Pegasystems, the customer relationship management and business systems specialist, recently rolled out Pega GenAI Knowledge Buddy, a generative AI assistant that provides natural language answers to queries.
Seeing an opportunity to harness the technology, Pega’s IT team tailored Knowledge Buddy to help support staff address customer queries more efficiently. Known as "Support Buddy" internally, it ingested different sources of Pega’s information to serve as a useful sidekick for addressing many customer queries.
By referring to product instructions and other information, Pega’s Support Buddy helps solve simple queries, reducing the load on customer service agents and freeing up human capacity to handle more complex tasks.
“It’s trained on product documentation, how to's and forums to generate the answers and be able to present that to customers,” said David Vidoni, CIO at Pegasystems.
Compatible with native Pega applications and other tools via APIs, Support Buddy can draw on different types of information belonging to an organization to generate responses. This allows organizations to customize it to assist in many areas, including customer support, HR, marketing, sales and contracts.
For example, sales people or customer service agents can pose specific questions about company procedures or product offerings and it will return answers without them having to spend time searching out the information themselves.
AI reduces ticket volumes
Pega has found that its Support Buddy is particularly helpful with a common type of account management task, where users have certain permissions but are blocked from carrying out other tasks and are not clear about why these are off limits.
They typically then turn to the customer support channel for help, but this happens regularly, creating an ongoing source of queries for the same type of help. Now, with an AI helper, the tech provides the customer guidance to help them understand and address the problem
“They put a ticket in and the bot will return information to guide them through solutions on how to resolve their issue,” Vidoni told CX Dive.
The business was receiving 300 to 450 of these kinds of requests per month, which are highly repetitive for agents to deal with and very disruptive for clients, according to Vidoni. It was the perfect kind of task to set AI onto and improve.
Implementing the bot has helped to dramatically reduce this workload, said Vidoni. In six months, Pega has seen a 65% reduction in tickets that customer support teams have to address.
The nature of the technology, which uses the retrieval augmented generation technique, avoids time-consuming LLM model training and fine tuning because it allows the business to just point the model at its documentation as the source material for responses.
To help with authenticating information, Support Buddy provides footnote-like citations to show the page where it’s sourcing the information in case the user wants to double-check it themselves.
Where support agents are addressing customer queries, the system helps generate the responses, doing away with the need to write it out each and every time. This is part of Pega’s continuous improvement process to identify areas that can be honed, whether it’s the algorithm or the content, to ensure it’s effectively addressing customer inquiries.
“This also helps the system identify patterns and where certain types of queries are coming in over and over again so it can help to drive prioritization around content that may need to be updated or expanded,” Vidoni said.
Rewriting the UX rules for customer help
Internally, Pega has created more than a dozen Buddies to help staff find information they need using natural language and to get a quick response. For example, a "People Buddy" helps employees more easily find HR policy information, and "Contract Buddy" helps find the latest information on vendor and client contracts.
Using these buddies for customer-facing services as well as internally creates a consistency in their approach to organizing the business’ information.
“Deploying it on the front end for customers, while also having it available at the back end for support agents, really helps to level that playing field for the business,” he said
Even so, integrating the generative AI internally and externally at the same time is a complex process not without its risks and challenges.
Spinning up a demo or proof of concept tool is relatively easy, but the benchmark is ensuring it provides credible answers, the system is secure and it can be connected and index information to provide meaningful answers quickly, Vidoni explained.
It relies on a good content strategy to keep the source information accurate and relevant and meet security and performance requirements.
“What we build needs to have a solid foundation and be able to be maintained and improved over time,” he said.
On a macro level, Vidoni believes generative AI-powered tools have the potential to change user experience taxonomies and information management. He expects to see a shift away from people needing to locate information themselves and instead relying on natural language queries and responses.
This has the potential to enable a unified experience for customer queries while meeting the needs of customers who search for information in different ways.
“Having the simplification layer of generative AI that can do the searching and summarization for you can improve the user experience for anyone interacting with the site,” he said.
