Conversation AI is the new UX
The use of chatbots is a fast-growing business that meets the needs of all stakeholders involved. According to some recent researches, 80% of custom interactions can be resolved by well-designed bots (Accenture), 60% of the consumers want easier access to self serve solutions for customer service (Ovum), and 50% of the enterprises will spend more on bots than traditional app dev in 2021 (Gartner). It is then a great market for all developers.
The use of various flavors of artificial intelligence in this specific area has been widely analyzed during the two days of Codemotion Milan 2019. One of the most relevant talks was held by Google’s Priyanka Vergadia.
“Why are chatbots becoming an increasingly powerful AI application? Because they make the users happy, so they are positively impacting every aspect of businesses”
Together with the use of natural language and related technologies, updated chatbots are becoming the new User Interface.
Target harware is in the IoT area
As a Developer Advocate at Google, Priyanka has helped multiple brands build and deploy Cloud AI solutions. Priyanka works with companies to build and architect their AI and machine learning strategy around customer experience. She has been working with different technologies in the past 10 years expanding her breadth of solution portfolio ranging from Contact centers, IVR technology to natural language processind, ML leading into modern conversational experiences.
Voice-based approaches exist since the 1950s, but they didn’t work that good. Coming to more recent times, “IoT is a key element today”, states Priyanka, as “we are interacting with so many devices at the same time”, thinking of coffee machines, washing machines, voice assistants, and cars. This is a paradigm switch we have to take care of when designing conversational approaches to customer care. The new interfaces must also be fully integrated with the overall business software inside the company.
More skills are needed to keep relevance today. “I would say that mandatory knowledges are Python, an understanding of AI and ML at a high level, and natural language processing”, tells us Priyanka. “Serverless functions is also something to learn and know about for those backend calls to a Database or 3rd party API”.
Why do conversational experiences fail?
The normal approach has always followed a dead path. “Flowcharts are disappointing” in describing the conversation, so there was the need for a quantum leap. Many aspects have to be put into account using AI and bots, and software code must be written a long time after the beginning of the bot’s analysis.
Artificial intelligence is still not that accessible and what’s needed here is something that doesn’t fit in a simple box. AI zone encompasses ML zone that, in turn, encompasses the DL zone: conversational AI can’t be confined in one of these areas, but is asks for solutions that can be thought as coming from all three zones (AI, ML, and DL).
Coming to bots, they try to handle too many things at once, and don’t normally communicate with existing business systems. “A proper bot solution is integrated with existing software in a multi-channel solution”, reminds the developer advocate.
Any voice chatbot must implement NLU, Natural Language Understanding, that today can be easily managed. AI and bots communications are technological problems that can be easily solved with correct analysis and its related business plan.
The largest part of the overall conversational user interface is thus outside technology. You need a correct approach to designing a solution. Today’s use cases are not that strong: “some bots lack transparency, they don’t understand the context, and they lack proper human escalation protocols”.
How can we untie all these knots? There is a need for a strategy.
Voice is a good approach to B2C, B2E
A world-class conversational interface is a multi-step process. It can be scheduled in seven steps:
- identify top user journeys that could be automated;
- build a business plan;
- design a persona;
- write professional scripts;
- connect to backend enterprise system;
- build a proof of concept and collect feedback;
- connect to human agent systems for human intervention.
This approach is giving his best in some important areas. Today’s most advertised area is at home and in the car, with IoT-based home assistants. The traditional area of business was B2C customer service, but there is a lot of space also inside the enterprises with B2E organizational knowledge, onboarding, and helpdesk.
DialogFlow is the tool towards conversational AI
What is needed to develop a solution? Google proposes DialogFlow as a tool to write conversational AIs. At least three points must be clear: intent (reading the important verb “You have to understand the way people asks questions; the same thing can be asked in thousands different ways”); entities, such as names, time, roots, synonyms “appointment at 3 PM tomorrow is an example from which to extract three variables to send to my backend”; and obviously the context: “we understand each other based on some assumptions” that make the context, and “you have to continuously train your model”.
Your AI bot should work on more channels: web, car, tv, home. And it must connect all to the backend through robust APIs. “The connection to your backend is crucial: without it, all the work you’ve done is a waste”.
