Episode #14: The Complete Guide to AI-Powered Intent Analysis

If you want to understand your customers, you need to understand their intentions. Modern channels can help. But how do you possibly sift through and analyze the billions of messages sent every day? Fortunately, help is here. In today’s episode we look at how AI can help you understand customer intent, and improve their overall experience.

All podcast episodes

Go Now

PODCAST TRANSCRIPT

It’s the CXM Experience. And as usual, I’m Grad Conn, CXO at Sprinklr. And today we’re going to kick off a relatively long series of discussions around AI. Let me tell you a little bit why AI is important. I will demystify AI a little bit over the next couple of weeks, and we’re gonna, we’re gonna dig into all sorts of different features and stuff. So it’s gonna be super fun. But I want to talk about AI at a high level. And then I want to specifically focus on one aspect of how AI is used inside Sprinklr for something called intents. And it’s not intense, its intents… as in I-N-T-E-N-T-S. As in what do you intend to do? What was your intent? And so the identifying intents, and bringing them to life has a very significant number of applied use cases, particularly in Customer Care. Not only but particularly in Customer Care. We’ll probably double click on that a little bit more today.

So why AI? So you’ve heard me talk if you’ve been listening, about listen, learn and love. This is what Sprinklr is all about, which is listen to what people are saying. Bring in the billions of conversations that are out there. They’re unstructured, they’re unsolicited. Unsolicited is good. Unsolicited is good, because it’s the truth. Unstructured is hard. Unstructured is hard because you have a mixture of emotions, a mixture of brands, a mixture of ideas in a single post. More complex, harder to parse.

The way most companies are dealing with it today still is they ignore it. Because it’s hard. Instead, they do surveys, or focus groups. Oh my god, I can’t believe people still doing focus groups. But there’s some people are just doing surveys because they’re easier. They’re structured data, they can put them in the CRM system, their relational database doesn’t cack on it. So easy, right? And so silly. If anyone’s learned anything from the US election, surveys don’t work. Polls are garbage. Solicited feedback is almost always untrue. You want unsolicited feedback, that’s what you’ve got to get. So that’s what listening is all about.

Learn is what we’re going to spend a lot of time on today. Because the problem, the great thing about pulling in millions of conversations is you’ve got millions of conversations. What people really think. The so called voice of the customer, but it’s true voice of the customer because it’s unsolicited. The problem is how do you read a million conversations that cover all the gamut, and cover emotions and brands. So I’m going to spend a lot of time on that because the only way to do it, the only way to do it is with AI. And Sprinklr has one of the most sophisticated AI platforms in the world. It’s been developing it for years, it’s one of our biggest investments as a company. We are using a massive database to train it and a massive user base to do day to day feedback on it. So we have an incredible platform. And I’ll talk a little bit about how that kind of comes to life.

And then of course, the love part comes from once you’ve discerned what people want and what they’re doing, you can actually do the right thing for them. So let me talk a little bit about intent. So let me talk about it first, in a customer care context. I’m going to talk a little bit about some of the challenges faced by care teams today. By the way, if you hear a little bit of a barking sound in the background, that’s my dog having a nightmare. So her name is Hester, very cute dog. Anyway, so challenges faced by care teams today: more than 50% of customer calls go unresolved or require some kind of escalation. 52% of customers hang up on a customer service call before their issue is actually resolved. And 32% of people expect a response to them 30 minutes, and 50% of people expect a response in an hour.

Customer Care is really not cutting it on a lot of the key metrics that people expect. See what you really want to be able to do to fix that is you’ve got to identify the top customer intents driving the volume of calls and inquiries. And then be able to enable agents with AI-based responses that are based on those intents. And if you can understand and process customer intent, you can actually speed your customer response by over 99%. So you can reduce your resolution time from say 10 minutes or so to seconds for most inquiries. So it’s really some one of the most fundamental things which is if you understand the intent of a message, and you can discern that with machine learning, then you can very quickly get back to someone with something that helps them solve their problem.

You know, AI is a big deal these days. 57% of businesses expect AI to enhance the customer experience. That’s from Forrester. And there’s actually a 91% cost reduction, which can be achieved by replacing human agents with virtual agents. That’s an IBM study. You know, Intelligent Automation can reduce response times by 80%, KPMG found that. And EY says that there’s a 20X reduction in subsequent resource requirements when AI is deployed in care. 80% of all executives say that AI boosts productivity. And I’m surprised that’s not 100%. But certainly, most people realize and recognize that AI is the key to driving the future.

