About Fabrice Martin
Fabrice is responsible for the vision, roadmap and go-to-market strategy for the Clarabridge CX Suite of products. He brings more than two decades of experience launching new products and business applications focused on solving large, complex analytical problems and delivering valuable insights.
Follow Fabrice Martin on LinkedIn
Follow Fabrice Martin on Twitter @fmartindc
This episode was also recorded in video format. To watch the conversation, tune in below.
Mary Drumond
Mary Drumond is Chief Marketing Officer at survey tech startup Worthix, and host of the Voices of Customer Experience Podcast. Originally a passion project, the podcast runs weekly and features some of the most influential CX thought-leaders, practitioners and academia on challenges, development and the evolution of CX.
Follow Mary Drumond on Linkedin
Follow Mary Drumond on Twitter
Follow Worthix
About Worthix
Worthix was born in the Experience where customers are the backbone, and customer-centricity is the soul of every company. Innovation is at our core, and we believe in welding technology to bring companies and customers together. Our purpose is to use cutting edge mathematical models and Artificial Intelligence to extract actionable, relevant, and easy-to-understand insight straight from your customers’ minds.
Transcript
Mary Drumond: Welcome back. We are on episode seven of Voices of Customer Experience podcast, and today I am joined by Fabrice Martin. He is the CPO of Clarabridge. Clarabridge is a really cool tech company, very similar to Worthix. So I’m really happy to have Fabrice on here today. Hey Fabrice.
Fabrice Martin: Hi, Mary, how are you?
Mary Drumond: It’s great to have you. Thanks so much for coming on.
Fabrice Martin: My pleasure and chapter seven is a good number to be in.
Mary Drumond: Yeah. So I’m going to start off this episode, giving you a chance to introduce yourself to our listeners and tell them a little bit about how your career progressed to the moment that you are today.
Fabrice Martin: Excellent, excellent. Well thanks for the introduction. And first of all, my pleasure. It’s really great to be with you and with your audience today on this chilly day here in Washington, DC. This is where I’m based. And so I’m originally from Mexico. Uh, I started engineering, computer engineering in Mexico.
And, uh, I was hired in the late nineties to come into the U.S. to work for a software company. So I started my career as a programmer, in some of the heavy algorithmic areas of that company. It was a business intelligence company. And what I worked on was a, an SQL generation engine. So we would feed it objects and clicks and abstractions, and it would generate SQL programs to go on what a query, what were the largest databases back at that time. So did the engineering thing for five or six years. Um, then I noticed that I really liked the business side of, uh, of the, the shop as well. I was very interested in understanding how customers were using our product, uh, what new features they wanted and why.
And, uh, little by little that pulled me more and more into product management, which is ultimately what I do today. Uh, so, in that same company, I moved to the product management side, uh, and then did a startup. Uh, I started an MBA to kind of fine tune my, more of my business skills. And here I am, you know, a few years later, uh, Chief Product Officer here at Clarabridge.
So, so, uh, it’s been a fun ride and I love that crossroads of software, high technology, analytics and business stats. That’s where I think the action is and where I love to work.
Mary Drumond: So do I, but don’t take my word for it. I work at the exact same type of company.
Fabrice Martin: That’s awesome.
Mary Drumond: So tell me a little bit about Clarabridge and Clarabridge’s mission with its technology. It was, it was pretty revolutionary and groundbreaking when it came out. Wasn’t it?
Fabrice Martin: Yeah, absolutely. And it’s interesting. We have, uh, uh, similar philosophies and similar approaches between Worthix and Clarabridge, and it is to apply AI, apply natural language processing to improve customer experience.
Right. So, the way we see it, our philosophy is that we want to let our customers, and those tend to be large B2C, companies in different verticals, maybe healthcare, financial services, CPG retail, online retail, mostly. And, um, we want them to understand very clearly their customers’ voices, right?
What are they telling them? Why are they expressing feedback? It could be a good feedback, bad feedback. Uh, what do they want? We let those companies understand that feedback at very large scale. Then we provide analytics to understand that feedback and that feedback, by the way, it could be surveys. It could be online, uh, social posts, for example, or ratings and reviews. It could be chats, emails, calls. We all consider all of that feedback. We bring it into Clarabridge, make sense of it through analytics and natural language processing and understanding. And then we provide reporting tools, analytics tools, workflows that enable those companies to respond, to improve product, to improve marketing, improve services, improve brand. Uh, so again, improve customer experience in general.
