Ever since Facebook made its big bot announcement in 2016, conversational AI (or chatbots) are all of a sudden everywhere — along with their promise to enable a smoother and more conversational way for a consumer to interact with a business.
Emphasis on the word “promise.”
But as most consumers who’ve had a bot experience in the last year can attest, the real-world experience doesn’t always, or even often, live up to that promise.
The conversational experience — billed as the faster, smoother and more seamless way to interact with a firm in digital space — often turns out to be a bit stilted, confusing and far from frictionless. Often, consumers report that they are working harder to use a bot than they would if they used an app or even a mobile-optimized website.
Other than the bragging rights that come from interacting with a bot, the bot experience often doesn’t add much to the consumer’s journey.
So what would?
That was Karen Webster’s first question for Zor Gorelov, CEO and co-founder of Kasisto. Gorelov’s been in the software platform/conversational interface space for nearly two decades, creating these conversational interfaces before the term for them was “chatbot.” His latest company, Kasisto, created a conversational AI platform that’s been powering “intelligent conversational interactions between financial services companies and their consumers” for years.
In response to what might make a great bot experience, Gorelov noted, “The real question is why are there so many bad bots — and by bad, I mean low IQ bots. If there are 100,000 Facebook bots out there today, I would say that over 90 percent of them are low IQ or dumb bots.”
As for why, it’s simple, he said: dumb bots are a lot easier to build than a truly smart multi-functional bot. With a thin layer of natural language understanding (NLU) software wrapped around a few pre-programmed responses, a functioning, yet not entirely helpful, chatbot can be built quickly.
A smart bot, on the other hand, does more than understand a few phrases and a couple of tasks — and requires much more specialized and sophisticated AI platforms than simple bot frameworks.
“High IQ bots are the ones that have deep knowledge [of] the company and its business … they also understand consumers and use that knowledge to answer questions and engage consumers in a way that actually helps them get something done conversationally and contextually.”
That is a much taller order — and requires thinking somewhat differently about real-time chat as a platform for business, which, in Kasisto’s case, is the banking business. Kasisto’s conversational AI platform powers omnichannel virtual assistants and bots on mobile apps, messaging platforms, web and IoT devices with deep domain expertise about finance.
That means KAI Banking is fluent in finances — understanding thousands of banking intents, all of the actions customers want to take. KAI Banking was trained on millions of banking sentences, all of the ways people phrase their requests. Without that deep domain expertise, the usefulness of a bot is very shallow and falls apart pretty quickly. Bots for banking and finance require something more than a simple bot framework trained on a few use cases and FAQ content. To hold truly human-like intelligent conversations, Kasisto’s bots are able to fulfill requests, solve problems and predict needs.
Upping the IQ of Bots
The basic difference between a smart bot and a dumb bot is depth, according to Gorelov. Low IQ bots are defined by their shallow knowledge and limited skill set. A smart bot is able to do more and infer context, which makes it capable of anticipating what a user might want it to do.
“As an example, say someone wants to make a credit card payment — so, with a KAI-powered bot, that is how a consumer can start the conversation. They say ‘I want to pay my credit card.’ Kai then says, for example, ‘You need $100 to pay what’s due, or you can pay off the whole balance with $1,000.’”
Gorelov says that’s just the beginning. Kai can then keep the conversation moving when the consumer might respond, “Do I have $1,000 to pay it off?” Kai can answer “yes” or tell the user that they don’t have the funds in their checking account — but that it can move the funds from savings to checking to pay off the balance.
That interaction is more complex and requires the bot to offer contextual recommendations based on other information that it has available to it — totally, attainable goals for AI, Gorelov says. Artificial Intelligence has more raw computing power at its disposal than ever before and more data to work with — the power and the potential richness of those data sources can be seen by taking a peek inside the backend of the bot.
“Often the bot that a consumer interacts with is really the top of the iceberg that is floating in the sea. Whether it succeeds depends on how deep the part is below the surface that consumers don’t see — that is where the magic happens including advanced machine learning, content creation, classifier building and knowledge composition. If the bot publisher doesn’t keep improving that underside, the top part melts pretty quickly and results in subpar user experience.”
Building for the Bot Future
Despite the predictions from some of the biggest names in tech that the bot is going to displace the app, Gorelov and Webster agreed that it seems doubtful.
“Bots aren’t the new apps,” he noted, “because consumers will go to the place where they can best get something done, and that won’t always be a bot.”
Bots can be really useful, however, in creating a user experience that “wasn’t possible or even imaginable before,” that actually trades on the power of using natural human language.
“Today you wouldn’t call your bank to ask a customer service representative, ‘So how am I doing with my finances?’ because they would hang up on you or be very confused,” Gorelov said. “Using a virtual assistant is rooted in giving people a way to do things they couldn’t or wouldn’t before that is tied to the power of human language.”
Done right, it is a powerful tool — but doing it right is hard, and lots of banks did it fast instead of doing it right, Gorelov remarked.
“Many banks rushed into it and created some interesting experiences and now realize that creating, maintaining and growing meaningful experiences is really hard,” Gorelov noted.
Banks want to unleash their customers’ data to create more contextual, personal and predictive experiences that do more to help their customers manage their money. But, Gorelov said, that financial data needs to be tracked, anonymized and analyzed. Further, he explained that banks know how to manage their transactional systems, but in a world powered by AI and data, they struggle.
That means creating a banking experience that is useful in more than one context, since, Gorelov said, not all environments and channels where consumers might like to interact with their bank are created equally — they are situational. A banking bot that talks to customers on Facebook Messenger needs to behave differently than one that talks to them inside their mobile banking app, which is also different from one meant to interact via a voice-activated platform in a car or one intended to interact with a connected device like Alexa.
“When we first moved KAI Banking to a messaging platform from the mobile app platform, it didn’t work at all,” Gorelov recounted. “We learned that mobile experiences and messaging experiences are … [two] totally different thing[s] with totally different needs. Believe it or not, we moved away from having engineers creating graphical UIs for iPhones or Android devices to hiring writers who could help us create conversational experiences on the messaging platform.”
People text, type and talk very differently — and do all of those things in different contexts, he noted. So that meant KAI bots had to learn the shortcuts and casual ways people typically text. The conversational experience you want to create across all channels should be as natural as texting a friend — a bot needs to track idiosyncratic phrases, rapid topic changes and interruptions. KAI bots and assistants can extract the meaning and intent of a conversation to stay focused on the customer’s goal. There is a “reasoning” element to Kasisto’s AI that makes the conversation very human-like.
“We are betting on the very wide adoption of omnichannel, multimodal conversational experiences, and the way the market has been moving, that was a correct bet so far.”
But as bots become more prevalent, Gorelov believes they will also become more invisible. Consumers won’t think about interacting with chatbots, because smart bots will interact seamlessly enough that consumers will be able to fully focus on the subject of their interaction instead of the mode they use to do it.
Given the state of bots today, it’s going to take a generation of high IQ bots to fulfill that vision. A mission that Kasisto is more than happy to rise to the challenge to deliver — one very smart bot at a time – as evidenced by their work with Mastercard and DBS digibank.