How conversational AI is transforming customer interaction
Suhas Uliyar is VP Digital Assistant, AI & Integration at Oracle. His passion for AI was born out of a history of working with mobile technologies stretching back to the 1990s. “I became fixated with the customer experience and digital experiences. That led me to start my own startup around mobile technologies, building up a mobile platform that allowed developers to build good mobile experiences,” he says. As mobile technologies evolved with the advent of the iPhone, the digital experiences they enabled led to what Uliyar describes as his first foray into AI with predictive analytics, before joining Oracle in 2013 to drive its mobile strategy.
With customers so used to interacting via mobile phone technologies such as SMS or WhatsApp or Facebook Messenger, Oracle identified an opportunity to use natural language processing to allow users to converse naturally with AI chat bots. “We quickly pivoted over to the chatbot world,” says Uliyar. “We used natural language processing to automate things like customer services, building the chatbot platform from scratch.”
Uliyar emphasises the fact that Oracle has four pillars that guide its AI approach, including building, training and managing machine learning models on Oracle Cloud Infrastructure, adaptive intelligence applications with out of the box AI capabilities, and building machine learning into the product it’s best known for: its database. Uliyar’s division constitutes the fourth, which is focused on AI services.
AI services involve facilitating the addition of AI capabilities to applications, without having to start from scratch. Conversational services are among the most popular, such as adding speech recognition capabilities, as well as computer vision solutions for image recognition. “There's a whole bunch of these AI services,” says Uliyar. “For example, a conversational AI is what we use for our digital assistant and chatbots. Our technologies are used by developers who want to do things like text classification, entity recognition, or aspect-based sentiment analysis and so on.”
One concrete example of its utility, applicable to most companies, comes in expense automation. “What you want to be able to do to your expenses receipt is take a picture and then apply optical character recognition (OCR) to it,” says Uliyar. “Then, once they’re recognised, you want to be able to automatically label them - that's the name of the restaurant I went to, this is the tip amount, etcetera. Being able to use AI to recognise those things is crucial, and that can also be used for invoicing, for example, or bills of lading.”
One of the ways Oracle differentiates itself from competitors is by considering the use cases of enterprise customers. “One really unique thing about our speech model, for example, is that we are able to handle a lot of domain- or customer-specific vocabulary,” says Uliyar. “If you look at some of the speech providers today, like Siri, Google or Amazon, they do a great job in open domains. But, if you ask questions with complex terminology, such as EBITDA, they break down. A lot of enterprises have their own very domain-specific vocabulary. With AI, we’re able to let our customers train those language models, without our direct involvement.”
The ‘Digital Assistant’ component of Uliyar’s title is one that will be familiar to most of us, having as we do such technologies embedded into our phones and other smart devices.
Oracle’s exploration of the technology started as an amalgamation of a large number of chatbots that had been specifically designed to excel in one area. The issue was that a company’s needs often evolved, and it was difficult to select the right one for the job.
“What we did was build an AI-powered digital assistant,” says Uliyar. “It's essentially your one window through which you interact with all these other chatbots, which are subject matter experts, and which know how to then connect to the system of record. With other solutions, you ask for a specific skill to do something right here. With us, it's all implicit. One day you might ask about your paycheck, another about your vacation balance and the digital assistant is smart enough to route your question to the relevant bot.” With this focus on natural language, it was equally vital for the assistant to be able to comprehend the tangential and nonsequitur-laden ways in which human beings speak and think. Consequently, the assistant is designed to recognise context, and figure out precisely what area is being talked about.
Companies are using Oracle’s AI technologies in many and varied ways, with particular demand for customer service and support. Customers include delivery firm Hermes Group which has partnered with Oracle for over two and a half years. “The net promoter score (NPS) has really gone up from the customer perspective, and from the enterprise perspective, the cost of support has gone down significantly. We are able to deflect 50 to 60% of the calls to be handled by the digital assistant, as opposed to going to a human.” Uliyar stresses that any escalations go to a human agent to avoid a frustrating experience. That same technology is also being put to use inside organisations, a particularly invaluable service in the age of COVID-19, where IT departments are pressured to support geographically scattered employees.
Uliyar sees the future of Oracle’s AI platform as allowing its customers access to truly intelligent process automation, rather than the more naive robotic technologies that involve screen scraping and emulated mouse movements. “We’re looking at a suite of business process improvements and tools that assist the knowledge worker by removing repetitive, replicable and routine tasks - for example recommending the next best action. An AI that’s able to look at how you've done things and provide recommendations and forecasting - ‘If you continue down this path you're going to have this problem’.”