Following an article published on conversational AI on UC Today, Nikki De Kretser, CMO at Cirrus, reached out to bring me up to speed on their artificial intelligence and chatbot capabilities.
In my original post, I mentioned that Cirrus’ initial demo of their AI tool at the time blew the room away. However, as the industry changes and matures, so do products. Today’s conversational AI is perhaps even more intelligent than first thought. Nikki teed me up with CEO, Jason Roos, and Product Director, Glen Blow, to catch up on the latest conversational AI happenings at Cirrus.
Cirrus and conversational AI in 2018
Jason was keen to point how that the Cirrus platform is not a technology lead approach. Their major focus is on customer outcomes – and that just so happens to leverage technology on the way.
Back in March 2018, at the Call and Contact Centre Expo, Cirrus took a ballsy but confident approach to conversational AI. The problem they aimed to solve was when contact information is buried layers deep so customers can’t actually contact a business. Jason commented on the state of Contact Us sections:
“Often a negative customer experience is created, but also sales opportunities are missed. Conversational AI, as a rapid release, would have transformational results”
Jason described the demo at the time as dangerous. The offer of spinning up a demo within 30 minutes of receiving a business card was a game changer in chatbots – before we were talking about conversational AI.
Jason goes on to say that the second key point to the Cirrus AI experience surrounded the knowledge base. The offering was unique in 2018 as their chat tool scraped the website and existing resources to gather information. However, information doesn’t always provide context and understanding.
Challenges from 2018 to 2019
Jason told me that Cirrus has found some challenges around rapid deployment and scraping sites for knowledge.
“We found that most websites are built poorly. You could have content in two different places that addresses a question but the language used is subtly different because the context is different. If you allow the knowledge base to use the website, you could end up with two outcomes for one question”.
He goes on to comment on another problem around linguistics. When humans spell words incorrectly or there are alternative meanings to words, chatbots need to be taught this. We cannot assume artificial intelligence is totally artificial or totally intelligent. Just like with humans, bots must be trained to an extent.
Jason used the example of an insurance company where claim is often misspelt. You don’t want your chatbots offering advice on how to make a clam. Understanding that humans make mistakes needs to be plumbed into bots.
Cirrus conversational AI in 2019
From the lessons learned, Cirrus has now launched Cai (Conversational AI). To streamline the customer journey further, Cai now comes provides a chat experience similar to an IVR.
“80% of queries are the same. In the insurance industry, it’s often quote, amend or claim. This gave us a light bulb moment in terms of the approach to delivering conversational AI. We needed to rapidly build vertical based personas. If 80% of the content is going to be the same, the training can be done in advance. All that then needs to happen is tie in brand personality and leave 20% scope for unique customer requirements. A result of this is that we can achieve quicker deployment speeds than ever before”.