Mary Drumond: So the cool thing for me about Clarabridge is that anything can be feedback. Right. Like, you can, you can pick it up. You can, you can scour the web or, or even your, your databases internally, or your call logs from your contact centers or, or customer service departments and, and everything is feedback. Right?
Fabrice Martin: Exactly. Exactly, exactly. It, it’s a very interesting, uh, point that you make. Everything, every conversation, uh, usually feedback tends to be categorized as solicited feedback or unsolicited feedback. Right, right. Uh, solicited feedback is where the company is proactively asking for feedback and it can be a survey. Uh, it can be, you know, sometimes in a video format or audio format or electronic forms. Right. Uh, but there’s also unsolicited feedback. Right. So, so it’s, uh, uh, stuff, conversations, posts, that customers are putting out there in the internet or through a contact center and expressing opinions, desires about a specific brands.
So it is just as important in, in my opinion, for companies to be listening to that other feedback, the unsolicited one, because it is something that a customer, uh, expressed and chose to express without any bias, they just decided to put a great review, for example. So why not learn what it has to say? Right?
Mary Drumond: Do you find that with unstructured data? Um, you, you tend to pick up on a lot of positive feedback that you don’t necessarily get from solicited feedback? Do you feel like in solicited feedback, when people have a problem, they tend to focus on that problem. Or they tend to focus on the more recent interactions they had with a brand.
Um, whereas on social media you can pick up on nuances that you might not, not necessarily get when, when solicited?
Fabrice Martin: That’s an interesting question. I think, uh, there’s, there’s definitely, um, well actually let me step back. It depends a lot on the type of source, right? Some sources can be super, super noisy, right?
If, uh, for example, we try to fish out information about a company on Twitter, uh, 95%, probably even more, is going to be noise, uh, that doesn’t really carry that much signal, that much information. Right. But you’re going to find some good nuggets there and that’s where the natural language processing technology is so important, because it can let you filter out the noise and get to the signal. Um, other types of channels, for example, the, the contact center, right, uh, are super, super rich in terms of information, right? Because when the customer is calling or chatting or writing an email to the company, the customer needs something, wants something, right. And you have to pay attention. And then the signal to noise ratio is going to be very high because again, there’s, there’s the full intent of the customer. They’re making their best effort to express what they want or what they need, when they need it. Right. Because they have that interest of getting something from the company.
So very important to listen. And then you have the historical aspect. You can look at one individual communication from a customer. But if you look at their entire historical, then you start seeing patterns, you start seeing needs, and you start predicting, if we get into technology terms better, what that customer might ultimately want or need. And then you can look at other customers like that one and start extrapolating. It becomes really interesting.
Mary Drumond: Hmm. Yeah. And the interesting thing is how do you guys work with like emotion? Is that something that you’re able to gauge with your technology? So for instance, how the customer feels when they’re giving that feedback?
Fabrice Martin: Yeah, absolutely. Absolutely. And there’s the, why this is important is because again, we’re all about customer experience and there’s three key pillars. This is a framework I really like about customer experience. It’s effectiveness, ease and emotion, right? The three E’s of customer experience.
And this is a term that Forrester coined, I didn’t come up with it, but it’s one that I use quite often. Effectiveness is if you’re a company and you’re providing a service or a product, how effective are you at providing value. How effective are you at getting the customer what they need, right. So you have to be good at that if you’re going to be a longterm business. Then ease is okay, your product is great. It provides value to the customer, but are you making it easy for that customer to do business with you? Right. If you make it very hard, you’re going to lose the loyalty of your customers. And there’s plenty of studies around that.
And the third one is emotion, right? So you provide a great product. You make it easy for customers to do business with you. Third one, you make them feel great about doing business with you, right? Uh, if you do those three really well, you are a champion company, you’re the best of the best, right? So that’s the why.
That’s why effort, ease and emotion are important in customer experience. And how we do it using AI, using technology, is we focus mainly on language. So the way sentences are structured, the way words are used will denote a certain level of emotion. So if I tell you, I’m very frustrated with the way your website works or doesn’t work, that expresses a high degree of one emotion, frustration, and it’s intense, right.
I’m very frustrated. Um, right. So, so that can be one way to get emotions. And then you can tease out different emotions. Uh, we have models about with 50 different emotions: love, anger, frustration, surprise. Um, and it’s interesting. Some of the emotions like surprise can be positive or negative. You can be pleasantly surprised or you can be horribly surprised or disappointed.
Right? So those subtleties in language are very important as well.
Mary Drumond: How about profanity?
