The worry is that AIs will replace humans, but the unexpected sometimes happens. Meet an AI that is increasing employment for humans.
Can we trust artificial intelligence now that some AIs are getting so sophisticated it’s beyond human capacity to understand them? Can we rely on AI programmers to be responsible enough to create AIs that won’t accidentally do something horrible? Will AIs take over not just some jobs, but most of them? There are a lot of stories to be told about the many valid reasons to fear AI – but this is not one of them. This is a story about AIs doing such a kick-ass great job expanding employment for humans that some of their meat-based co-workers have given their virtual colleagues their own honorary cubicles.
The term “artificial intelligence” encompasses so very many different things. When we’ve focused on AI lately, we’ve often been talking about a chip that can learn with a minimum of data input, or the systems that enable an autonomous vehicle to consistently identify obstacles, or a program that can analyze impossibly vast amounts of data in real-time to discern subtle patterns that might reveal a financial trading opportunity or detect the activity of a malicious hacker. That’s not what we’re talking about here.
This is a story about the type of AI that hasn’t got much press lately, the kind of AI that can interact with humans as if they were human themselves. These AIs can be plenty sophisticated, but the technology they rely on is hardly new, let alone bleeding edge.
Conversica’s flagship is an AI sales assistant that corresponds with humans via email or text. Think of Siri or Alexa, but if you were corresponding with them instead of speaking to them.
These assistants are designed to initiate contact with people who have been identified as leads, evaluate the responses, and continue to interact until the lead either severs contact or indicates a desire for a more complex interaction that requires a human to complete.
Conversica’s corporate antecedent was a company that helped commercial clients with lead generation. The clients all shared a common shortcoming: none of them were very good at pursuing the leads that were generated for them.
“Our research showed that with most companies there is no follow-up on 30 to 40 percent of inbound leads,” said Conversica CEO Alex Terry. “This is someone saying ‘Hey, cable company, I want to subscribe,’ or ‘Hey, I want to buy a time-share’ – or whatever. That’s 30 to 40 percent with zero follow-up. With another 30 to 40 percent, there’s follow up only one or two times. And this is when people are asking for information.”
Companies were leaving money on the table. But what if there was a way to follow up on those leads? Well, there could be. Someone could develop AI assistants that could make initial contact and qualify leads.
Since no one else seemed to be doing it, however, the company decided to do it itself. AI assistants soon became its main business, and the company renamed itself Conversica.
Over the last three or four years the company has created five different AI assistants (there are more in development), trained for different applications, and fluent in a half-dozen languages. The idea is to have them do entry-level stuff, for example making the three, four, or five contacts necessary just to qualify a lead. Clients set up the AIs, giving them not just names but often also titles.
Which really sounds like replacing humans. So much so that even Conversica thinks it sounds like replacing humans.
So Conversica got into the habit of polling every single customer after four months. The vast majority that adopt Conversica AIs report that they had to hire people to handle the additional sales that result from the AIs qualifying more leads – the ratio of those that hire more people to those who don’t is 5:1, Terry said. Conversica AIs are being set to do the drudge work that people hate doing and which consequently rarely gets done. Some of Conversica’a AIs have even been voted Employee of the Month by their human compatriots.
Conversica partners with a number of companies that sell customer relations management (CRM) systems. Terry said that one of those (which he declined to name) independently verified Conversica’s results. Why would the CRM company go through the trouble? It sells software seats, and if Conversica’s AIs were replacing humans, the CRM company would end up losing seat licenses. The opposite is the case, Terry said, and this particular CRM company remains a business partner.
Conversica now integrates with at least 50 different CRMs and marketing automation platforms, Terry said, with more added every month.
Becoming a Gre(AI)t Employee…
The AI software is capable of natural language processing. Conversica assistants are trained, and their training is updated as the company logs additional transactions, which number in the millions. The training includes not only how to evaluate natural language and how to choose the appropriate responses, but also learning new languages (the company is currently working on Portuguese).
The AIs are designed to extract intent (“Okay, I’m interested”) and entities (“My name is William, but call me Bill,” or “please call me on my mobile instead”).
The decision engine is a distinct operational element. That enables each customer to set their version of a Conversica AI assistant to respond to specific requests in the way each wants. For instance, Terry explained, Company A might be happy to give a lead the price of a product or service in an email; Company B might decide that under no circumstances will they ever give a potential customer a price until that customer steps foot in a showroom. Or Company A might change its policy – that’s a simple update.
While the training is extensive and ongoing, the technology necessary to implement Conversica’s AIs is widely available. Conversica’s assistants are mounted in the cloud, currently on Amazon’s AWS platform.
No GPUs For You
“We don’t have to use graphical processing units,” Terry said. “We don’t need 100 millisecond response times. Calculating a response in less than a few seconds – that’s just fine. This isn’t an onboard system in a car where a lag means hitting a stop sign or a pedestrian. Sending an email in minutes is okay. Even with a real-time chat, that’s okay. The kind of latency you need to support this kind of conversational AI is supported by standard web services. Frankly, that means it’s cheaper for us to solve our customers’ problems.”
What comes next? What Terry referred to as a "customer success assistant." Another place where sales organizations tend to lag is in following up with customers, often to gauge satisfaction with goods or services provided, but also to discern if there are cross-sell or up-sell opportunities.
“It could be for anything,” Terry said. “’Hey it’s time for your monthly business review – do you want to schedule it now?’ Or, ‘Hey, we’re going to be at a trade show near you. You want to schedule a meeting?’”
Another new product is customer care, which Terry said is currently being trialed by a handful of customers. “’Hey, your bill is due. What’s going on?’ Maybe the customer needs more time. Or maybe it’s ‘Hey, your credit card has lapsed – can we update that for you?’” Handling those types of calls takes different skills, Terry explained, but the conversational platform is the same.
“We’re using technology to help people be more successful,” Terry continued. “It’s better than a human alone, and better than an AI alone.”