Highlights from the Twin Cities Startup Week panel “The Future of Marketing in the Age of Machine Learning” hosted by Cludo at our Minneapolis office.
As part of Twin Cities Startup Week, Cludo hosted the panel “The Future of Marketing in the Age of Machine Learning” right here in our Minneapolis office.
Our panel was populated by experts in the field of Artificial Intelligence and Machine Learning, from both the corporate and academic worlds. Addressing a crowded house, a robust discussion was had around various facets of the world of AI and marketing.
The panelists were:
Matthew Versaggi, Senior Director of Artificial Intelligence and Machine Learning, Optum Technologies (ATC) / United Health Group
Brianna Noland, Consumer Insights Data Scientist, General Mills
Dr. Manjeet Rege, Big Data Analytics Researcher, Educator & Consultant, The University of St. Thomas
Corey Christensen, Data Scientist, Cludo
Moderated By: Philip Andersen, Former CEO, Cludo
Dr. Manjeet Rege (University of St. Thomas): I think one of the biggest misconceptions is the two terms that we often hear used, AI and ML, and the fact that they are probably the same. Since many of you are working in start-ups, one tweet that I had read recently is, “If you want to seek funding, you should be using the term of AI. If you want to build your team and hire, you should be using the term ML.”
So, I think that is the biggest misconception. You can think about AI as a field, like for example, physics. ML is a series of tools that help you enable that larger goal. If you think about AI as physics, then ML could be like Deacon’s Laws that help you achieve something. So I think that’s my take on it, what are the biggest misconceptions.
The other is the fact that just because you have data, you can throw ML techniques at it, and there will be treasure underneath that, and you will end up discovering something. That may not always be the case as well. So that kind of takes us into whether you have data that is relevant and whether you have data that is clean and prepped that can be fed into a machine learning model, and how do you align your business goals with a particular machine learning technique, for example.
Brianna Noland (General Mills): I also think that one of the misconceptions is that it has to be extremely complicated. A lot of what we do at General Mills is really simple applications of machine learning, just taking traditional analytics and putting it into production, even automating things as simple as consumer segmentation that people have been doing for years and haven’t really maybe thought of as machine learning.
So I would say it can be simple. It doesn’t always have to be a big earth-changing endeavor. Sometimes it is, but it doesn’t have to be. And I would also say that having the idea that artificial intelligence is going to solve all of our problems … it can solve a lot of our problems, but I think especially in marketing, there’s a human element that we can’t replace with machines yet. I think eventually we’ll get closer, but I think there’s still some nuances that artificial intelligence can’t quite account for in marketing yet.
Corey Christensen (Cludo): Yeah, I think one of the biggest misconceptions I’ve seen is that machine learning is a catch-all solution. When people say, “Yeah, we’ve got a problem. Let’s just throw machine learning at it and it’ll solve our problem.” Which, sometimes it does and it can be used quite well, but a lot of times what you’ll find is that simpler approaches are faster, cheaper, and more efficient, and can solve the problem with a similar ballpark performance as what you’d find from the new, shiny, brand new machine learning algorithms that you find out there.
Philip Andersen (Moderator/Cludo): So based on that, when do you decide if you should throw the new, shining features at your data source, and when should you use a more simple approach?
Corey Christensen (Cludo): Usually, we’ll start with doing some exploratory data analysis and see, first of all, do I have the data necessary to make the machine learning worthwhile? And if yes, then I’ll start doing some comparison between doing a simpler approach and some of the newer approaches, just to see if a simpler and cheaper approach could get us the same results without having to invest as much into the machine learning itself.
Matthew Versaggi (Optum/United Health Group): So I’m going to take a completely different approach from the rest of the panel, because somebody has to do that. My biggest issue with misconceptions of artificial intelligence is that data science, machine learning, and deep learning is the entirety of artificial intelligence, and if you think that, and you would if all you read was the hype that you would get in social media and then in consumer-based journal articles, as opposed to cutting-edge research and coursework and whitepapers, you would be … If you thought that, you would be dead wrong, because that is probably the most hyped part, but it’s hardly it at all.
