Conversational AI, the patterns in it..

This picture of my son talking to Siri when he was just fours and half years old left a lasting impression on my mind and that is when I starting exploring the potential of AI and its lasting impact to our daily routine.

So you see, he is talking to SIRI or shall I say conversing with an AI BOT? In this particular situation I was sitting in the living room and this guy was in the bedroom and I could hear him talking to someone, at that time he was a single child so he used to talk to his toys a lot, just like any other kid of his age, so initially I thought he is doing the same, but something caught my attention and I decided to go over to see what exactly he is up to. I realized he is holding on to my iPad and speaking to SIRI and the best part was he was asking questions that he would usually ask while making new friends. The kind of interaction that was happening between him and SIRI was funny and interesting at the same time. He actually asked SIRI, where do you live and SIRI said right here.. Then he was like do you live here in my house…I would term that as Conversational AI!!

But what is it anyways as per the technology pundits? Conversational AI refers to the use of messaging apps, speech-based assistants and chatbots to automate communication and create personalized customer experiences at scale.

With every passing day Conversational AI is emerging in more and more aspects of daily life, benefiting consumers, employees, and businesses. Few examples are:

• Automobile: In-Vehicle assistant
• Retail: Shop Assistant, Sales Agent
• Home Automation: Home Assistant, Elderly Companion
• BPO: Customer Care Agent, Agent Coach
• IT Security: Cyber Security Expert, Compliance Expert, IT Helpdesk, HR Advisor
• BFSI: Financial Advisor, Mortgage Advisor

There is an important word associated with Conversational AI, how do we define the elements of conversation that is usually covered in the AI?

Easy definition as per the below picture: Conversation = Channel + Content + Context + Collaboration

We will go through each of these terms one by one to get some more details nnd then we can realize the key attributes of conversational AI.

Channels: Channels have evolved from simple phone call to website chatterbots to today’s plethora of channels. Choosing the right channel is key to make virtual agents successful. Matching right use case with right channel is recipe for success. Customers start to expect omni-channel experience, start from webchat, continue on messenger & close on email.

Considered as natural to human, voice channel is challenging to implement & scale compared to chatbots. We are already realizing many smart things (like Alexa) acting as customer facing channels for virtual agents. They are handy, easy to use & easy to integrate. We already have advanced bot channels using ‘all senses’ for complete view of customer experience and relying on emotional analysis to alter virtual agent responses, interesting isn’t?

Now let us talk about Content or shall I say Intent that we tend to exchange over these channels?

Content/Intent: the ideas that are contained in a piece of writing, a speech, or a film

Content is the information exchanged during natural dialogue with customers while context (will cover just after this topic) enables conversational applications to anticipate user needs. Thus, the content enables conversations.

Every bot is different – different persona, tone of conversation (formal/ informal), use of emojis, abbreviations. And their usage is also diverse, providing general information or notification, question/ answer solution. The key trait is to be able to detect and respond in user language and it is applicable to multi-lingual websites, applications.

All of us must have used guided conversations, that was very restricted in nature, users were expected to follow specific path through use of UI elements, such as buttons. With Conversational AI free conversations has taken up that space and that gives good feel of natural conversation (key to note highly dependent on NLU effectiveness).

How would you feel when you don’t have to run follow-throughs, give simple actions or commands to virtual agents – creating tickets, sending emails, etc. Enabled with transactional dialogs, capturing customer inputs to arrive at a decision, e.g. mortgage advisor, trip booking, Conversation AI powered with right “Content” is a game changer.

Content must be complemented with Context Setting as explained above.

Context: The challenge is to interpret behavioural and emotional interaction in a voice conversation in order to deliver an impactful experience. It helps to understand both customers’ communication style and suggest the next-best action to agents in a centralised view. What is the context in plain English? This is what I found on dictionary.com ??

the parts of a written or spoken statement that precede or follow a specific word or passage, usually influencing its meaning or effect:You have misinterpreted my remark because you took it out of context.

the set of circumstances or facts that surround a particular event, situation, etc.

So, you see, if is critical that we stay within the context while using our Conversational AI and avoid going out of context.

We have context defined at the enterprise level for ex. various Industry specific, purpose specific, customizable/ trained domains for ex. book a ticket vs raise a ticket, apple phone vs apple fruit.

• Jumping in and out of use case flows, for ex. checking account balance, checking nearest branch, checking credit card balance
• Remembering last conversations and trends to offer hyper-personal service ex. knowing user always uses specific airlines, or restaurant.
• Walking a decision tree to support conversation flow. Executing pre & post step operations within conversation context.
• Supporting multiple intents/ entities in the same utterance and slot fitting/ prioritizing intents.

Let us see some example where the bots have gone wrong with lack of context/intent training:

Not able to restart/ switch context

Does not remember the context

NLU is not well trained

The last but the most important part of the conversational AI is the collaboration.

Collaboration: What would you like a AI bot to do if in case it does not have the content or unable to evaluate the context?

the situation of two or more people working together to create or achieve the same thing

Escalate to human in case of incomprehension or request for escalation and this can also be driven through sentiment analysis.

You will find scenarios where the teams are formed where the first point of contact is a bot and if bot does not respond within SLA, “Human” team member picks up conversation. Handing over conversation context to human during escalation. It would help in:

• Stop annoying repeats.
• Supporting agents to serve customers better – agent coach, FAQ bot for agents.

That is not all, AI Bots can keep on learning on their own based on Human responses.

How does it impact the role of customer support agents? It has already evolved their role and now they are working with AI bot as their team members and the expectation from agents is to possess ‘Analytical’ personality profile, to be able to work on more productive areas than at the transactional level.

It is clear if we can check all the boxes (as mentioned above) with Conversational AI we would get:

• PERSONALIZED Predictive, dynamic experience based on customer profile
• ENGAGING & EVOLVING DIGITAL EXPERIENCE Multi-media, contextually aware interactions that learn and adapt over time
• QUICK, EFFORTLESS RESOLUTION Guided conversations through interactive, multi-media, one touch formats with faster resolution than human or regular chatbot
• SEAMLESS TRANSITIONS Persistent conversations that carry all context throughout escalations and channels
• HIGH CUSTOMER SATISFACTION Consistently scoring 2-3X higher customer satisfaction than human-only experiences

Huge opportunity for the contact centre business to realize the benefit of Conversational AI and ultimately add to opex savings. Are you in for it?