In the new digital ecosystem, utility customers expect from their utilities the same customized and responsive experiences they now enjoy from online shopping, social media, entertainment services, and apps. Today’s technology innovations like advanced analytics, made possible through the cloud and artificial intelligence, have the power to unlock customer insights like never before. Utilities can leverage these insights to deliver the most relevant, timely, and personalized customer experiences that help increase engagement, participation in products and services, and satisfaction. Lirio’s Jeremy Floyd talks with Fiveworx Chief Product and Technology Officer, Patrick Hunt, about the opportunities availability through machine learning to help utilities meet these objectives.
Jeremy Floyd: Well, hello. This is Jeremy Floyd, CEO and President of Lirio, the parent company of Finworx and Fiveworx, and today I’m sitting down with Patrick Hunt, our Chief Product Officer. Specifically, we’re going to drill into something that I think excites both of us, and that is machine learning and artificial intelligence. We’ve spent a lot of time and work thinking about machine learning and artificial intelligence and how they enhance the communication experience between the client and their clients, and we actually have that at work in our core technology. So, welcome Patrick.
Patrick: Thanks, Jeremy. It’s great to be here. I’m also very excited to talk about this particular topic, you know. We’re not going to go quite so far as the computer in the movie, “Her” did, but it should be a lot of fun to talk about machine learning and AI.
Jeremy Floyd: So, there’s all these terms out there. There’s “big data,” there’s “data analytics,” “machine learning,” “artificial intelligence.” So, help me understand: What really is the difference there? Kind of explain that landscape a little bit.
Patrick: Well, don’t forget “natural language processing,” “natural language generation,” “machine intelligence,” I mean the list of these terms and phrases can go on, and on, and on, and there’s a lot of similarities, but there are some important distinctions to understand about the terms as well.
Specifically around machine learning and AI, I think. Intel’s Nidhi Chappell had a really simple explanation about the difference. That AI is really the higher level science about machines thinking more like a human being and machine learning are the software algorithms that actually enable that AI to occur. And, to me, that really is the key distinction between those two terms.
And then, all the others that we mentioned are various flavors, or enabling technologies, or similar concepts that would all play together nicely.
Jeremy Floyd: So, bring that down for me a little bit. What’s an example of machine learning that we might experience in everyday life?
Patrick: Well, one of my favorite examples is my iPhone and Apple watch. Nel knows that I typically—and I’ve never told it this explicitly—but it now knows that I take my daughter to school most mornings. It knows that I go to work from there and it knows I go home from there, typically. And so, I will get alerts that say, you know, “20 minutes for school. You should leave now. Light traffic.” And I’ve never told it any of those things. I don’t have a calendar item for taking my daughter to school, I don’t have a calendar item for going home in the evenings. It just has figured that out, based on my past behavior: where I travel, when I travel there, how often I repeat those trips. And I find it incredibly fascinating, especially when they overlay traffic data to let me know if I need to leave a little bit earlier than I typically do, for example.
I find that really, really fascinating. But, obviously there are tons of other examples that we could cite, you know, like some of the things that spam filters and recommendation engines are doing, like for Netflix movies or things of that nature. And, you know, it’s happening so fast, at such a pervasive rate, you almost don’t know what’s going on if you’re not really following the topic.
Jeremy Floyd: So, your phone’s telling you, “Hey, you’re here, and it’s going to take you 17 minutes to get to your next destination.” First time you see that, does it freak you out on your phone?
Patrick: I’m probably not your typical user. It didn’t freak me out. It delighted me. It fascinated me that it knew that, but I could see how it might concern some folks. But, I made a conscious decision to invest in these technologies specifically so they could enhance my daily life. I’m actually a big fan of services that understand who I am, what my needs are, what my intentions are, and deliver me valuable information at the moment in time that I need it.
So, I’m probably not the best case. Now, take your average user. The first time that happens to them, maybe it causes them concern. But, if you think back to just digital advertising on the web, and what’s called re-marketing—if you visit a website, then chances are in the near future, you’re going to see ads for that website or that product or service on many of the websites that you visit.
It’s because you express an explicit interest in the product or service offered by that website, and therefore, you’re going to see ads related to that explicit interest. And at first that might have caused concern for some people, kind of freaked them out a little bit, but I think it’s now become common practice. And I think you’ll see the same type of thing in these more personal, intimate interactions that AI enables.
