Interview: The pursuit of loyal chatbots

By: Ryan Daws

18, September, 2017


Artificial Intelligence - Chatbot - Chatbots - Consumer - Featured - Machine Learning -

istock peshkova

AI Tech spoke with Alex Bunardzic, senior software developer at Staples, about the future of chatbots and the pursuit of their loyalty.

“You go where the market is. Now, businesses are realising the market is in the chatting channels.”

“People spend a lot of their online time chatting either through SMS, Messenger, or other channels,” says Bunardzic. “Previously, it was the trend that people were spending more time on social media and that’s where businesses were flocking to – you go where the market is. Now, businesses are realising the market is in the chatting channels.”

Chatbots are not a new concept but recent advancements in AI are enabling them to become far more useful. Bunardzic has three categories for chatbots; Non-Stateful, Stateful, and Loyal.

  • – Non-Stateful: This category represents the most basic form of chatbots and they can only respond by looking for scripted questions. They do not remember anything of a conversation with a user. Most of the earliest chatbots are non-stateful.
  • – Stateful: These chatbots are generally more useful as, for that session at least, they will remember earlier answers from a user. Many of the latest chatbots fall into this category.
  • – Loyal: Chatbots in this category represent the most complex and they can remember details of previous conversations with a user even after the chat has closed. These bots are considered the ultimate goal and recent AI developments are helping the industry to reach it.

An example of a non-stateful chatbot could be a weather bot which listens for a city and responds with the forecast. A stateful delivery chatbot may ask for a name and order number to provide tracking information. A loyal car insurance chatbot may keep your details on file and send you a message when you’re due for renewal if it’s found a better deal.

“You can have a really simple chatbot which treats every message as if it’s coming in for the first time — you could be chatting to a bot for hours, and it will not remember your previous messages,” comments Bunardzic. “They’re not very useful and people don’t find them engaging.

“From that, you can move into chatbots which do a little more heavy lifting and remember previous responses in the conversation. Then you can build what I call ‘loyal’ chatbots which actually learn about you, stay with you, and maybe form a relationship over numerous sessions to give a more quality response to queries.”

Moving from non-stateful and stateful chatbots to loyal will also enable them to become reactive and even proactive. For example, with a customer’s postcode details on file, a chatbot for an internet provider could alert specific users of an outage or planned maintenance which impacts them. This, in turn, provides comfort to customers and reduces the number of calls made to service centres which allow more staff to be focused on real issues faster.

“It’s not easy to offer common sense dialogue between machines and people. Machines tend to get confused or go into a loop and become less effective. Right now, to remedy that, a lot of businesses are thinking about escalating quickly to a human operator,” says Bunardzic. “We start a dialogue with a chatbot and when something becomes iffy it will escalate to a human.”

Most readers will have reached an automated voicemail at some point and listened to a multitude of options before thinking “Just connect me to a human operator already!” Then, to make things worse, the dreaded hold music comes in…

When asked if there’s a concern that chatbots which ultimately escalate queries to humans could become a similar source of customer frustration, Bunardzic says: “In a way, it’s worse, because people today [in a text-based chat] feel they can start chatting about anything. If you’re in an automated voice menu they cannot branch off into other topics.”

“In today’s freeform input boxes, we cannot constrain the conversation and often we get really weird or off-topic messages. People can also get frustrated in other ways like if they ask a question that has nothing to do with your service and you fail to respond they can feel betrayed.”

Businesses small and large are waking up to chatbots’ potential. Microsoft has been among the most prominent advocates of chatbots and envisions a future where one can even speak to another. For example, you could ask Microsoft’s virtual assistant, Cortana, to order a pizza and it could bring in the chatbot for Domino’s Pizza to complete and track your order. This is similar to how ‘Skills’ function on Amazon Alexa or ‘Actions’ on Google Home in the voice assistant realm.

“If you are offering services via chat that means you don’t need to worry so much about things such as graphical interfaces and performance,” explains Bunardzic. “If you are building a traditional app, it needs great performance otherwise you will lose your customers.”

Voice assistants, of course, remove the need for a graphical interface altogether.

“We call them ‘screenless interfaces’ because you don’t need a screen to accomplish something,” says Bunardzic. “Especially in situations like when you’re driving, if you can interact with your voice and listen to responses then it’s very desirable and I think will become very popular.”

When asked who he sees as being a current leader in the chatbot space, Bunardzic responds with Berlin-based open source conversational AI company Rasa.

“They are offering a very good platform for machine-learning probabilistic service that’s open source and in my opinion, from what I’ve seen, it gives the best chance to bot builders looking to offer services,” explains Bunardzic. “They offer an opportunity to train up the bots and teach the bot to learn about your contacts and answer various questions.”

“When you ask something like ‘Where is the nearest hospital?’ people can ask that in a variety of ways and the challenge sometimes with a bot is that it’s too rigid and can only understand certain ways of saying it. Rasa allows us to teach a bot how to respond to various ways of saying it and I think it’s very promising.”

Back at Staples, where Bunardzic currently works as a senior developer, the company is looking at using chatbots to help with the diverse work of the organisation. He notes there are many smaller teams within Staples and the manpower is not scalable so customers can end up being left on hold for a long time and getting frustrated, or emails not responded to for days.

Bunardzic and his team are being tasked to automate services so that chatbots can respond immediately in many cases to alleviate customer frustration. Some, of course, will need to be escalated to a human but these queries should be answered quicker due to others being automated.

The excitement around chatbots is growing, and our chat with Bunardzic confirms it’s for good reason.

You can hear from Alex Bunardzic at AI Expo North America at the Santa Clara Convention Center being held November 29th – 30th. He will be giving the ‘Chatbots – From Chat To Full-Serve Experience’ talk and will also be talking on the panel ‘The Realities of Bot & VA development & implementation’.

What are your thoughts about chatbots? Let us know in the comments.

© iStock.com/Peshkova