Vodafone transforms its customer care strategy with digital assistant built on Azure Cognitive Services

An evolving business  

Vodafone knows telecommunications. One of the world’s largest telecoms, the company has 300 million mobile customers who speak more than 40 languages in 26 countries.

Historically its business centered on mobile, but Vodafone has more recently expanded to include internet, TV, Internet of Things (IoT), and other offerings, aligning more with digital-native companies than with other telecoms. But as a 35-year-old company, Vodafone still had legacy technologies and disparate systems across its IT landscape as a result of various acquisitions over the years. The company adopted a digital transformation roadmap and set out to streamline and improve its environment. 

A smart, effective digital assistant 

As part of its digital journey, Vodafone wanted to incorporate AI into its customer care approach to efficiently serve its many customers across languages, products, and markets. The company’s commitment to an engaging, consistent customer service experience led it to create a digital assistant that customers could reach instantly and easily through many channels, including chat, popular messaging apps, or through the My Vodafone app.

Early on, Vodafone understood the potential for how chatbots could transform traditional customer care, but when the company started this journey in 2017, it wasn’t obvious which direction to take. After a number of small-scale experiments in different countries, it saw that it needed to take a comprehensive, global approach. Vodafone conducted a worldwide assessment of leading AI vendors, evaluating business impact, technical capabilities, a scalable commercial model, and a language capability fit for a truly global organization.

Vodafone quickly developed a bot persona named TOBi, using Microsoft Azure Bot Service. The company wanted to make TOBi as fast and accurate as possible while creating a natural and frictionless interaction to assist customers with everything from billing questions to contract updates. To give the bot natural language understanding capabilities, Vodafone turned to Azure Cognitive Services, using its Language Understanding (LUIS) API to apply AI that understands overall meaning and pulls out relevant, detailed information. This solution is part of a flexible modular platform that can also connect with non-Microsoft components.

“It was clear that Microsoft was a great fit for what we were looking for,” says Kevin Knowles, Leader, Care Transformation Program at Vodafone. “We needed a flexible, scalable natural language understanding solution that could match our growth ambitions for TOBi. It was essential that tools would be easy to use for the conversational designers in our business, built upon a modular technology to support our ambition to make TOBi the biggest chatbot in the world.”

Focus on natural, conversational language 

Key in the success of TOBi for contact center automation is the ability to engage on multiple channels, from the chat-based ones like the Vodafone website, the My Vodafone app, and social media to voice-based ones like telephony. To do this, Vodafone uses both Speech to Text and Text to Speech within Cognitive Services. Text to Speech includes leading-edge neural voice capabilities, so the bot voice sounds highly realistic and natural, and its speaking style can be adjusted to express different emotions. Vodafone ensures that customers have a smooth interaction while always making it clear that they are talking to a bot.

By using Cognitive Services and working with its in-house conversation design team, Vodafone successfully created a natural-sounding TOBi. For example, after a customer gives their name, instead of a dry request like, “Now tell me your address,” TOBi might say, “Hey, that’s a great name. Now I’d like to know where you live.” Says Paul Jacobs, Group Head of Operations Transformation at Vodafone, “We used the AI and natural language processing capabilities in Azure to give TOBi a clear personality that could make conversations natural and fun, which drives better engagement. TOBi now handles 60 percent of our customer interactions.”

During TOBI’s initial creation, Vodafone developers worked closely with Microsoft developers to customize the bot, troubleshoot challenges, and influence the product roadmap to prioritize future solutions that Vodafone needed. “My team had the opportunity to collaborate with Microsoft and make long-term plans on what to develop together, which was very fruitful,” says Nina Lange-Richter, Head of Digital Products and Portfolio Management at Vodafone.

Since Vodafone first deployed TOBi in Italy, it has rolled out 15 language-specific versions of its bot to 15 other markets using Translator, part of Cognitive Services, for real-time, AI-based text translation. 

Advanced multilingual capabilities

Microsoft knew that companies like Vodafone would want to engage as many customers as possible in their preferred languages—without adding a lot of development time and expense. So, it created a new feature, conversational language understanding (preview), to expand customers’ capacity to quickly deliver natural and fun conversation like TOBi’s in many languages. The Vodafone team in South Africa eagerly experimented with this new set of APIs and highly intelligent models while the feature was in private preview. The company plans to augment TOBi with the new multilingual capabilities within conversational language understanding once the feature becomes generally available in early 2022.