Jeremy Floyd: Yeah. So, I think technology’s constantly streamlining, creating efficiency for the human experience, right? And, you know I’m a big fan of Kevin Kelly’s, “The Inevitable,” and one of the things that he says there is that, basically, any cognitive processing that can be outsourced, will be. And so, what you’re describing there is sort of this latent technology processing information in the background, trying to make your life better. It’s not like when—and this is going to age me a little bit—because I had a few years where, at the house, you had to get up and actually turn the dial on the television.
Patrick: Yeah. I’m a little older than you.
Jeremy Floyd: And, you know, you hit the remote … the remote was a technological advancement to improve the human experience and we’ve seen that even going back to thirty years, fifty years, technology’s constantly doing that. So, put in that context, it’s not scary. But I think there’s some fear out there. It’s really about application. Is that fair?
Patrick: I think that’s fair. I think the big cause for concern that people have is more when things happen auto-magically by a system mining their personal data to make that human experience better. And so, what you have to really be cognizant of, or what technology partners are you entrusting that ability to? Whether it’s the big players in technology today, like Microsoft, Google, Apple, Facebook, and others, and what you’re allowing them to do, and how they access your data is really the most important part of that puzzle. And that’s an individual decision for everyone.
Some people have no problem that part of Google’s model is to track all sorts of data about you and use that to deliver up the most relevant advertising on websites, right? No problem.
Some people have no problem with that. Other people have a huge problem with the fact that Google is tracking your data in that fashion, and refuse to use Google’s services as a result. But, again, that’s an individual decision and I think everybody just needs to make that for themselves, and be aware of it.
Jeremy Floyd: So, there’s a lot of engines out there that’re doing some different processing. You said “natural language processing” earlier on. Can you describe what that is?
Patrick: Yeah, in its simplest terms, “natural language processing’ is the ability of a machine to look at written text and discern meaning and value from it. So, for example, I might be able to do sentiment analysis on that text and, if someone writes a paragraph review of a restaurant, or a hotel on TripAdvisor, I can discern sentiment of whether that person enjoyed their experience or didn’t. And I can break it down further—maybe they enjoyed their experience with the restaurant, but not with the pool area of a hotel—and really deliver some nice analytics back to the hotel in terms of how they improve their experience for customers. Again, that human experience, right? So, it’s all in that service of improving the human experience.
Jeremy Floyd: So, take that into our business a little bit. We’re using machine learning. So, describe for the listeners, what are we trying to improve? What are we trying to enhance? I think of the idea of a human torque, or these human torques, or these cogs that are out there doing a lot of manual processing. What is it that we’re augmenting by using some machine learning?
Patrick: Yeah, we’re actually doing quite a few different things. So, let me try to summarize it and then drive down into a few specific examples. Ultimately, what we’re trying to do is help our clients improve the effectiveness and quality of the communication they have with their customers.
Jeremy Floyd: So that’s a great … that’s the why, right?
Patrick: That’s the why.
Jeremy Floyd: Things to improve it, right?
Patrick: Yeah, that’s right. And who doesn’t want to improve their communication? Whether it’s a mass email send or a one-on-one phone conversation, or somewhere in between the two, we all want to have more effective conversations. We want them to be of the greatest quality, of the greatest relevance and meaning to the two people that are engaged in this conversation or this dialogue. And so, offering up tools that help improve the quality of those conversations is really important for lots of different kinds of businesses or organizations and how they relate to their audience.
Now, that’s the why. But, if you get down into the specifics of exactly what we’re doing, the first thing is, we’re huge believers in the idea of persona-based communication: where we can define a pretty small number of broad groups, who are behaviorally and psychographically defined, who have different attitudes, opinions, perceptions, motivations for why they might engage with an organization. And we use that to drive predictive analytics, and we use that to drive messaging strategy. So, that’s the first thing that we’re using machine for.
Jeremy Floyd: So, what does that mean? To get the right content, right person, right time? Is that how that works?
Patrick: That’s exactly what that’s about, because you and I are both motivated by different things, right? So, you might be motivated more by long-term financial gain and I might be more motivated by, you know, paying my bills every month, right? Two simple examples. So, how do we frame the communication between an organization and their audience that taps into those different motivators?