So let me talk about what are intents. So using AI Sprinklr intuition can automatically classify messages to help brands better understand the customer’s intention. For example, someone might say something like, I bought this last month, recently, it stopped working and there’s a red light, which keeps on blinking. Where can I get a new one? That’s kind of a classic post, right? That’s like, the kind of thing people say all the time. It’s very difficult to kind of pull that out, unless you’ve got a really strong AI engine. Because the AI engine can read that and say, ah, someone needs a store locator. And they have a device malfunction. They parse out that and that is the intent of the message, even though the words are sort of sloppy. So intents, analyzes the messages, and identifies whether it’s an opinion, a query, a marketing note, news, a complaint, a suggestion, an appreciation, and many, many, many, many more things and then classifies the content into a set of predefined intent categories. And we work with each of our customers to figure out what these are, and they can be 40, 50, 60, 100. There can be lots of different intents, depending on the brand.

Let me give you another example. Someone will say something like I want to get my device replaced. When are your stores open? Right? Intent here is store timings. And again, device replacement, right? So how do we get a device to this person and get her the store times. And so this idea of being able to pull intents out and identify them allows us to be much better with routing, because we know where to send the message, much better with response management, because we know what we need to say to them. And we can assist agents in being able to respond to things really quickly.

And so basically, the way it works technically, if you’re sort of into that kind of stuff, is someone else have a message where essentially they’ll say something like I bought this device last month, and it’s taking too long to charge now, where can I get it replaced. And then each word in the message is tokenized. So the message is broken down into tokens. And then phrase detection occurs. And so the phrase detection actually will pull the words out that are associated with different types of intents. And then it will basically… things like taking too long to charge would be a phrase that you can extract from the initial message. And that is translated to an intent called slow charging. And where can I get it replaced is a set of tokens, which is a phrase, which means replacement in an intent. So that replacement is generally deemed to be the primary intent, because slow charging is a complaint, whereas replacement is a request, right?

So the to the intents will actually prioritize themselves as well. And so then what we can do is set up these things for routing. So depending on the intentidentified the message can be routed to specific agents specialize in handling that type of intent. You can actually create conversational chat bots. So they’ll use intents to provide automated replies to messages belonging to a particular intent. And then you can also use identification for agent assistance. So FAQ chat bots can use intents to answer commonly asked customer queries, which is really helpful. I mean, I often have questions and don’t really need to talk to somebody, I just want to know what times the place open or when does it close? Or how do I get it fixed. And if a chatbot can do that, for me, I’m all over it.

And then the best thing, of course, is for reporting. And so intent analysis can be used to provide consolidated numbers around prevailing intense behind the different customer messages. And this allows companies to take all this mass of messages that are coming at them through all these different channels. Remember, it’s modern channels, right? It’s not just social. So it’s anything coming across the forums that are out there, the review sites that are out there, blogs that are out there, all the messaging platforms that are out there, all the social platforms that are out there, chat on your website, all that stuff that’s coming in can be parsed. And what you can do is you can look at, hey, here are all the different kinds of things that people are querying about. And these are the primary intents. And so you could have a really simple pie chart. And you could have, you know, a bunch of people are generally satisfied, a bunch of people have general complaints. There’s delivery issues, location issues, advertising issues, checkout cart issues, career issues, job issues, order details, branch and store information, loyalty program questions, greetings, assistance needed, availability… goes on and on… apologies, attachments, loyalty damaged, broken, like you can just, you can add as many as you want. And depending what kind of business you’re in, obviously, if you’re in a service business, you’re not going to see a lot of replacement. But you will see a lot of maybe make betters or make dos or you know, do overs and that kind of stuff.

And so, you know, this for me is one of the more exciting things that’s coming up today from Sprinklr because you can actually see amazing increases in case resolution, the number of cases that are handled, and the number of inbound messages that are handled by bots. So we’re actually seeing in one case, there is a WeChat account, and they were able to get a 2.4X increase in case resolution speed, they are able to handle 25% more cases, and of 1.3 million inbound messages, almost 100 thousands were just able to be handled by a bot. And they’re able to do case deflection, and that makes a lot of sense. And as you do this more you get better at it. You get better at identifying intents and you get better at having the right responses for people that they need.

So this is a for purchase feature in Sprinklr. If you’re interested, you know, contact us through the website. And that is what we’re going to talk about today in AI. So I’m going to sign off. For the CXM Experience. It’s Grad Conn and I’ll see you tomorrow.