Enhancing digital assistant TOBi with generative AI

As a global leader in telecommunications, Vodafone is always looking for new opportunities to improve the customer experience. One great example of its digital innovation is TOBi, a virtual assistant that is available 24x7 to answer customer questions on a wide range of topics in approximately 14 different languages. It is powered by IBM® watsonx® Assistant, a product that helps build better virtual agents to drive enterprise productivity, overcome the friction of traditional support and deliver exceptional experiences for their customers.

Vodafone began exploring where generative AI could make its conversational journey building processes faster and improve the quality of the content created for each journey. Through a series of workshops led by IBM, the Vodafone team identified five key use cases that would reduce the time conversational designers and testers spent creating journeys for TOBi. They prioritized two use cases that offered the most value and were the most feasible: testing journeys and gap analysis, which is the process of identifying discrepancies between the journeys in Vodafone’s Central Library and various markets.

IBM Client Engineering and IBM Consulting® then led an accelerated five-week build phase with Vodafone. The project spanned several interconnected workstreams—data, design, architecture, development. An orchestration service was built and deployed using IBM Cloud® Code Engine, which was responsible for executing multiple test scenarios in Voice Flow (link resides outside of ibm.com). The service relied on IBM watsonx.ai™, a next-generation enterprise studio for AI builders, to simulate customer interaction with TOBi. The transcripts generated by these interactions were then analyzed using watsonx.ai. The solution provided sentiment and emotional analysis to assess customer satisfaction and identify areas to improve journey content. It was also used to deeply analyze journeys from Central Library and Markets, providing meaningful analysis on differences to help conversational designers improve the journeys.

The results of the pilot showed great potential for the impact generative AI could have on TOBi.  For example, the team observed a 99% improvement in the turnaround time of journey testing. Before the pilot, it took the team around 6.5 hours to test each new journey before it was ready to be deployed into production. The pilot demonstrated it could take less than one minute per journey per persona. Before the pilot, there was no capacity to conduct gap analysis between the Central Library and Markets. The pilot demonstrated that a comprehensive gap analysis could be conducted in less than five minutes per journey with the help of watsonx.ai. The Vodafone team also expects to achieve meaningful time savings by using AI to conduct conversation sentiment analysis. This should also improve overall customer satisfaction given the additional testing and knowledge of the journeys prior to open them to actual live traffic.