Fabrice Martin: Oh, very associated with emotions. Yes, absolutely, absolutely. That’s a great point, right.
Mary Drumond: But like it can be a negative or a positive. Like I’m the world’s biggest potty mouth. The people who listened to this podcast regularly know this. But, so I’ll express profanity when I am super happy. When I’m super sad. When I’m frustrated, when I’m annoyed, like I’ve got multiple uses. Um, so like how does the NLP do that? Like use like just the word itself.
Fabrice Martin: I love that question. Love that question, Mary. So we call those modifiers. And by the way, you know, modifiers don’t have to necessarily be profanity, but let’s say, uh, your podcast is great, right?
Uh, so that’s one piece of feedback that you’ll take and, and by the way it is. Uh, so then if I say, um, your podcast is not great, right? The not is a modifier that completely changes the sense, right? So we keep track of those. And by understanding semantics, by understanding grammar, we can understand where that is going.
And now if I said, Oh, wow, your podcast is freaking amazing. Right. And notice that I changed the profanity there. Uh, it, it denoted the same emotion, that I liked the podcast, but I said it’s freaking amazing. So the modifier made that intensity of the emotion and of the feedback that much higher, more intense, right?
So, so that’s how we apply modifiers to sense, and frankly understand the meaning of feedback and of conversations in general.
Mary Drumond: And, and was there, I mean, I imagine that a lot of training went into the module in order to be able to grasp the subtleties of the language. Um, is there a training that goes into like exclamations? Like what the freak?
Fabrice Martin: Absolutely, exactly. And you’ll notice that profanity can go in a good way or a bad way. Right. When I said this is freaking awesome. It’s very positive. Right? So, so profanity, can also, context will provide you whether something is positive or negative.
Mary Drumond: Right.
Fabrice Martin: Now, in terms of the training. There’s a few factors.
It is important to understand the words and the sentences and keep them updated. So we use several techniques, uh, to, to refresh those words, right. And to make sure that we keep them current, uh, we do that in multiple languages. So it’s also important to understand the subtleties of the language.
For example, people in the UK, if they use a word, the same word as in the U.S., it could be really intense based on their culture. Right. There’s there’s cultural subtleties as well, uh, versus here in the U.S. So you also have to be aware of, who’s speaking and that context will give you a lot of hints about how to interpret profanity, modifiers, any type of, sentiment or feedback.
Mary Drumond: And when it comes,
Fabrice Martin: I don’t know if I answered your question really
Mary Drumond: Well no, it was more about training the machine, right? The different ways to train. So like how, how do you train for regionalism even within the U.S.? So if you consider the English language and you take, um, colonial english, let’s say, so Australia, New Zealand, South Africa, et cetera, versus from the United Kingdom proper versus the U.S., that’s one thing.
But then even within the U.S., you have certain expressions that are super-duper regional, right?
Fabrice Martin: Yes.
Mary Drumond: So how are you able to train the machine to notice almost I’m going to say cultural subtleties of the language or idiomatic expressions, for instance.
Fabrice Martin: Right. Great question. And there is a limit to how you can train any algorithm, right?
There’s the tension between precision and recall and not to get too nerdy here, butthe more precise you make analgorithm to understand certain words and subtleties, the less recall it’s going to have. Right? Recall, you can think of it as a net of how much language you can capture. And the more precise you make it, the less language you’re going to capture with that.
Right. So, there’s a trade-off, and we try to strike the fine balance where we get you enough precision, enough understanding, but broad coverage. Right? So, we take a very specific approach, which we call a hybrid approach. Right? So there’s only so much you can do through training an algorithm.
Eventually there’s a point where somebody needs to modify it or override it. Right. So, so that modification or ability to override,is based on rules, right? So somebody can create a rule. If this word is pronounced or mentioned, then change the rule and bias it more towards, I don’t know, negative sentence.
Right? So that hybrid model is what allows us to override the machine and the training, if you want.
Mary Drumond: So you started getting really technical and that’s when I started getting really happy. Cause that’s when it gets interesting for me, like we can’t make it too technical, but, but slightly technical is interesting. Because it kind of explains a lot of the mystery behind what people consider to be AI. You know, people are like AI and they start imagining the freaking Terminator or some like, weird robot with like alien- like figures.
Right. But in, in general, in general, what do you Fabrice consider AI? You.