Right communication, right time, right channel, right message—all of those types of things, really narrowing down to what’s that specific thing. But it’s all about persona. So, what we’re doing with machine learning is inferring what that persona is: which of the five or six broad groups of the population does an individual human being belong to, and therefore, what are their likely motivators, psychographic profiles, psychometric profiles, if you will, that are going to cause them to pay attention to a message and engage with that message.
Jeremy Floyd: So, for years, I think, in digital marketing, we’ve heard this idea of: right message, right time, right person. But it feels to me like it’s always in the context of still using the blunt force tools. Use a mass email system in order to do that. How is our system different?
Patrick: We process all of the data that we gather at the individual user level. So, if you’re a contact in our data-base, we’re looking at your digital footprint, your social profiles. If we’re sending an email, or somebody else is sending an email on our organization’s behalf, we’re looking at your opens and clicks. Not just that you opened and clicked an email, but what was the subject of the email that you opened, and what was the intent of the link that you clicked, so that we can discern what your level of interest is in different content topics.
We’re tracking data from our clients’ customer relationship management software, where they’re recording human interaction, so that we know what they’re engaged around, in terms of one-on-one contact. And, at the time we send a message, we don’t blast the same message out to every single contact in the database. We use all of that data I just described, at the individual user level, to dynamically and in real time determine what message to send to what individual user at that moment in time to increase relevance. Ultimately, that’s the business that we’re in.
Jeremy Floyd: So, just to go back to the comment earlier. What you’re describing right now, from a human element, and from using some type of other system at that level of analysis, would require lots of data analysts going in, looking at the micro-level and then setting up multiple types of communications for each one of those recipients at the time of each send. Is that right?
Patrick: That’s right, or trying to do it ahead of time so they’re all stored in the database. But it’s a massive effort. And part of what we’re doing is reducing that effort by orders of magnitude and making it far easier for organizations to deliver that kind of personalized, relevant messaging to customers, without such a massive workload or such a long lead time. Right? And to do it much more efficiently by using machine learning and by using that dynamic assembly of the messages.
Jeremy Floyd: So, that’s great. So, that’s the area of machine learning really around the actual send, around learning more about the behavior. Are there other areas where we’re using machine learning or natural language processing in our system?
Patrick: Yeah, absolutely. One of my favorite examples is surfacing insights for the users of our software. So, a financial advisor, for example, might be dealing with 150 to 250 clients that they have on their roster right now. They want to know how they’re doing with those clients and so we surface an insight that talks about which percentage of their clients are engaged by virtue of all the interactions they might have had, whether it’s opening emails, viewing landing pages, or the advisor recording an interaction that they had with that person. And we can also identify those that might be at risk because they haven’t engaged with the advisor for some time: maybe they’ve unsubscribed from receiving future emails or there’s some indication that something’s not quite right in the relationship, and the advisor should reach out to them.
That’s just one example. This idea of personas that we talked about earlier, you know, we’re surfacing insights around what percentage of an organization’s total audience fits into which persona and that can tell us lots of things. How does that compare to the expected? And therefore, what does that mean for your business?
So, we’re really working hard to continually evolve all the kinds of insights that we’re able to identify in the data, and present those so that human beings that really do need to make decisions and take action in a compelling way don’t really have to think about it. It’s kind of a set- it, forget-it—just like the email campaigns we were talking about before. I get all of these great recommended actions that I can take to improve my business and improve the quality or the outcome of the communication I’m having with my audience.
Jeremy Floyd: So, talk for a second about how this is applied within the utility space and how Fiveworx really helps benefit energy efficiency. And how does machine learning play there?
Patrick: Well, just in case some of our listeners don’t understand how and why utilities would want you to use less of what they actually sell, there’s a very complex system of programs and policies and procedures across the country called energy efficiency, where utilities are required by their state regulators to help their customers use less energy, and it turns out that most customers, 68% of utility customers, aren’t even aware that those programs exist, and they typically think about their energy usage for about 10 minutes in an entire year. So, it’s not very top of mind for them, right? So, to get them to pay attention, to hear your message, you’ve got to get to persona-based messaging that we talked about before. So, we have personas that range from, on the one hand, caring about climate change and saving the environment to, on the other, only really being concerned about energy independence and return on investment payback period.
