Q&A: Today’s Chatbots are Still “AI Infants” That Rely on Committed Parents

Tae Moon, Senior Manager of Digital Banking at TD Bank, chats with us about the corporate challenges of "infant" chatbots and their position on the path to adolescence.

Tae Moon chatbots

Vendor promises about the scale of chatbot capabilities should be taken with a pinch of salt, because, according to one buy-side expert, these technologies are still in their infancy, and far from the industry application level.

Tae Moon is a Senior Manager of Digital Banking at TD Bank, focused on artificial intelligence (AI) and chatbots. He oversees and manages the end-to-end development and deployment of chatbot service, so has developed a keen eye for what is working and what is still lagging behind.

We sat down with Moon to discuss corporate challenges with these chatbot infants, and their position on the path to adolescence.

Disclaimer: The opinions and views expressed below are solely those of Tae Moon and do not represent TD Bank or its affiliates.

Nearshore Americas: In today’s customer experience environment, is chatbot support something that large corporations can rely on, or is there still a lot of work to be done?

Tae MoonTae Moon: If I had to make an analogy, I’d say that we are currently where we were with internet during Y2K, when everybody is excited or on some kind of dot com, yet people don’t fully understand why or what the power is behind it.

The infrastructure is there and it’s being adopted, with lots of financial institutes investing into the chatbot space. However, for me, the next “internet” is the AI wave – chatbots just happen to be getting the spotlight because the so-called machine learning in the back-end is not seen by the end customer, and they really don’t care about it.

Chatbots are the only AI solutions or applications that the customers are able to engage with, so in essence it is the next wave. Maturity wise, it’s still early, but if you don’t get your feet wet now, you’re going to have a lot of catching up to do.

Nearshore Americas: What does “getting your feet wet” involve at this stage in the maturity of chatbots?

Tae Moon: I would say, just start off with a POC (proof of concept), try it out and find out what the problems are. Only by doing that will you realize what you want, what you need to do, and what vendor to go with.

Right now, going back to the example of the internet, people think that if they put their companies on the web they will magically get more exposure or sales, but that’s not the case. AI is not a silver bullet – you can’t just adopt or implement AI into your company and hope that it solves whatever problem you’re having.

There is a significant amount of adjustment and growing pains – it’s an AI infant. You’re going to have to train it, you’ll realize that whatever AI solution vendor that you go with is not optimized for your process, and there’s a chance that you’ll have to adopt another AI infant that does fit your needs.

Nearshore Americas: Chatbots are on the right path to continued evolution, but what are the most common problems you’re still noticing with them?

Tae Moon: Every corporation is different and has unique use cases that will not be covered by the out-of-the-box solution – there will definitely be some customization required.

Secondly, we need to talk about the NLP portion (natural language processing), which has definitely come a long way. Many vendors I have assessed are presenting very good NLP products, capable of interpreting various utterances and variations of speech. But it’s the generation and response half that has a long way to go.

I have not yet met a single vendor that was able to form free-text speech, meaning you take elements of a sentence – verb, subject, or noun – and actually create a sentence. As an example, the phrase “my name is…” is a static, canned response that the bot will always retrieve. The “blank” that comes next is how vendors are claiming to be dynamic at this point, which is not true NLG (natural language generation), but just a “fill in the banks” response.

I have seen this working in academia and within a controlled environment, such as with Google and IBM Watson, but at an industry application level it is not there yet. If they were to get there, we’d be dealing with true, independent machine learning, which is akin to teaching a child – they need to know not to use profanity, to use the right grammar structure, and how to answer questions, so there is a whole suite of training that will need to be provided if and when we get to that point.

 

Speaking at Sourcing Decisions 2018, Tae Moon was part of a panel of clients that discussed what was holding them back from chatbot and RPA adoption. Check out the video here.

Nearshore Americas: Is this good news for some industries, but bad for others? How might financial services be impacted?

Sign up for our Nearshore Americas newsletter:

Tae Moon: Banks are highly regulated, so every word they say can make a difference to how a customer moves or spends their money. For now, banks are comfortable with going for controlled, , pre-scripted responses, but when the free-form speech comes along, there will likely be a whole other governing body that oversees the language that is generated.

This will have to be a collaborative effort between the banks and the vendors, as it’s like taking a kid fresh out of college and teaching them what they can and can’t say within the boundaries of company policies.

Nearshore Americas: Are customers convinced that chatbots are good enough to meet their customer service needs, or do they not even care?

Tae Moon: From a financial services perspective, customers don’t find banking and finances “fun” per se. It’s a very goal-focused interaction, such as paying a bill, transferring money, or checking a balance, so for customers it’s not about how smart a chatbot is, as long as it can help them achieve those goals in the most frictionless, convenient way.

Customers don’t care that the bot is a true AI and can create sentences from structures. Even if it’s a pre-scripted response, if the bot can help pay a credit card bill with two text messages, they will be satisfied.

Once customers get used to it, and the chatbots have their trust, then banks and financial institutes could start using bots for higher value tasks like cash flow analysis, or spending pattern projections.

Nearshore Americas: Improved customer experience and improved efficiency are the main selling points for chatbots. Are they being fulfilled, or are they false promises?

Tae Moon: There are some immediate savings, for sure, in volume reductions, and operational costs from the phone channel point of view. But, the real value will come when the trust is built, and customers start treating the chatbots as their “Jarvis” or “Alfred”. Once they get to that stage, they can be scaled at a fraction of the cost of any human private banker.


More on chatbots:

Tags

JOIN THE CONVERSATION

JOIN THE CONVERSATION