Episode #84: How to Use AI to Classify Data and Drive Insights

Yesterday we learned how to discover the billions of conversations happening right now in modern channels. But then what? How can you possibly sift through those billions of conversations to find the proverbial needles in the haystack? Spoiler alert: You can’t. But, AI can. And it can do it amazingly well. Today’s episode is all about classifying the data so you can use what you’ve learned to drive actionable insights.

All podcast episodes

Go Now

PODCAST TRANSCRIPT

Okay, okay, okay. Here we are at the CXM Experience. I’m Grad Conn CXO, chief experience officer at Sprinklr. And today we are continuing our series on discover, classify, and engage.

A quick recap of what we’re doing here. We’re motivated, inspired, I think inspired is probably the best word. Maybe excited, by the kinds of things that are happening in marketing today, around one-to-one. I talked a little bit about Marc Pritchard, and his commitment to mass one-on-one at Procter & Gamble, and talked about the evolution of marketing communications from what used to be pure face-to-face one-to-one, to mass communications in the 20th century, to what we now have in the 21st century, which is conversational marketing, which is one-to-one and mass at the same time.

I was talking to a customer today, and they would talk about two-way communication. Another great way of putting that. They had a great phrase. It was so well put. And I thought the person who said this was brilliant the way she put it. What she said was, don’t forget that when you think past the business, and you think past the applications, and you think past the phone, and all the other mechanisms of communication, remember you’re talking to a human. There’s a human sitting at the other end of the table. And that human-to-human connection is what really counts.

And this is to me what’s so exciting about where we’re going in the 21st century. We’ve got a whole bunch of new types of communication that enable a back and forth that has not been possible between brand and customer before. So great times.

We’re talking a little bit about what it takes to make a mass one-to-one marketing platform work. And we’ve got three stages right now, which is discover, which is you need omni-channel access to all digital channels. Classify, which is you need to be able to take all these billions of conversations and somehow classify it and sort it and make sense of it. And then you need to engage. When you hear somebody in pain you need to solve. If someone is happy, you need to amplify it. Somebody who’s got a question, you need to answer it. One-on-one.

I think conversational marketing is a good word. Because a lot of this is what you normally do in a conversation. When someone is talking to you, you need to listen. If you’re not listening, you’re not going to be very effective in the conversation. You need to sort out what they’re saying to really understand it properly. And then, if they ever request or if there’s something that you need to do to help them, you need to be able to respond to it.

So that’s the setup. Now what we’ve done in the last show is we spent a bit of time on discover, and drilled in a little bit on that one. Today, I’m going to drill in a little bit on classify. So, this is going to be a discussion on AI.

A little bit on AI for a second. AI is really composed of three things. There are the algorithms that are used to essentially create the neural network. And those are actually reasonably standardized. And there there’s not a lot of differentiation possible in the algorithms alone. We’re getting pretty good at these. So there’s the algorithm.

The second thing is the training set. So, the data that you’re going to train the algorithm on. That’s very important. We’ll come back to that in a second. And then the third piece is the actual training itself. The feedback. When the AI makes the recommendation, if that recommendation is off, there needs to be a mechanism to say, that’s not quite right. And so it can adjust. I thought that was red, but it was really blue. Next time, I won’t make that prediction, because I knew I was wrong. And I’ll get more accurate as we go down the path.

We talked a little bit about discover, and one of the things we talked about was the 400 million different data sources that are out there. And I obviously speak to this from a Sprinklr standpoint, because that’s what I know and that’s the context. You’ve got this giant haystack of stuff composed from social platforms, all the public stuff, right? Forums, things like Reddit, review sites, news sites, blogs, all the broadcast that’s out there. TV, print, all the print that’s out there. Radio, all that stuff’s all piled into the haystack.

Now, the trick is, I need to find needles in that haystack. And it’s a reasonably complex task. We’re actually talking about more than 16 petabytes of information. And literally billions of conversations. As Carl Sagan would say, “billions and billions” of conversations out there. And so, what we’ve done at Sprinklr to solve this problem, and we’re generally considered to be probably one of the most sophisticated AI platforms in the world, and head to head will beat any platform that comes our way. We’ve got eight layers of AI, more than 100 languages being covered, including some of the more complex languages like Japanese and Chinese, which are a little trickier. There are 10 billion predictions per day, with an 80% accuracy coming out of Sprinklr. 10 billion a day. And we have a little more than 1,200 AI models across 60 different industry verticals now. So that’s a sense of what what’s going on there.

And then we route this into a database. And the database, just to give you a sense of scope here, okay. And Sprinklr is obviously, way out there. I mean, we are the world’s leading CXM platform. And we are the world’s leading distributor of all this kind of stuff. And we’re pulling in more information than anyone else on the planet. So these numbers are big numbers. But that’s because they’re Sprinklr numbers. So in the CXM database, there’s a billion records ingested per day. We do 15 billion automation runs per day. We do real-time reporting on 600 billion records. And there’s real time alerting on more than a billion different time series. Just to give you a sense of the scope of this.