And so, we’ve created all of our content to appeal to those two and to the two other personas with very different key words, phrases, motivational tactics, to get people to pay attention to that message and act on it. And that’s number one. So, infer a persona: we identify which persona someone belongs to and then we deliver messaging that’s specifically related to those personas. Even if it’s about the same thing, by the way.
In example, you might be a cautious conservative and I might be a true believer, and the utility wants both of us to participate in new tankless water heaters, because they’re very efficient. But, we’re both likely to be in a more affluent demographic category, and therefore, both more likely to spend on a tankless water heater. So, we’re going to promote that same thing to both of us, or the utility will promote that, but they’ll promote it in very, very different ways. That’s sort of the first step that we do in utility communication.
The second is we really try to motivate someone to go from just participating in one energy-efficiency program to five or more, for lots of good reasons that we’ll cover in another podcast. But, that requires really understanding what a customer’s interests are, what a customer’s likelihood to participate is, how to motivate them from participating in one thing to participating in the second thing, the third thing, and so on. And that five number becomes a tipping point for customer satisfaction with the utility and other metrics. So, we’re really using it to define that customer journey and to personalize that journey over a period of time to motivate participation in those programs.
Jeremy Floyd: So, can you not achieve the same thing by sending out a blast email promoting energy efficiency?
Patrick: Well, yes and no. You can get some program participation by using traditional batch and blast email techniques, absolutely. But third-party, independent analysis of our methodology, for how we use persona-based messaging and personalization to drive the customer journey, shows that we’re actually able to increase program participation by 150% when compared with those traditional methods. So, if you actually want to succeed, then you use a methodology like that that we’ve just created rather than the more traditional method.
Jeremy Floyd: Have you seen any results that would indicate that we’ve had a higher performance because of using this system?
Patrick: Yeah. So, if you take that program participation number and you break it down to actual engagement with the communication methods that we’re deploying, emails specifically, we see about a 68% increase in email open rates and a 100% increase in email click-through rates (and these are based on industry benchmarks that have been provided by another firm), significantly higher in every metric across the board than those standard industry benchmarks.
Jeremy Floyd: Great. Well, is there anything else? If you’ve got a client that’s in the utilities base that’s listening, is there anything else that you’d share about Fiveworx?
Patrick: Yeah. I was just describing energy efficiency in particular, but the persona-based approach that we have and the research that underpins it all—and that’s an important point, it is research-based—is applicable to a broader array of utility and energy sector products and services that they might offer. So, a utility might be getting big into community solar or roof-top solar. There are utilities that we work with that have a big push in electric vehicles. There are new products and services, like home energy management, that companies like Google and Apple and Comcast are making a big play in, and they’re disintermediating the experience the utility has with customers.
So, we have the ability to apply the approach to communication to home energy management as well, and really strengthen the loyalty and the relationship that the customer has with the utility over time. And the last thing I’d say is, we apply this approach, not just for residential customers, although that’s the easiest to talk about. But the SNB market is probably one of the hardest markets for utilities to serve in terms of energy efficiency and extended products and services, and so we have a sector-based approach, as opposed to a persona-based approach, that allows us to remain relevant and then still use all of the machine learning and personalization capabilities that we’ve been talking about to engage that small business owner or that small business manager with their utility.
Jeremy Floyd: Good. Well Patrick, sounds like there is an ever-increasing amount of development going on in this space. Seems to me, from things I’ve heard, we’re in a very elementary or an infant stage of what is happening in machine learning, although it’s advanced significantly. Sounds like we’re going to have more of these podcasts to come in the future.
Patrick: Yeah, it’s a really exciting time and, although we are in the earlier stages of deploying machine learning, that should not suggest that we’re not doing really sophisticated work, that we’re not really on a great path. We are. And it’s going to be a fun journey. The technology’s changing all the time. The libraries that are available to us are changing all the time. More and more people are focused on it, therefore the talent pool is increasing, and we’ve just got to remain flexible and nimble in taking advantage of the market opportunity.
Jeremy Floyd: Great. Well, thanks Patrick.
The opinions expressed are as of the date of this podcast and may change as subsequent conditions vary. This podcast is not intended to be relied upon as a forecast, research or investment advice, and is not a recommendation, offer or solicitation to buy or sell any securities or to adopt any investment strategy. There is no guarantee that any forecasts made will come to pass. Reliance upon information in this podcast is at the sole discretion of the listener.
© Lirio, LLC