This AI Tool for Lawyers Is All About Sorting Through Millions of Documents Fast

In the last year, multiple lawyers have learned the hard way that generative AI tools like OpenAI and Google Gemini can hallucinate, or make things up. But that’s not to say there’s no place for gen AI within legal circles.

For instance: electronic discovery, or e-discovery. That’s the process of collecting and exchanging electronic evidence like emails, voicemails, chats and social media posts for civil lawsuits.

That’s where data company Hanzo comes in. It helps legal departments sort through unstructured data from resources like Slack, Microsoft Teams and email to identify relevant documents and to make sense of those records in the e-discovery process.

“All this industry is based on billable hours,” said Hanzo CEO Julien Masanès. “They are very expensive lawyers at the end of the day, so we’re trying to reduce that bill.”

It’s yet another example of AI being used to bring new efficiencies to a variety of industries, including agriculture, ecommerce, media and utilities. Now it’s coming for civil law in what could be one of the clearest examples of using AI as a “superhero sidekick” yet. 

Let’s say you work at a tech company and receive a complaint from a competitor about an IP breach. A judge orders you to provide records of internal communication between employees.

According to Masanès, a large company could easily have 5 to 10 million Slack messages to be sorted through for a request like this. That’s a lot of messages to pore over to determine which are relevant. Plus, you don’t want to share any more than you have to and risk exposing your own trade secrets.

If you were to embark upon the e-discovery process on your own, you’d have to use keywords or train an AI model with some sample documents. But in the latter case, you’d still have do a manual review to ensure the model picked out the right electronic evidence. That’s tedious work for anyone.

Instead, Hanzo said, its platform can analyze these documents and identify what you need.

“People always think about chatbots when they think about AI,” Masanès said. “But there’s lots of other applications that are possible and really changing the game.”

That includes automating e-discovery, a market projected to grow from $15.5 billion in 2023 to $39.9 billion by 2032, thanks in part to the proliferation of electronic records, as well as AI and automation.

Earlier this month, Hanzo announced the availability of its Spotlight AI technology, which it says helps lower the cost of applying large language models to businesses’ legal cases.

Instead of using a popular model like ChatGPT or Google Gemini, Spotlight AI uses smaller models that are less expensive and uses them only when needed, which a Hanzo spokesperson said “better aligns cost to a company’s usage and the value derived from AI.”

Large models have rate limits and usage tiers for their APIs, which range from requests per minute (or day) and tokens per minute (or day). A token is equivalent to about four characters in English. AI models break down text into tokens to help them process and understand it.

Hanzo argued the token-based pricing model can be cost-prohibitive for legal teams in data discovery, given the vast number of documents that need to be processed. Hanzo instead charges by gigabyte, which it said lowers the cost substantially. (To be fair, Google Gemini 1.5 Flash has a free tier with rate limits of 1 million tokens per minute or 1,500 requests per day.)

“We believe that being able to scale millions of documents very efficiently is super important,” Masanès said. “I think bringing this much more interesting price point will be something that really changes the game in our industry at least.”

This is one of a series of short profiles of AI companies, to help you get a handle on the landscape of artificial intelligence activity going on. For more on AI, see our AI Atlas hub, which includes product reviews, news, tips and explainers.

Editors’ note: CNET used an AI engine to help create several dozen stories, which are labeled accordingly. The note you’re reading is attached to articles that deal substantively with the topic of AI but are created entirely by our expert editors and writers. For more, see our AI policy.

Read More

Lisa Lacy