There’s a book called “Artificial Intelligence: The Modern Approach”. It’s about three inches thick, and it’s about this big. It’s really good for beating your kids with if they get out of line. (laughter from the audience) It is a standard textbook in universities these days for teaching artificial intelligence, and it’s written by one of the core researchers at Google, and this is one of their main thrusts. In the first couple pages into it, it comes out and says, “Look, this entire book is an exploration of intelligent agents.” What does it mean to be an intelligent agent? Consumer-wise, just walk into Target, go that hallway of the vacuum cleaners and look at those ridiculously priced round vacuum cleaners. You push a button and it figures out how to actually vacuum your floor in the best way. That’s an intelligent agent. It learns from its environment.
So, perception systems and navigation systems and reasoning systems and all these other kinds of things from game AI and swarm-logic and natural language in all of its forms, all of these things are part and parcel of creating a robot that goes around your house and does intelligent things. In addition to the old stuff from the ’60s like constraint satisfaction and logic and search and all that kind of stuff, those three areas, the data-oriented part, data science, machine learning. Neural networks. You stack neural networks, you get deep neural networks. The old stuff that’s been around, in your phone and transportation systems, you don’t even know it’s there.
And all of the stuff surrounding intelligent agents, which is really deep and really vast, that’s AI. It’s not just machine learning data science.
Dr. Manjeet Rege (University of St. Thomas): So just a kind of extension of the ideas that have already been shared by the panel, I’m bringing the conversation back to how to align machine learning with the business goal with regards to misconceptions. Often you’ll end up believing that to achieve a particular business goal, one particular model is enough, but that is most probably not the case. It is a sequence of models that help you achieve individual business goals. If I’m to predict if my product would be going out of stock at a particular store, I think the common approach in terms of misconceptions to think about, building that one model that can directly do that, as opposed to I can have a sequence of models. That one model can perhaps predict what the demand would be? What is my inventory in the nearby warehouses? How much time would it take to restock? Is there any seasonality to this particular behavior? And all of these models together can eventually help me predict.
Dr. Manjeet Rege (University of St. Thomas): I think with the chatbot applications, you go online, to any website, and then you are right away greeted by a chatbot that acts and behaves much more like a human. The chatbot engages you in conversation, and based on that interaction, a chatbot also improves over a period of time. But if that part of your experience did not go well, that part for your customer, we always have machine-learning algorithms that your interactions are being fed into. So that has been one of the major customer facing [applications].
If you think about how ML has evolved over a period of time, you can probably differentiate that into three waves. One of the first ones that engages successful applications was the recommendation engine. That was pioneered by Amazon and Netflix. Then you have more of the operational intelligence works that are more about the mundane manual activities that perhaps could be automated. And now you have applications where, is it possible for an intelligent device to be hearing what I speak? Things like Alexa, Google Home, and that is kind of the third way of this. So all of these are extremely customer interactive and customer facing as well.
Matthew Versaggi (Optum/United Health Group): All right let me grab this one. So, I’m going to approach this from a Fortune 6 perspective. We see AI and machine-learning and integration, IoT, everywhere in our market. Anybody think that the health-care system needs to be revamped in this place? Exactly. We do too! So we’re seeing things in our space that are combining the machine-learning and data science and analytics part with IoT, and a lot of these other highly networked things that are voice controlled, in emerging markets, that are really focusing in the next few years, on the customer experience. Everything from your hospital care, all the way down through Rx.
