Machine learning will scale personalised customer interactions: Flamingo

Following a successful listing on the Australian Securities Exchange (ASX) on November 17 with a market capitalisation of AU$23 million, conversational commerce company Flamingo is now looking at Australia as a key market, with online lending marketplace DirectMoney being announced as its first local client.

Born in Australia and based in the US, Flamingo is a software-as-a-service (SaaS) company that provides an "intelligent guided selling platform" to help financial services firms address the problem of low online sales conversion rates. Currently, the average online quote-to-sales conversion rate in the US insurance industry sits between 1 and 3 percent, compared to 20 to 60 percent conversion in call centres.

"The reason for this is that buying products online such as financial services is enormously complicated and customers have a large tendency to abandon doing things online and eventually will go to a call centre or to a broker," Flamingo founder and CEO Dr Catriona Wallace said at a media and investor meeting last Wednesday.

"One of our large US clients report that within the first 90 days that a new customer comes on board, 80 percent of those customers will call the call centre four to five times because they don't understand what they've bought. They don't understand the product, they don't understand how to use it, they don't understand the value of it.

"In addition, there's pressure to reduce headcount and get efficiency in call centres."

According to a recent study conducted by Fifth Quadrant and commissioned by Flamingo, 60 percent of the Australian consumers surveyed said they had abandoned an online purchase in the last three months due to prices not being competitive, complicated checkout process or poor shopping cart functionality, complicated website navigation, security concerns, and lack of information about the product or service.

By combining web chat, web forms, and artificial intelligence, Flamingo guides customers through their purchasing decisions, Wallace explained, coupled with a data science capability.

The company claims its edge is a conversational commerce agent called Rosie that is more sophisticated than a chatbot.

"Conversational commerce agents are not generic chatbots, but virtual assistants bound to specific business processes ... Rosie automates this process to sell insurance or onboard customers. She doesn't hold aimless chats with the partially engaged general public," Dr Jack Elliott, chief data scientist at Flamingo, said at the media and investor meeting.

"Chatbots require intense knowledge engineering to deploy. Rosie, on the other hand, starts learning from day one."

Rosie learns from real-time human to human interactions -- between customers and customer service representatives -- and from conversation logs or other historical data that companies have recorded.

"As Rosie observes conversations between consumers and vendors, she learns how customer service representatives solve problems for consumers. We start to see patterns in the language, repetitions in the behaviour emerging from the platform," Elliott said.

"Once we've seen enough, we can start to automate human interactions by responding to queries to human operators with responses from our conversational commerce agent. This is no different to training any other operator. But the new operator being invisibly trained behind the curtain is not human, it's software."

Flamingo believes that human assisted virtual assistance (HAVA) is the best option for successful interactions with customers. HAVA mode is where Rosie largely guides the customer; however, a human can take over from the machine if there is a difficult question that needs to be addressed and is beyond what Rosie is capable of assisting with.

Joe Waller, CTO of Flamingo, said the company's edge over artificial intelligence players such as IBM Watson is its flexibility, speed, and price. An enterprise implementation can take multiple years and multiple million dollars before a business can reap the rewards of its investment. With Flamingo, it's about fifth of the cost, the company claims.

"One of the reasons why [Nationwide Insurance] selected us was because we offer a low barrier to entry. You don't need to configure the machine learning capability. You can get started straight away in human to human mode. The training starts immediately." said Waller.

Rosie can learn within 100 to 200 human to human sessions, though this varies depending on the complexity of the business and its products.

When questioned about Flamingo contributing to the issue of humans losing their jobs to robots, Wallace said many companies are looking to redeploy their staff so they can deliver value via other means rather than completing menial tasks that can be automated.

Matt Henderson, omnichannel program director at Greater Bank, who is investigating the potential use of conversational commerce, added that it's not about forcing everybody into self-service.

"It's about humanising that digital interface by being available when people need to speak to humans," he said at the media and investor meeting. "The reality is it's not an efficient or effective way to operate 24/7 voice enabled contact centres."

"There is a greater expectation for real-time support for their everyday banking needs regardless of the day or time of the week."

Wallace said now that the company has listed on the ASX, Flamingo will be sharpening its focus on Australia and the broader Asia-Pacific market. Flamingo will not always be an enterprise implementation, Wallace said. Towards the end of 2017, the company will go to market with a product more suitable for mid-market companies.

Much further down the track, Flamingo will play in the machine to machine interaction space, when humans rely on bots speaking to each other on their behalf.

At the media and investor meeting, Anthony Nantes, CEO of DirectMoney, an ASX-listed marketplace lender, said in the future bots will organise deals on behalf of consumers.

"I'll wake up one day as a consumer, I'll want a quote on a new kitchen to put into my house, and I'll say to my personal master assistant chatbot to get me some finance for that. My bot, my Siri or whatever it might be, will come and talk to DirectMoney's bot and get me the best personal finance deal to finance that kitchen. That's only a couple of years away," Nantes said.

"My personal bot will know my income, my finance threshold, my history and what I'm looking for in terms of financing my kitchen and will transact with another bot to make that happen."

Flamingo, which is one of two companies on the ASX led by a female CEO and female chair, believes it will be able to ride off the back of increasing enterprise interest in human to machine conversational AI in the next two years.

"Customers are demanding individualised treatment and experience. How do you do that scale without machine learning? It's almost impossible," Wallace said. "Nothing's been really changed in the contact center space for the last 10 to 15 years as far as being able to get to scale beyond human to human interaction. We think now is the perfect time for the coming of machine learning to actually start to scale individualised experiences."

A similar sentiment was communicated by Patrick Malatack, Twilio's VP of Product Management, at a startup conference recently when he said there's a belief among businesses that personalised communication experiences are not scalable, that you can't humanise communications and serve a wide customer base at the same time.

"That's no longer the case. The same technology that created a lot of this inhumanity for us is now actually enabling us to build more human experiences," he said at StartCon 2016.

Conversational AI company MindMeld launched its Deep-Domain Conversational AI platform in November, claiming it to be the next stage in the evolution of conversational AI. The platform makes it possible for companies to create voice and chat assistants that can demonstrate knowledge and expertise around any custom content domain.

The platform offers capabilities such as broad vocabulary natural language understanding, question answering across any knowledge graph, dialogue management and dialogue state tracking, and large scale training data generation and management, with cloud-based or on-premises deployment.

In September, LinkedIn launched new bots to help people make the connections and build the skills they need to advance their careers.