Updated: February 19, 2020 4:33:34 pm
As many of us have discovered, interacting with chatbots, especially when trying to solve a business-related query, can be a trying experience. This is often because the technology is still unable to understand the ‘intent’ of the user.
PV Kannan, co-founder of 7.ai and co-author of the book The Age of Intent: Using Artificial Intelligence to Deliver a Superior Customer Experience, believes understanding ‘intent’ is what is crucial to determining success for a virtual agent deployed by a business to solve customer-related problems.
The book, co-written with former SVP of Forrester Research Josh Bernoff, includes examples of how some businesses have gone to successfully implement chatbot systems when it comes to interacting with customers, and where others have failed. The book makes the argument that instead of just creating a simple chatbot that follows a given trajectory for set questions and answers, virtual agents will need to solve more complex tasks in order to serve customers better.
Indianexpress.com spoke to the author over a video call to understand why intent matters for chatbots. Below are edited excerpts of the interaction.
Best of Express Premium
Why is intent so crucial?
Think about what a call center agent does in the first few minutes of any conversation. They authenticate you and then they try and understand what you are trying to do. For example, I want to pay my child’s school fee, so I need to transfer some money. Now, the human agent is converting those words in his mind and saying, “Oh, you know, guess he wants to make a balance transfer.”
We don’t consciously think about the number of steps that it takes for the agent to understand intent. But when you translate it to a machine and the machine has to understand what a human being is saying, if you don’t understand the link, it doesn’t matter what other AI you have, whatever natural language processing you have, you are solving the wrong problem. Intent has to be understood, and only then you are solving the right problem for the customer. If you don’t get it right, then you are just doing something else and frustrating the customer.
What are the challenges when creating virtual agents for a certain company? In the book you also mentioned how a simple path-driven approach does not necessarily work when creating these.
So there are two types of bots. One is what I call a simple bot, where you know you use it to order pizza, where the conversation path is fairly predictable without much complexity to it. Because there’s only so many things you can have in that conversation for that given intent.
Where the complexity comes is when you take a brand such as a bank. Today, most virtual agent designers seem to obsess more about the dialogue and the flows, and try to make it very prescriptive. This is not how real human beings converse. So what ends up happening is that most of these bots end up not getting used.
The focus should be on the intent. Once you solve the intent, the flows for that intent actually become very simple, right? For example, if I can figure out what you’re trying to do is transfer balance in a bank, then I can quickly get on. And I will tell you why figuring that out is hard.
Most customers don’t converse the way bot designers think. They think that ‘hey, I want to do some balance transfer’ is how a customer will talk. But customers will say something like ‘hey, I need to pay my credit card bill tomorrow, so I need to move some money from my savings account to checking account’. The word balance or transfer are never talked about in this.
Express Tech is now on Telegram. Click here to join our channel (@expresstechie) and stay updated with the latest tech news
And then most of these bots, they’ll flip out.
But that’s how humans express what they’re going to do. If you know how real conversations take place, and if you have the ability to map those sentences and utterances and understand them deeply, and combine them with what we call backend insignals, then you can solve the intent. Today, bot designers have not given any thought on how to integrate backend signals.
Complex conversations are not easy to figure out. You need a more sophisticated approach to doing it. That’s why the book starts with saying intent is actually the building block for building great virtual assistants.
You’ve also given examples of a hybrid future where both humans and virtual agents are working together. Wouldn’t the argument be seen as a little too simplistic, that it doesn’t often work out that way, especially as agents they get better at figuring out intent? The fear that humans would be pushed out once AI comes up is still there.
When the web came out and mobile apps came out, people stopped calling the different helpline numbers for brands. They started relying on them to do simple things like checking your balance, etc. That’s one of the reasons why I say these simplistic bots are a waste of time because people already know how to do that. You don’t need to go to a bot to ask what your account balance is.
What has happened in the last 15-20 years is as the web and mobile web has become more powerful, the queries that get escalated to a call center have become fairly complex. If the customer could figure out how to do it themselves, they wouldn’t be calling. So that’s why these bots have to be super smart.
