Making dreaded TnCs easy with LLMs
Terms and conditions are all around us. We are so surrounded by TnCs that we don’t notice the same while buying products.
The main pain point of reading TnCs is their length. Can that be solved via LLMs?
In many industries like financial services, reading TnCs/product descriptions/agreements is very important for the customer.
Can we garner trust by not hiding TnCs in legalese and long length but instead making it the most transparent thing for customers by allowing them to interact with the document via LLMs?
LLMs are great at retrieval. And the field is rapidly advancing. Customers can ask any question about their products/agreements and that can be answered by the LLMs with the exact reference from where it found the answer from. So customers will have all their doubts cleared and they will know exactly where in their document is the answer referenced from.
This idea was inspired by a conversation during a session wherein I asked people what are the biggest challenges they feel they face in digital lending today. And many people responded to the fact there is a deficiency of trust and most people don't have the time or the patience to go through the Key Fact Sheet (KFS)- a document mentioning all the details of the loan that the user has chosen, leading to either drop-offs or reverting to an agent who might not have the best intentions at hand.
Technologies like the above will help reinforce trust making it easy for consumers to trust the entire digital process and make them feel like they have a truthful interested digital assistant at hand.
Below is an image from a demo that I built using retrieval augmented generation(RAG)- an emerging concept in the field of LLMs. The document fed to the LLM is a sample Key Fact sheet as part of loan processing which mentions all the details of the loan.
For example: if the user asks ‘what is the cooling period?’, the model will search for the answer from within the document and answer it for the user as an output.
The output answers the user’s questions and also tells from which page the answer is being referenced from reinforcing trust.
You can find the Github code here.
Hopefully with the spread of LLMs and by making them more accessible, users will be able to get their key points of concerns cleared from large documents like agreements or Terms & Conditions (TnC).
Thanks!
