those who expect the AI to “write me an SPA” (Share Purchase Agreement) with minimal input [00:38:29]. Ongoing training, onboarding, and office meetings are vital to bridge this gap [00:38:13].
Product Roadmap and Model Evolution
Balancing product development with rapidly improving AI models is a challenge [00:13:31]. The strategy is to “skate to where the puck is going” but also be pragmatic [00:13:36]. If AI labs (like OpenAI, Anthropic, Google) are likely to build a feature and make it available, Lorra avoids building it [00:14:24].
- Example: Lorra built a “playbooks” feature for negotiating documents with defined rules. While valuable now, if models become so advanced that they can understand playbooks from simple files and apply them for redlining, the feature might become unnecessary [00:14:46].
- Workflows: Complex multi-step workflows, traditionally requiring technical builders, can now be handled by LLMs that plan and execute their own workflows with tool use based on a simple instruction [00:15:51].
Pricing and Cost Structure
Lorra currently uses a seat-based pricing model, which is easy for clients to predict [00:19:23]. However, the variable cost of LLM usage (one user once racked up $10,000 in LLM costs) suggests a future shift to a platform fee with a usage element [00:19:32]. The expectation that LLM prices would continuously decrease has not fully materialized; while older models become cheaper, newer, more powerful models like 03 (GPT-4) are very expensive, though “incredibly good at a lot of legal work” [00:19:51]. This necessitates using “model pickers” to choose the best (and most cost-effective) model for a given task [00:20:29].
The Evolving Lawyer Skillset
The future lawyer will need different skills:
- Entrepreneurial and Creative: Instead of “underconfident overachievers” good at following instructions, the industry needs individuals who challenge existing methods [00:42:18].
- AI Fluency: Lawyers will be “managers of AI agents” from day one, requiring fluency in working with AI [00:43:05].
- Self-Augmentation: Individuals will be expected to augment their own work with AI tools [00:43:21].
Technical Aspects and Moats
Lorra’s architecture is designed for flexibility, allowing the system to decide which model to use for specific tasks [00:33:30]. They use tools like “Brain Trust” for thousands of evaluations to test new models quickly [00:23:44]. All models used by Lorra must pass rigorous security, privacy, and legal reviews to conform with data processing agreements due to the sensitive nature of legal data [00:23:57].
The company tests permutations of different models for each step of multi-step workflows and for the complete end-to-end product [00:24:14]. High-quality reasoning models are excellent but expensive [00:24:26].
Max Unistron believes that the “moat” (competitive advantage) for AI applications is not in fine-tuning models, as these are becoming commoditized [00:23:15]. Instead, it lies in becoming a “system of record” and a central platform for collaboration between designers, PMs, marketers, and directly with clients, similar to Figma [00:21:50]. This involves integrating with customer data, external databases (case law, legislation), and client-specific tools like CRMs or knowledge bases [00:22:12]. Lorra aims to shift lawyers from spending 80% of their time in Microsoft Word, Outlook, and document management systems [00:22:40].
A significant architectural shift for Lorra was moving from a button/workflow-based product (where users clicked buttons for structured flows) to a chat-based interface [00:39:03]. This allowed for a more flexible interface and the ability to integrate functions or tools that the chat could then utilize, despite requiring the deletion of 95% of their initial source code [00:39:41].
Company Building in the AI Era
AI companies are experiencing a new expectation for growth velocity [00:35:49]. They are not just replacing existing software but creating entirely new categories [00:35:56]. The fastest builders with the highest velocity, best product, and service will dominate [00:36:09].
Lorra’s culture reflects this rapid pace:
- High Urgency: Attracting employees with a strong desire to achieve and a belief in the mission [00:32:12].
- Founder-Driven Pace: The co-founder’s intense coding schedule (14 hours a day, 7 days a week) sets a high standard [00:32:20].
- No 9-to-5: Lorra is upfront in recruiting that it’s not a typical 9-to-5 job, focusing on building for the future rather than maintaining [00:32:44].
- In-Office Culture: Mandatory onboarding in Stockholm and a fully in-office policy foster momentum and a winning mentality [00:34:01].
The shift from a small Swedish startup (10 people a year ago) to a global company with 100 employees, onboarding 90 people in a year, highlights this unprecedented growth rate [00:34:50]. Lorra’s decision to pause sales for 4-5 months after raising $35 million to focus on product development for new markets demonstrates a mature approach to scaling during this rapid expansion [00:00:09, 00:28:28].
Other Exciting AI Application Ideas
Beyond law, Max Unistron is excited about AI applications in Clinical Research Organizations (CROs) within the pharma industry [00:40:17]. CROs are described as manual, data-rich, and slow-moving consultancies with structured workflows, presenting a massive disruption opportunity for a fully AI-powered CRO [00:40:20].
He is also excited about multimodal AI use cases, particularly working with voice and audio transcripts [00:41:00]. This includes instructing Lorra via voice and processing audio files, such as depositions, to transcribe and interrogate them as if they were text documents, eliminating the need for manual note-taking [00:41:03].