Once deployed, conversational language understanding will help Vodafone offer TOBi across more markets and communities, serving customers quickly in multiple languages. There are 11 official languages in South Africa, and the local Vodafone team is starting by supporting Zulu. “We have rural areas in South Africa where some customers only speak one other language apart from English,” says Avi Moodley, Principal Machine Learning Specialist at Vodacom, the South Africa Vodafone subsidiary. “It can be difficult for them to get the customer service that they deserve, and now we’re taking the next step to give that to them.”

With the advanced multilingual capabilities, Vodafone data science teams can train models in one language then apply the models in another supported language without retraining—thereby reducing development effort and accelerating time to value. The South Africa team can take advantage of TOBi’s previous machine learning in a variety of other languages to train new models in Zulu and, eventually, other local languages. 

“This would be a huge competitive differentiator for Vodafone South Africa,” says Moodley. “We can go to market quickly using conversational language understanding multilingualism and would probably be the first telecom company in this country to provide a Zulu chatbot for customer service. By offering TOBi in more languages, we improve the customer experience while also reducing call volumes and essentially saving costs.” 

Flexible, fine-tuned models

Conversational language understanding is based on a large language model, which gives Vodafone data scientists the opportunity to fine-tune TOBi’s conversational skills and raise the quality of its responses even further. For example, the bot can already recognize slang terms and synonyms, and through conversational language understanding, it will also be able to differentiate between similar requests. TOBi will provide a distinct, appropriate response when, for instance, a customer says they want to “delete a service” versus “cancel a service.” 

“Those two statements are semantically similar, but in some cases, one word makes the difference in how a customer gets routed to a specific product or answer,” says Moodley. “We’ll gain huge value because we can build these two entities in completely separate models, so they don’t get confused with each other. We have more flexibility in the model, and we can pick and choose where we want the models to go, which is quite powerful.”

With added flexibility and ease of use, Vodafone predicts it will decrease labor and increase productivity. “We find conversational language understanding to be a very streamlined experience, and it’s easy to bring new people onboard because they can understand a common process of building machine learning models from start to finish,” says Moodley. “In terms of production, I foresee cost reductions on the business side because we’re able to quickly build such sophisticated systems with conversational language understanding to support authentic, helpful interactions with customers.”

Ongoing improvements, personalized service

Vodafone customers benefit from the continuous improvements the company is making to TOBi. For example, Vodafone wanted TOBi to use available data to predict customer intent and say things like, “We think you need help with this—are we correct?” rather than just asking customers why they contacted Vodafone. So, it took advantage of data compiled through Cognitive Services to boost the quality and speed of its bot’s service. “After TOBi began to predict customer intent and what query needed resolution, we saw more adoption and greater digital containment,” says Jacobs.

With transactional conversations digitized, fewer customers need to speak to a live agent, freeing up more time for those agents to care for emotional or complex cases. “If you ask TOBi to connect you to an agent, he will because he’s there to help you, not to prevent you from getting through to a human,” says Lange-Richter. “With TOBi, we’re not simply trying to automate everything. Rather, we consider our bot part of a mature strategy for value generation and high-quality experiences for our customers.”

TOBi now holds 25 million to 30 million conversations a month with customers, and Vodafone expects that to grow to half a billion conversations over the next few years. “Since launching TOBi, we’ve reduced the frequency of customer contacts to call centers by 12 percent year-over-year,” says Jacobs.

Continued digital transformation

By focusing on digital transformation, Vodafone gained more agility and flexibility to respond to market and customer changes, such as during the COVID-19 crisis. Customers in many countries under orders to stay home made connectivity more critical than ever, and the Vodafone call centers experienced very high call volumes. To ensure that all customers quickly gained access to the services that they needed, TOBi made the same sales offers that Vodafone’s human agents made, such as unlimited data boosts.

“Because we built our digital assistant on Azure, we could react really quickly in response to COVID-19,” says Jacobs. “Our customers had a trusted relationship with TOBi, and he made offers when appropriate, such as add-ons and upgrades, that added value and provided practical advice on digital service when our retail stores closed during lockdown.”

Given the success of TOBi, Vodafone will continue to add capabilities to its bot and explore more potential AI and cloud benefits through Azure and other technologies. Concludes Jacobs, “We used Azure Cognitive Services and Bot Service to deliver an instantly responsive, personal expert into our customers’ pockets. Providing this constant access to help is key to our customer care strategy.”

The South Africa team is actively pursuing this mission through its early adoption of conversational language understanding. “We’re excited to use conversational language understanding to get to the point where Vodafone customers chat with TOBi and forget it’s a bot because the conversation is such a natural experience,” says Moodley.

Find out more about Vodafone on Twitter, Facebook, and LinkedIn.