Agiloft launches AI Trainer to deliver “fully individualised AI contract analysis”

Legal IT Insider speaks to Agiloft’s chief product officer Andy Wishart about the new release

Contract lifecycle management vendor Agiloft today (15 August) announced the release of AI Trainer, which enables non-technical users to customise the way they review and analyse contracts. 

Where corporate legal teams have largely had to rely on pre-trained generic AI models, AI Trainer offers users a no-code environment to create their own AI models to better automate the process of reviewing large numbers of contracts. AI Trainer is a built on proprietary Agiloft technology but at a foundational level, the language model that underpins AI Trainer is a derivative of Google’s BERT. 

Speaking to Legal IT Insider, Agiloft’s chief product officer Andy Wishart said: “The AI Trainer is about empowering and enabling our customers to train an AI model on the unique data points or the unique ways in which they craft their clauses and make use of that trained AI on their content.

“It can identify clauses during use cases like third party review and negotiation, and also help with the remediation problem: when a change happens in the environment you need to be able to quickly identify those contracts within the repository with that particular clause, or a particular key piece of data. So we’re putting it into our customers hands to have AI their own way so they can take our ethos of having no-code configurability across the platform, but now applying that to artificial intelligence as well.” 

AI Trainer, which is a commercial add-on, is found within the Agiloft AI platform. Users upload their documents and tag the clauses that they are interested in, such as a Force Majeure clause. Once tagged, the user can mark the documents as ready to train and Wishart says: “You train the model to identify those clauses in documents that you have not seen yet.” Users train the model on three quarters of the documents uploaded and test it on the remaining quarter. 

While pre-trained models will continue to be available ‘out of the box’, Wishart said: “We think it’s important to provide customers with the ability for them to train models based on their own content because they may have variations in how they write a particular clause, or within their industry there might be specific data points or clauses or parts of clauses which they need to be able to identify quickly. While pre-trained labels are important, I don’t think they are enough to provide a complete AI capability for customers, which will be a combination of pre-trained and the customer being able to train their own models.” 

Agiloft in January this year launched ConvoAI, powered by Cognizer’s Genius platform, to enable users to interrogate their contracts in natural language without key words or filters. 

However, Wishart told us: “The difference between that solution and what we’re talking about enabling here is that with AI Trainer, once you train that model and apply it to your documents, you’re enriching the contract records. You’re taking the information that the AI has found and you’re storing that information and data within our search solution. You might not have any data about your contracts but we’re going to ingest them, put them through that large language model and place the content on a knowledge graph.” 

To learn more about the new Agiloft AI Trainer you can join its 2023 Summer Release webinar