Internal Process Optimization using GenAI and LLMs
Business Challenge
As part of a transformation initiative aimed at better servicing their clients, PwC began a department wide overhaul of their internal knowledge portal. Prior to working with Shorthills AI, associates would manually search through thousands of legal cases, old judgements, and opinion pieces to guide their clients on probable outcomes of tax notices. This process took hours per case and limited scalability. The two areas of focus were to improve internal search capabilities and to enable their clients through a self-serve portal. The goal was to have clients submit their notices and leverage GenAI search functionalities in real-time to gauge probable outcomes of their tax notices.
Proposed Solution
Based on the above requirements, Shorthills AI assembled a team of specialized resources including a product manager and a solutions architect to map the new process and select the appropriate AI methods to deliver on objectives, while still being cost effective. AI aside, the expertise to understand the complexity of taxation rules and laws in a short period of time were critical in the execution. The core techniques used were semantic search, summarization, chain of thoughts and natural language response.
Key Outcomes
After the deployment of these models, PwC has seen a drastic decrease in the time it takes for associates to compile findings from historic cases and offer guidance to clients. On average it now takes 6 minutes to retrieve and summarize cases versus 1.5 hours prior. Furthermore, the customer experience has evolved; enabling clients to quickly become educated on probable outcomes - something that was not possible prior to the creation of the portal. Scalability, cost optimization and data engineering expertise were the foundation of a successful phase one transformation with PwC.
Tech Stack
GPT 3.5/4, RAG, Intent recognition, PubSub AI, Topicfinder