Now, what’s cool about this is once you’ve got a system like this running, and this has taken six or seven years to build, a very focused effort. A significant percentage of our R&D budget and a very significant part of what we do every day. Once you build something like this, then what can you do with it. And one thing that we do, which is great, is you can do location insights. So, if you’re a fast food restaurant, and you want to know which of your locations are performing well, we can actually isolate the sentiment, positive and negative, to different locations. It’s great for banks, it’s great for hotels. And people express their sentiments, so you know what’s going on.

You can also get media insights. We’re actually replacing some of the more traditional PR earned message tools out there, like Cision. Those older tools are being replaced by Sprinklr, all over the place, by some of the biggest companies. Because we access more information and more broadly than they can. And it gives you insight into what’s happening in your media, your earned media. It allows you to see issues before they become a problem. So there’s crisis management. There’s a whole bunch of stuff around governance, the use of mark, that kind of stuff.

Product insights, this is a very exciting area. We’re working with a lot of product groups in tech companies, hardware companies, packaged goods companies. People that have a strong product focus use Sprinklr to be able to find out what things they should build next, and what kind of things people are saying about the products that they already have. Makes a huge difference.

Competitive insights, it’s pretty obvious. But what is the competition doing? And we see that rolled up and we benchmark against the competition, and we understand how they’re working and what they do well. That makes a big difference. There’s Audience Insights. Who am I talking to? What do they care about? What do they do? Where do they go? What words do they use to describe me? What emojis do they use to describe me? There’s visual insights. Visual insights are really cool. We can actually see things. I see many examples of this at Microsoft, where we would see the logos but we wouldn’t see words in the post. And we’d be able to react to it and respond to it. We do this for lots of different clients. Some for example, we’ll use our visual AI to basically OCR serial numbers on say, a computer case, and be able to get the computer serial number into a customer care person’s hands without the customer having to do it themselves. And those numbers are tiny, right? So it’s a lot better if we could just do it for them.

We can see when people are trying to bypass a text-based system and do hand drawn flyers and post those to run parties and things like that. We can see that recognize it and then be able to alert. We’ve had a lot of really great success with that as well. Industry insights. Many teams will use us to sort out what’s happening in an industry and then publish that as a regular feed and turn what we’re doing into the definitive source of what’s happening in an industry.

And then finally, my favorite one is sentiment insights. And sentiment is, what is the sentiment that someone has about you? How do they feel how they feel about your ads, what are they reacting to? All these emotional elements are really, really powerful. As part of sentiment we can also now detect intent. And intent is a really important thing to detect in comments. So, you know how to route someone correctly. Whether you’re going to send them to customer care, or community management, or whatever that is, the intent of their message is really important. And there’s a lot of nuance and subtlety in there, that AI is really good at separating out.

And so, just to give you a quick sense of the flow of all this stuff, and all the different things that are moving through it. And you really do have to have one of the most advanced AI systems possible to do it. I think that where I see people fail in modern channel management, is where they try to manage it manually with human beings. The scope and volume of messaging is so high. I mean, that could have sort of worked 10 years ago. But the volume of messaging is so high now that if you have community managers, and manual intervention, you’re maybe going to get to 1% — maybe 1% of the messages coming at you. Probably not even that high. You just can’t manage it any other way, you’ve got to bring AI into the fore. And you’ve got to have the classification and routing and all the other kinds of things that AI does in order to make it work well.

Again, thinking about a mass one-to-one system, maybe one of the reasons sometimes people balk at it a little bit, is that conceptually, if you take the mindset of the broadcast world, and you apply it to the one-to-one world, it doesn’t make any sense. Because there’s a very tiny team of people, mostly manually producing marketing materials in today’s broadcast universe. And then you think, how are those people going to respond to a million incoming messages? Well, they’re not, there’s no way they can. And so then people give up or throw their hands in the air.

That’s why any mass one-to-one system has to not just have discover, which brings all this stuff in, it’s got to have the classify stage, which has to be a deeply sophisticated AI system, to enable people to make sense of what’s happening and be able to respond. Now you’re still gonna have to engage more people in the company. But the thing is, if you’re routing stuff correctly, and if you’ve got things classified correctly, and if you’ve got smart responses in place as well, then you can actually more broadly bring customer input and feedback and commentary into the company. You can have a more two way communication pattern with the customer. Because it’s been made sense of. It’s not like you’re turning everyone in the company into community manager.

That’s classify. So, we will be talking about engage tomorrow. It’s obviously the really cool part of it. And I think engage is actually what makes CXM, CXM. It’s the “M” in CXM. And there are a bunch of folks out there saying they’re CXM companies, but they’re not. They’re just really maybe CX. And are in many cases, just CF — customer feedback. We’ll talk a little bit about that. And we’ll talk about why the “M” in CXM is so important as you think about a platform and defining the category.

So, that was a very serious talk today. But that was fun. I enjoyed that. I hope you enjoyed it as well. Lots of the use of the word “billions.” And I did use “petabytes” as well. So, lots of big numbers today. And hopefully you enjoyed that as well. And I’ll see you tomorrow, on engage. For the CXM Experience. I’m Grad Conn and I’ll talk to you next time.