Fabrice Martin: So for, for the domain that we’re talking about, which is customer experience, I see it as a way to give analysts and businesses superpowers, right? Uh, you know, they can understand- actually, if I take a step back, imagine no AI and no technology, like the one that we’re talking about. And we’re back to, you know, paper forms, uh, after you have an experience, they would have to be scanned or manually transcribed, and then somebody would try to make sense of, I don’t know, maybe a 1,000, 10,000 pieces of feedback. Right. And that was it. That was the capability that a person or a team of analysts could have. Now with, again, social media, millions of conversations flowing through chats, through emails, through contact centers, millions of surveys circulating around the world.
It’s just not humanly possible to process that information, but it is there. And if you want to successfully provide the experiences that your customers want, the products that you need to offer so that the competition doesn’t beat you, you have to tap into that information. You have to listen. And the only way is with this type of technology, with AI.
So, so again, I see it as a way to augment your capacity and give you superpowers in some way.
Mary Drumond: A lot of people talk about, when it comes to machine learning, especially with natural language processing that involves, um, training the machine. When you have to be very, very careful at the start because any sort of bias at the beginning may be imperceptible, but as time goes by that curve starts getting more and more clear.
And at some point I remember when, when the first company started dabbling a little bit more in that, they sometimes had to reset entirely because the machine had gone really far off course. So how do you guys avoid that type of situation?
Fabrice Martin: Uh, great question. And I think I go back to that hybrid model, right?
Where there’s always the ability to, change a rule or point the machine in a completely different way. So that, that is a way to enable, or to find that bias that could have been inherent in the data that you use for training or for modeling. And certain algorithms we have, for example, algorithms that will be used to score, for example, a phone conversation, right? This is very common in contact centers. Um, one way to score how a conversation went is by sending a survey to the customer that called after the call and say, okay, well, how would you rate that conversation right? From 0 to 10 or 1 to 10.
Was it great? Was it not great? And then give us some other feedback, right? That’s one way. The other way, one of the issues with that way is that there’s very low response rates, very, very low response rates. So probably 95 to 98% of those conversations go unscored or unrated. And if an agent is going to be evaluated or even compensated based on that score, I mean, it seems kind of unfair, right?
That maybe it was that one call where things didn’t go that well, but that customer filled the survey,
Mary Drumond: Especially cause people go the extra mile when they’re angry. Right.
Fabrice Martin: Exactly, exactly. Speaking of emotion. Right. So, we think that we have a better approach, which is, well, let’s score the conversation itself.
Let’s look for patterns of emotion, of effort, of empathy from the agent, from the customer, and try to come up with a set of rules that can transparently, and that’s the keyword, transparently, score that conversation, right? So, uh, for those types of algorithms, we use completely explainable and transparent algorithms.
So that if there’s a question of, okay, well, how, how come I was scored in this way? Uh, there is no black box. It’s fully explainable and you can trace it back to exactly why the algorithm decided to give you a 5 or a 7 or a 10, right? So there’s going to be situations where you want to be a 100% transparent.
There’s going to be situations where it’s okay to have a little bit of a black box there, still with the ability to override ultimately what the machine decided.
Mary Drumond: What do you think is the very best use of the intelligence provided by Clarabridge platforms, for companies?
Fabrice Martin: Hm, that’s a good question.
So the best, there are a few, but one that I think is very valuable, we try to frame it very simply when we speak to customers or prospects. Right. Um, in the end it’s about business, right? So our technology has to save you money somehow, right? Through improved operations. Or it has to help you make money, right, through improved sales processes, or you know, better product positioning or segmentation.
Or it has to help you reduce risk or liabilities. Right? So, so for example, if our software helps you reduce the potential of government fines in a compliance use case, right. Because language can be detected, uh, then that’s another area of ROI. So is there, within those three, a better one? I don’t know, but, but again, the way we try to frame it is always, okay, what’s the value that we bring into a company. And we’re going to help you, again, make money, save money or reduce risks. Uh, ideally all three of them.
Mary Drumond: Cool. You know, pulling slightly away from Clarabridge, a little bit more into like, you as a person or as an executive and as a professional, do you ever get questions or maybe pushback from people because you work with artificial intelligence? Do you ever get the claim of artificial intelligence is stealing people’s jobs and a future of hell and damnation. Do you, do you get that?
Fabrice Martin: We do. We do. I do. There’s a lot of literature out there. I mean, you talked about Terminator, right? Basically AI is coming to take over.
Mary Drumond: It’s so scifi-esque, right?
Fabrice Martin: Yeah. Take our jobs, rule the world, et cetera. And that’s a way to see it. Right. Of course. And there is no denying that there are certain areas where AI can look a little bit scary. Right. But, but at the same time, when applied properly when applied ethically, it can be incredibly liberating.
