Evolved AI and Live Escalation Inquiries

Can AI really satisfy 95% of customer inquiries?
16 August 2022

TD Garden, home of the Celtics – has AI finally got itself a ball game?

Everyone’s talking about AI. In particular, everybody’s talking about how AI, as part of a solid digital transformation plan, can – or will be able to – replace the tedious types of human work, like repetitively answering the same questions thousands of times a day, or dealing with intense, repetitive, in-depth ticket or venue questions ahead of big games, or everyday spectacular events on which a tourist venue depends for its livelihood.

Except we the humans know what AI has previously been capable of, and so we take the potential of the technology with an in-built pinch of salt. We’ve all run into chatbots when trying to get tickets booked, or venue information, or phones upgraded, etc, etc – you add the user-case, we’ve been disappointed in it by a so-called state-of-art AI chatbot.

It’s always wise to treat extravagant claims with a degree of skepticism. So when a company claims it has an evolved AI, underpinned by a sophisticated proprietary Natural Language Processing unit, and a new way of delivering hypergranular information solutions, with AI able to satisfactorily answer 95% of human questions, you have to sit down and talk with them.

We sat down to talk with Don White, CEO of Satisfi Labs, which made the claim, to talk about the power of evolved AI and in particular, the hypergranularity of information that AI can deliver in a ‘live’ scenario.

What It’s All About

First, some background. Satisfi’s system is based on an answer engine that works on data from hundreds of existing clients – whenever a new client uses Satisfi’s system, it helps educate the answer engine. Then there are a host of (mostly pre-built) AI assistants, each dealing with a particular type of enquiry – ticketing, venue information, local area information, etc. And then there’s the NLP, which helps the AI assistants talk naturally to the humans who ask them questions through any number of methods – messengers, websites, etc.

As a set-up, there’s no getting away from the fact that it’s pretty slick.

And Satisfi has an impressive roster of big name (and big challenge) clients on its books to bolster its claims. TD Garden (home to both the Bruins and the Celtics), Oakland Zoo, LA Zoo, the Tampa Bay Buccaneers, and the long-running musical theater show Wicked, to name just a few, all use the company’s evolved AI solution to manage aspects of their interaction with visitors or potential visitors.

We caught up with Don White in the wake of the launch of a new element in the system, called Bridge, which amounts to a roster of human monitors who can step in whenever the AI meets a human whose questions it can’t satisfactorily handle – down to an average of just 5% of the time.

THQ: Your conversation AI assistants – these are not the chatbots we’re looking for, are they?

Don White: The thing about chatbots is that they started off with high keyword, high flow, and that was a way the initial vendors found to offer a product with a low investment in technology. And it was good enough – or at least the market thought it was good enough, because back then, the market didn’t know anything.

We’ve always sat on the virtual assistant side of it, and our AI assistants are more like a Siri or an Alexa. Our AI assistants were designed to understand hypergranular combinations, which standard chatbots can’t give you. The AI element and the constant training of the answer engines mean they’re ready for pretty much anything. Our NLP is a minimum of five layers deep on any subject. So if you think about intent, our intents aren’t like intent entities, the intent side could be four layers deep.

The Deep Dive

Don White: Just imagine that it’s not good enough for the assistant to understand you want to buy a ticket. Maybe you want to buy a ticket for your son who uses a wheelchair, so you need accessibility information tied into your ticket purchase. Maybe there has to be easy access from the subway to the gig. Maybe you need to be sure of the routes to the bathrooms, in addition to the access from the subway, and the accessibility for wheelchairs, and… so on.

That’s where hypergranularity and a multi-layered NLP gets you – the ability to answer more and more complex questions, either in a single query or a set of queries, to deliver a better, more rewarding customer experience than you’d get from any old-style chatbot, and quite a lot of the time, a more rewarding experience than you might get from a time-pressured human operator, too.

So, the short answer is no, these are not the chatbots you’re looking for. And the rejoinder would be why are you looking for chatbots in an evolved AI world?

Who’s Using It – And Why?

THQ: What’s the business case for using evolved AI assistants?

Don White: Most of the time, it’s organizations in the leisure, venue, and tourism sectors that process huge numbers of people, and even larger numbers of queries from those people. To staff that sort of response takes lots of salaried, highly trained, people. Or it takes an evolved AI solution that can deal with 95% of queries effectively and deliver that kind of granularity that allows people to feel safe and satisfied with their interaction.

We probably over-engineered the system at the start, but that’s paying dividends now, because all the chatbot people are trying to scale up to evolved AI to deal with complex, 21st century questions, and we’re already here. But we probably suffered early on because our system was too complicated for the needs of companies.

THQ: Do they ask “What do we need this for?”

Don White: Sometimes. I say “Well, this thing understands 50,000 unique queries and has over 1000 intents.” And they say “But… we only have 10 FAQs on our website…”

I say “Well, how do you know? You don’t know.” And they go “We know we don’t have 10,000!” But these days, people care about data. If you don’t have hyper intent tracking, you actually don’t have great data. Again, it comes down to that thing where a chatbot can sell you an available ticket. A conversational AI assistant can sell you the precise seat you want, taking all your many and varied requirements into account, just like a human could, but probably faster.

Getting Out of the Chatbot Mindset

THQ: Is there a sense in which the availability of this level of intent and this granularity and complexity of questioning is something the customers have to be trained in, having grown used to the limited functions of a standard chatbot?

Don White: Quick idea that’ll answer your question. We used to have quick replies that were one word, say ‘tickets.’ Now, our front end is designed for questions, and they can be simple or complex, long or short. We’re trying to train the user into the idea that you can go deeper, and you can expect more, because with an evolved AI with a sophisticated NLP, we can deliver more. As soon as people realize that there’s a deeper layer, they’ll do it.

He’s right – they will. And of course, given that the company keeps adding use cases to its portfolio, it’s clear that they are. As the world evolves to expect more from its technology, it’s these kinds of ‘probably over-engineered’ solutions that will take the public understanding and expectation of what you can achieve with AI forward.

A Measured Backlash

There’s a factual backlash against science fictional interpretations of AI in the informed tech world right now, but that doesn’t stop us doing some pretty fascinating and life-improving things with the technology.

The Satisfi AI assistants are accessed via a simple web snippet, so venues can create as many entry points to the system as they want – from websites to WhatsApp – so it’s an immensely powerful AI information system, accessed through a tiny piece of web coding, with minimal technological investment by clients.

While White is skeptical himself over whether AIs can get past the human factor in some applications – ‘for purchases over a certain dollar amount, you probably want to see it a store and touch it’ – with the right use case, he’s confident that AI can get up to answering 99% of inquiries, even if having humans that can step in is always a useful layer to deliver customer satisfaction, rather than abandoning the inquiry.

The Next Ten Years

He sees opportunities in the next ten years to add voice function in remote work set-ups, to add conversational AI assistants to technology like connected cars – and then to take it into the Metaverse.

What is clear is that while people are generally right to be skeptical about the out-there claims sometimes made about the likes of AI technology, with the right application, and a set-up that’s ‘probably over-engineered,’ tech companies can deliver impressive results with it today – and already have their sights set on tomorrow.