How Fiserv improved a cumbersome client service experience with a Watson-enabled virtual agent
- The IT team at Fiserv uses AI to streamline customer and agent interactions
- Watson Assistant efficiently handling Fiserv’s unique acoustical challenges
- With an AI solution already in place, reacting to market disruption is a snap
Promise and pitfalls
Fiserv is an exciting example of AI’s power to increase employee and customer satisfaction at scale across the entire enterprise. The Fortune 500 company provides its clients with financial services technology such as ecommerce solutions, ACH solutions, and point-of-sale technology. After a brief R&D period, Fiserv successfully deployed an AI virtual agent named Mave for customer support, then rapidly scaled the AI solution to significant effect across the entire business.
At the outset, Fiserv’s operational and technical leadership approached AI through one of their most challenging use cases: legacy credit card readers. As a financial services company, Fiserv works hard to keep transactions made by merchants and their customers secure. Legacy credit card terminals require software updates to remain secure. At the time, these devices didn’t have an easy solution to trigger updates. There was no app store to pull updates from, so updates required a phone call to a Fiserv contact center, which could often take 20 minutes or more to complete. Once the merchant connected with a live representative, the agent walked them through a time-consuming software update process. It was a frustrating solution that impacted the overall client experience.
Technology that learns with you
In collaboration with IBM and the Watson team, Fiserv began to gain understanding and confidence in AI as a mature technology capable of handling their specific needs. It wasn’t long before they recognized that AI could serve as a reliable solution to the challenging and time-consuming process in the Fiserv contact centers.
Intrigued by this new possibility Ryan Susman, Director of AI/ML at Fiserv and his team dove right in to explore how the three core components of IBM Watson technology — Watson Speech to Text, Watson Text to Speech, and Watson Assistant — could solve their problem. In a few weeks, a proof of concept was authored by the team, which easily incorporated several key customizations that adapted the technology to meet Fiserv’s expectations. These customizations included precise acoustical data modeling, teaching the AI to understand the structured data within Fiserv’s internal corpus, and natural language modeling.
To better understand how the solution operates in near real-world conditions, and to make adjustments pre-launch, Susman’s team asked associates from across the firm to have a verbal conversation with the bot. They then conducted several exercises to improve exchanges and continuity between the merchant and the solution, which they have since named Mave. These exercises evaluated the fidelity of the voice, intonation, listener fatigue, and most importantly, whether the solution understood what the merchant was saying. It was important for Mave to understand the merchant in any context, and merchants will sometimes be in a noisy area, perhaps even on a low-grade phone line. So Mave’s choice of words was carefully crafted to be well understood by all kinds of customers in all kinds of situations.
Another top priority was Mave’s ability to have a smooth conversation and log incredibly fine detail about each interaction. The team employed natural language processing (NLP) strategies to analyze every interaction in the large data set and quickly find actionable opportunities to improve Mave’s understanding. This enabled Fiserv to make quick adjustments to Mave and reach measurable success in hours rather than days.
A win/win situation
Ultimately, the deflection of a 20-minute phone call from the contact center to a digital channel is a win/win for merchants and contact center agents. The merchant receives a consistent service, and the Fiserv contact center associate has time to focus on more complex calls and deliver better client experiences. Neither has to deal with a tedious update process. Fiserv improved reported contact center agent happiness and merchant NPS. Traditional calls outside the use case also saw improved service, as Mave freed agents to handle more diverse cases.
Since this first implementation, Fiserv has developed Mave year over year. In the first half of 2021 alone, Mave has serviced tens of millions of unique customers. In addition, Mave now plays a large role in Fiserv’s IVR and chatbot systems. Susman and his team have also applied further AI tools: Watson Discovery and Watson Studio grow Mave’s proficiency in solving complex challenges in natural language and beyond.
How they did it
Fiserv’s specific business model offered two substantial challenges.
Cognitive understanding: With much of their data housed as an audio stream, Fiserv needed to ensure that the AI could recognize specific nouns in the context of Fiserv, so teaching the NLU to recognize a variety of contexts was important.
Acoustical understanding: In a global business, successful implementation must recognize auditory models for merchants in any location on the globe to separate words from noise, accounting for background noise and dialects.
With the onset of the COVID-19 pandemic and lockdown, many organizations transitioned to a digital-first philosophy. This change caused a ripple effect in the language customers and merchants used to make a transaction. Many terms new to the domain of global business became commonplace and had to be reflected in interactions with Mave. Terms like quarantine, PPE, lockdown, pandemic, contactless service, flattening the curve, and more. Best practices with NLP enabled the Fiserv team to respond quickly to the changing dialog and leverage Mave to see success through the pandemic.
How you can do it too
Susman credits Mave’s success to the overall attitude of curiosity at Fiserv. When Fiserv introduces the technology to customers, they get two kinds of response: those who are excited to embrace and explore, and those who must understand everything about the technology before taking the first step. Fiserv found success by diving right in and seeing the actionable opportunities living in their data.
Fiserv’s five steps for success in AI:
- Use the data you have on hand to target an area for experimentation.
- Don’t be afraid to experiment fast — your results will likely be actionable.
- Build your AI use case from those results.
- A modern business is AI-literate at a holistic level. Be sure to involve stakeholders and business leaders across the organization.
- AI technology resists silos. It can bring untold benefits to nearly every facet of your organization.