Right. And so instead of doing a repetitive, sometimes boring work. Uh, suddenly it can free you up, right? If I think of the contact center, for example, and you think of a chatbot technology or that type of technology, is it taking jobs? Not necessarily, right. It’s helping solve customer issues or prepare the customer for a conversation with a human.
And then the human is not doing the same conversation every time. Hey, how do I change my password on the website? Right. It gets boring at some point. So if that can be solved through a self-service webpage or a chatbot, then the human is still there, but it’s going to have much more meaningful, interesting human conversations with the customer.
Right? So, again, there’s many ways to see it, but I think AI is a technology for the better in general.
Mary Drumond: Do you think that in the future, let’s look a little bit into the future. Not way out, cause we’re also just get, go down like a rabbit hole. But so you know how, when surveys first started, like the whole idea of the 10 point scale or the 11 point scale and eventually customers learned how to manipulate surveys. So they know that if they give a 10, you know, or versus if they give a 0, et cetera, do you think that at some point, customers are going to learn how to manipulate an NLP, like by specifically saying certain words that trigger a response. Like for instance, I learned, my husband worked with call centers for a while.
Um, so he always gives me these pointers of which words to say, and normally it’s like very inflammatory words that you use during like a contact center call that will immediately like push you, into a priority or urgency line to solve your problems quicker. And this is like stuff that we learn as customers, like how to hack it, you think that that’s going to happen with NLP as well? Where people will eventually learn how to say certain words just to try to get some sort of advantage.
Fabrice Martin: Yeah. It’s interesting. I would imagine, you know, we’re clever and we try to beat the machine at its own game. So, so I would imagine that there’s going to be situations like that.
And, you know, I think of it even in the world of surveys, right? I mean, sometimes you go take your car to the shop and after you get the surveys, the person tells you, hey, I need a five because blah blah blah, right. So it can go both ways. It’s the person submitting the survey, but also the person on the other side, trying to game the system as well.
So, so we’re human. It’s part of what we do. We try to game the system and game the machine. So I would not be surprised if some clever hackers, try to hack NLP as well.
Mary Drumond: We’ve been talking for 38 minutes now, and that really flew by.
Fabrice Martin: That’s amazing. It was a fun conversation.
Mary Drumond: Yeah, this, I’m really fascinated. You know, being a part of this world of bringing technology into customer experience, it’s such a fascinating world. And I really think that we’re on the right track of really making customers’ lives easier by giving them, I’m not only going to say quicker access to things like you said with chatbots where you just, it just speeds up the process, but also in finding ways to value customers time and be respectful of their time by inquiring the least amount possible and maximizing that feedback so that the companies are able to make good decisions that benefit customers lives and give them a better experience as a whole.
Fabrice Martin: Absolutely.
Mary Drumond: Without having to solicit these big, huge, long surveys that have, you know, 35 questions and take, you know, 45 minutes to accomplish. So, I’m really happy at how much this industry is growing and pushing, and I’m glad to have companies like Clarabridge out there, really working hard to make technology, a part of customer experience.
Fabrice Martin: Yeah, absolutely, absolutely. And what you guys are doing is also fascinating, right? As you say, it’s trying to maximize the information that a company gets while minimizing the effort in the customer’s part to provide that information. And it’s the information that the customer is interested in providing.
Not what some analyst, somewhere in a desk decided to put as a question. So I think we’re doing the right things, right? Moving the industry forward, using technology to improve both the customer experience from a customer perspective, as well as with companies that want to dedicate their efforts to that as well.
Mary Drumond: Yeah, that’s amazing. Fabrice, if our listeners want to connect with you somehow to either continue the conversation or shoot you a question about something that you said, what’s the best way to connect with you?
Fabrice Martin: Oh, the best way to connect with me, and by the way, I would love to connect with folks who listened to this podcast. I’m always fascinated by these conversations. And I love nerding out on NLP and text analytics as you saw. So, LinkedIn, LinkedIn is probably the best way. So it’s a linkedin.com/martinfabrice. That’s my link.
Mary Drumond: Awesome. Well, I’m going to wish you a wonderful day and wonderful holidays because we’re nearing the end of 2020.
And you are welcome to come on this podcast anytime and nerd out on NLP, because I love it.
Fabrice Martin: Love it. Thank you so much for hosting Mary. This was a real pleasure and happy holidays to you as well.
Mary Drumond: Thank you Fabrice.
Fabrice Martin: Take care.
