Insights from our Conversational-AI roundtable v.2

DataSeries |VRT200820

In August, DataSeries, an OpenOcean led initiative hosted a Virtual Roundtable about the future of “Conversational AI”.

Key Challenges:

Maturity of technology and dialogue

The level of improvements in the space was debated intensely where some argue that bots are still almost as broken as 5 years ago. It is clear that bots need to be more contextual and personalised in order to increase the user experience. Rasa proposes 5 maturity levels here:

Level 1: Static with no flexibility — User needs to input all information

Level 2: Basic and FAQ based — User is guided a little with questions to get an answer

Level 3: Contextual — User can ask more specific questions around a specific goal

Level 4: Consultative and personal — User can express their situation in own words

Level 5: Adaptive — The bot adjusts the level of detail based on user, picks up cues and adapts

Most bots 5 years ago were primarily FAQ based and only got more contextual during the last few years.

Customer and user expectations

Customers have high expectations on average due to increased technology adoption, business promises and media attention combined with hype. Rather than approaching the implementation of conversational AI technology thoroughly, many businesses dedicate limited resources to take out of the box functionality of vendors to a basic level.

On the other side of the equation, the user is not very patient. Companies are losing a lot of users in flawed chatbot conversations and, for instance, close to 80% will drop during a purchase in case of early bot identity disclosure before human handover. For more general platforms there still exists a high level of uncertainty of what the customer can say vs. what they actually would like to say.

Design is receiving too little attention in the context of user engagement

We’ve entered a new era of chatbots where the task is to create more complex non-technical chatbot experiences. Design is increasingly important. Both for the whole customer experience, making the implementation extremely easy, and for the end user having an appealing interface to engage with. Most bot interfaces at the moment are very similar and there is little innovation happening when it comes to the visual user engagement with the bot.

Convert language and speech effectively into the right API calls

We need to get better at understanding the mix of language and speech and then try to convert it into symbolic API-calls. Also, it was mentioned that we should be more careful about how we do symbolic type representations.

Lack of trust

Trust is still a major item and privacy a concern. Can you trust the data sources, or the models for that matter? What data sources should you aim to integrate and what information should you use and how? Also, the users has to be sure that they can trust the chatbot/assistant with goal-directed tasks. In many use cases you can elevate this concern by falling back to a human, Trust can also be increased by the right design and visual UI’s.

Making it more personal
Choose the right API integrations to collect data that enables the conversation to be more relevant/personal and tailor the content based on what the customer has already said. Some companies also use cookies (possible grey zone) to personalise based on your surf history.

Quote: From a DataScience lens, data is not the new oil, it’s the new dirt that makes people search for diamonds. These require vast efforts to be found.

Opportunities: For user and business adoption

• Enrich input- take elements such as intonation and expression into consideration: Could we increase personalization by taking additional data such as non-verbal data? Studying the tone of voice, or face expression when the user is on camera for example?
• Interface/UI innovation: (From a holistic approach) Every chatbot looks the same (layout). There is plenty of opportunity for creativity left in the UI-space and not just in the chatbots intelligence.
• Guided user discovery: Framing the user experience possibilities for the user from the beginning might help the adoption (also: expectation management).
• More blended engagement: A hybrid version between voice-commands and visual-clicking (preferably in an Extended Reality format)
• Corporates are increasing their knowledge bases, are increasingly automating tasks and need the right interfaces for this
• COVID has been accelerating the IT ticketing automation
• Gaming is definitely an under-explored area atm, that will increasingly adopt conversational AI in the future and also be able to act as a pioneer for many use cases.
• Represent more brand identity and culture in conversational interfaces. Which also includes the UI. It’s not only about the chat offering, but platforms will gradually move into a broader representation.
• Bot-2-Bot communication. Enabling companies to essentially integrate more effectively and thereby also offer more personalised experiences. Longer term from a user perspective: having a concierge bot that helps you to solve certain tasks, but acts primarily as an assistant that communicates with other bots to offer you the best possible outcome.
• Participants also mentioned transactional analytics and quantum linguistics