So we’re seeing things like smart hospitals, where end-to-end care is mapped out and it’s optimized analytically all the time. We’re seeing things like genetics and genomics experiments being done. Drug trials being automated through data science and machine learning. We’re seeing everywhere care, through IoT devices that hooked up to your mobile phone that you can wear, that streams clinical data all the time, so that we can begin to intervene with the bad behavioral aspects of folks who aren’t taking care of themselves. And it’s en route to going from this archaic feeder service, kind of like environment, to making this transition, which is ripe with disruption, all the way to a mature value-based self-care system. All of that stuff is under-girded with the AI machinery, as well as IoT, big data, and all of these other things. Right now we’re greasing up an old dying machine but we know we have to make this change. It’s going to create these opportunities, and then sustain these opportunities as well.
Philip Andersen (Moderator/Cludo): Would that mean that since I’m not exercising a whole lot, I’ll have to pay a higher premium, than maybe Corey who’s exercising more?
Matthew Versaggi (Optum/United Health Group): We can talk about that on the ethics side. (laughter from the audience)
Dr. Manjeet Rege (University of St. Thomas): Not just only in higher education but overall education as well, right from K-12 up until higher ed, we now have data about how a student has been learning. Is there a behavioral pattern? In the past, we worked with a school district in upstate New York. We had data about the tests that students were scoring on, and we were able to predict that if a particular student scored in a particular range, then there was a higher probability of that student getting incarcerated in a high school, for example.
So, these things can perhaps be used by administrators for coming up with intervention that, if there’s a higher probability, can we take some remedial action and prevent from that particular event occurring, for example?
Brianna Noland (General Mills): We are working on a lot of anomaly detection. There are a lot of different applications within marketing for anomaly detection. A few of the things that we have a specific focus on are being able to understand where we sit relative to our competitors with how we’re appearing, how we’re approaching our consumers. Are we meeting them where we should be meeting them relative to where our competitors are? And also understanding how is our spend, our marketing spend, as it compares to our competitors, and where is that marketing spend?
There’s a lot of applications of anomaly detection that we’re looking at. That’s kind of a big one. But when you think about marketing, there are a tremendous number of applications of anomaly detection. We’re starting to integrate that within lots of different areas of our marketing organization to allow us to more quickly be able to understand the landscape. What is the consumer landscape, what is the competitive landscape? Detecting those events really helps.
Corey Christensen (Cludo): I think one of the biggest trends we’re starting to see nowadays, too, is people are starting to take a more holistic approach to understanding their data and how to use it, and realizing that data isn’t just generated in a vacuum. There’s people on the other end or there are systems on the other end that create that.
A good example of that is with Cludo, since we do website search… we know people search because they are searching. What businesses are starting to understand is that when someone does a search on your website, it’s because they’re not finding what they want easily. That is a flag for you as a business that you should fix your content, change your content, add content, and make it more end user-friendly. It’s really the understanding that those searches and that data is coming from an actual human on the other end and understanding their background for doing what they’re doing.
Brianna Noland (General Mills): So a lot of the machine learning that we do is really internal-facing, so of course as a marketing organization we have a lot of external-facing applications as well. One of our big pushes right now is to utilize machine learning and artificial intelligence to inform, within our organization, to enable faster decision making.
And so, that means, what can we do as an organization to bring the right insights to the right people at the right time? That’s sort of our mantra at the moment, and so that means, how are we enabling those at the top to easily make decisions on where to go with our marketing efforts quickly? But external-facing, if you think back to… Decision Sciences is the name of my team, one of our goals is also to understand not only how we can make decisions quickly, but how our consumers make decisions.
And so, a lot of the data that we’re leveraging is consumer panel based. We wanna understand how are they making decisions. What trends are we seeing? And how can we speak to those consumers at the right time and meet them where they are? So I can’t talk about specifics, but both from the internal and external perspective, we’re focusing a lot on getting the right information to the right people at the right time, whether those be our internal customers or our external consumers.
Philip Andersen (Moderator/Cludo): So, how’s this changed? Before we used to say just “analytics”… how’s that changed your department and so on and so forth?