Granted, as these bots get smarter and smarter and as AI gets more powerful, a bunch of conversation will go away. For brands, the question is how do I use humans in ways where there is a bigger touch point, that doesn’t exist today, and one which is a lot more productive and thoughtful. Today brands just don’t do that because their calls are already busy, they don’t have the time to tackle it. I do think these virtual agents free up money and time to invest in customers, and companies who fail to do that are essentially just a company run by robots at that point.
You’ve also compared the chat box with like the popular virtual systems that a lot of people might be familiar with which is Alexa, Facebook Messenger’s chatbots. But you’ve also pointed out that you know the inadequacies. Could you elaborate?
In the book, I talk about general purpose, virtual assistants like Alexa and Siri and other things where you know you can do so many things. Today if you look at Alexa’s actual usage, 85 per cent is for playing music. The rest is asking for the weather, and doing some transactions. The reason is there isn’t much to do and most of the Alexa devices don’t even have a screen.
So even if you say ‘hey I’m thinking about buying the brand new Bose headset’, you can’t always see a picture of it. That’s why the functionalities are fairly limited. Having said that, I think, as these devices get attached to the screen and they become cheaper, you could do more interesting things with them.
For customer service and specifically for brands to enable customer service to Alexa, it’s a very cumbersome process to link a brand login with Alexa. So until they solve that problem.
And as you pointed out, there are so many chatbots and no one uses them. For many of these other chatbots, you can only ask stupid questions, which anyway you can figure it out on the web yourself. That’s why they don’t get used. So until these devices can actually recognise and authenticate you, the use cases are fairly limited.
But I do see them solving the authentication problem soon. These companies are not going to let go of the fact that these devices are in every room. They will solve the problem.
For organisations how important are these virtual agents and chatbots?
A good example in this is your healthcare provider. Let’s say you are a customer of Apollo hospital. These (chatbots) are really useful if there is a lot of frequency of using the website, the app, and there is a monthly transactional relationship with the brand. What we see is a massive amount of interest in the banking, insurance segment, and a lot of interest in the telco segment. There is almost a daily relationship with these service providers.
Travel and e-commerce is a big area where there is a lot of interest. Pretty much all the top players are trying to figure this out. I mean, not all the efforts have come out well. In the book I talked about some of the not-so great efforts like Expedia, compared to what the very thoughtfully done efforts of Dish and Hilton and others. Without a doubt, these industries are all adopting these because their competition is doing it.
There’s also the question of privacy with chatbots, and how crucial security will be, because these virtual agents can access so much information.
At a very high level, the chatbot cannot access any information unless there is an authenticated customer talking to it. So there’s no way it can go and randomly start accessing information. It’s really similar to how the current call centres with humans are set up, where the agent is not allowed to go search and someone’s information.
There has to be a call connected, even if the call gets hung up for whatever reason signal issue or whatever, the access to the agent is cut off. That’s the same layer we are exploiting. As long as a live customer at the other end, we can access this information.
Companies do need to rely on all the data to build the chatbots, but not none of them includes your personal data. The only data that you study at a macro level for a given intent, how does the customer go about asking questions, what does the agent do. None of your personal data is needed for that. So in fact when we train our models we strip all that data out, because it is not needed.
How easy is it to train these specialised models for companies?
It takes effort. It’s not an impossible thing, it’s just that you have to systematically go about solving and tackling the problem as opposed to saying hey bot is the cool thing.
Everybody in the world built a bot on Facebook Messenger, and there are like 30,000 bots, but all of them are useless. For companies, it is about building a bot that is actually going to be capable of handling, most of the stuff that the customer wants to do that is fairly complex. If they put their mind to it, and then do it in a disciplined fashion. Otherwise it is a big buzzword, just a cool shiny toy that we want to tell the world, I’ve also done a bot and it’s gonna be a failed project.
🗞 Subscribe Now: Get Express Premium to access our in-depth reporting, explainers and opinions 🗞️
- The Indian Express website has been rated GREEN for its credibility and trustworthiness by Newsguard, a global service that rates news sources for their journalistic standards.