Brianna Noland (General Mills): Less reporting, a lot less time spent on reporting and digging around into different data sets. Our marketing associates would traditionally spend a lot of time preparing for regular reporting and update meetings, and one of our big pushes is to be able to remove a lot of that exploratory element, and help them understand what it is that they need to be focusing on for their regular meetings, to reduce the amount of time they’re spending getting ready for meetings with the executives.
So, definitely, we’re taking away a lot of the need for reporting and a lot of the need for human intervention when we’re speaking to our executives.
Dr. Manjeet Rege (University of St. Thomas): And if you also think about how AI has transforming marketing, so, marketing segmentation, what you refer to, has been done for a long time. But now with AI based marketing tools… you can have a marketing tool that a marketer can subscribe to, and then it can do autonomous media buying. You can specify what kind of demographics you are interested in, and then it can have a much more targeted and focused market penetration. You have marketing tools that can recommend something for your followers. So if I’m a skin product company, and I would like to recommend many other skin improvement blog posts online through my Twitter feed, I could have a marketing agent, again autonomous, that could be making those recommendations to my followers as well.
Brianna Noland (General Mills): And I would add to that, that it really just makes us a more agile marketing organization. We know what our consumers’ interests are much more quickly and so, we can shift, almost in real time, to adapt to those consumer interests. So, we’re much more agile, we’re able to change things as they need to be changed, whereas in the past, we wouldn’t know about changing consumer interests until after the fact. And at that point, your consumers are already interested in something new that you don’t even know about yet.
Corey Christensen (Cludo): I think at Cludo too, we have a slightly different approach, because we are ultimately a B to B company, so we’re not focusing more on marketing to end users, we’re focusing on, how do we engage with different businesses to get them interested? So a lot of our focus has been on taking a look at the sales and marketing pipeline, and identifying the different decision points throughout the course of it, and trying to figure out, where do people drop off? And then, using things like natural language processing to figure out, do we know why they’re dropping off at each of those decision points? So that we can more tailor our marketing and sales work towards that.
Matthew Versaggi (Optum/United Health Group): So I don’t pretend to be a marketing person, but I’m married to one. So I get some really good insights. And I take more of a macro point of view. From a big Fortune 6 company, when we look at applying any of these cutting-edge technologies, we go to the individual business segments and really get to understand what it is they do, what their pain points are, what the low-hanging fruit is, where to apply the technology reasonably to get a proper return on investment. In the marketing space, you’ve got a lot of stuff here.
There’s a little phrase, “What keeps marketing professionals awake at night.” And the list is really long, from distribution, to exposure, impression, recall, attitude shift, response, lead qualification, engagement, sales, profits, loyalty, it just goes on and on and on. Each one of those segments have their own unique data needs, and are ripe for disruption. Each one of those segments have done what they’ve done, historically, over time, using just cutting-edge analytics, though. At this point they’re ripe for changes. So any one of those can be looked at in depth, and these tools apply.
Dr. Manjeet Rege (University of St. Thomas): I think being a startup is a great place to be, especially with this huge interest in AI and ML. I think a startup is always well positioned to have an AI strategy laid out, right from the beginning. I often see these with huge companies that struggle to adopt an AI strategy, so I think there has to be a data strategy, and then part of that has to be an AI strategy as well, but how this particular goal, my business goal, could be accomplished with the series of ML models that I will be building.
So I think staying relevant is really important. With newer and newer advancements being made in ML and AI, any industry could be, I think, probably disrupted within a matter of, a couple of years, and they would have your competitor embracing that same technology very quickly, and putting you out of that business.
Marketers dedicate a significant amount of time and resources driving traffic to an organization's website. But what happens once visitors are there?
When planning your next marketing strategy and related marketing funnel, be sure that your site search is shaped-up to handle the searchers.
Introducing Private Data Masking! The feature masks any private data from MyCludo analytics that users may accidentally enter in your search.
Be the first to know about the world of search and analytics.