Search the website

Pini Yakuel, CEO and Founder of Optimove, discusses the future of innately intelligent marketing, including catering to Self-Optimizing Customer Journeys.

Video Transcript

Hello everybody. It’s great to be here at the fourth Optimove Connect. Super excited. I want to thank Amit and Tal and all of the wonderful people of the marketing department to make it happen again. I want to thank our customers, our partners, people from the Israeli ecosystem, from Europe, from the U.S., who took long flights.

Really thank you, and I hope you learn today. Enjoy Tel Aviv, enjoy the beach, and we can have a wonderful experience together. So today I’m going to talk about innately intelligent marketing. It may sound a bit fuzzy at the beginning, but it will become clear. And this notion of innate intelligence is something just… I took it from… I’m a big biology buff and, you know, our body has this notion of innate intelligence, which means our immune system knows when to kick in, and our metabolism, and our ability to heal, everything happens innately with us kind of like guiding the intelligence.

So that’s something that I want to relate to marketing and I’m going to talk about how we take it from concept to reality. So, first of all, every year here we do this notion of State of the Union, the Optimove Union, so a little bit about what goes on in Optimove. So, first and foremost, this year, we moved to a new home.

We have our new offices in Tel Aviv. Some of you guys have visited us. We have our folks from New York and London coming in for the first time for our Tel Aviv office. And it’s a wonderful home for us. Many, many years we were more grungy in our kind of like home and offices and now we’re a bit more grown up. So that’s exciting. Some facts and figures about Optimove.

So, this year, again, we enjoyed a very high growth and thanks to everybody who worked so hard on it, so 60% growth. We are now 200 strong, so 200 employees globally work in Optimove. And, this year, we’ve also done our first acquisition ever. We’ve acquired DynamicMail, which, if you think about Optimove throughout the years, we typically solve the personalization problem via segmentation and customer modeling.

And with DynamicMail, we essentially venture into solving personalization through content. So that’s the first time that we actually addressed personalization through content. And, specifically, DynamicMail, our technology today, we can do on-open optimization of email. So, for example, when you open an email, the most relevant content will surface at the moment of open, whether it’s related to inventory, price points, some kind of an event.

And that’s really exciting, so I feel also now that we’ve successfully assimilated the DynamicMail team within Optimove, so that’s been a big thing for us this year. In terms of new clients… So when Optimove started, we were typically adopted by, what you call, the pure plays. So, at the time, they called them internet companies, mobile first, however you want to call them, but young companies who were born online.

And as any disruptive technology, that’s the path that happens very naturally. And today, we see a lot more mature brands, the omni-channel retailers adopting Optimove, and still, the very innovative brands which are pure plays. So folks like Family Dollar, and Lamps Plus, and Adorama, and River Island, all of those omni-channel retailers, the same thing we see in gaming and in social games.

So we continue to work with brands who are customer-centric by nature across multiple industries. Product. So this notion of a solution. So I don’t know if you guys know, but there’s a big difference between a tool and a solution. So, this year, we have solidified our kind of like us becoming a solution.

And a solution, it means that we solve a problem. The relationship marketing problem is a big problem, it’s a big endeavor. You need to understand machine learning, you need to understand marketing, data, the talent that you have in your department in order to operate. Relationship marketing needs to be diverse. You need to coordinate tasks and abilities between a few different departments.

It’s a big thing. It’s a hard thing to do really, really well. And now, we are a one-stop shop for all things relationship marketing. And I want to show this through a schematic view. Don’t worry about it, you don’t need to look at every little box. But essentially, our core is this notion of having this customer data platform, a kind of a warehouse for the marketer with all customer data taken from multiple places, flattened, deduped, cleansed, and modeled in one place.

And around that, we have our predictive modeling, we have the orchestration layer that does both real-time and scheduled messages, we have the optimization bot. We’ve recently added the business intelligence capabilities, and then we have all of the channels that do communication, whether it’s our own channels or third-party channels. And our strategy here is to innovate around our core and enhance our solution around known problems.

So, for example, email, it’s a known problem. There’s hundreds of email vendors, it’s something that the world knows how to solve, and we offer this as well. But it’s not where we’re going to innovate. We’re going to innovate around our core, around our predictive capabilities, around our ability to orchestrate, and be a very smart and actionable customer data platform. So speaking of customer data platform, it just become such a big hyped thing in this recent year.

I just want to address it very specifically. So, at the heart of Optimove is our CDP, our customer data platform. And CDPs are all the rage these days. There’s loads of people talking about CDP this, CDP that, there’s multiple tenders of CDPs. This is one of the Gartner analysts, Martin Kihn. He wrote about…many people write about CDPs, but, just like this article specifically, and there’s a big hype.

And I can tell you, as a founder of a company where, at the beginning, I was always looking for a category, right? Investors would ask me, “So where do you fit? To which Magic Quadrant in Gartner do you belong?” And I was like, “I don’t know.” I mean, I never really knew where do we fit? And finally, we fit here. So there is a category around what Optimove does.

So CDP has become a category, and we were a CDP before the world actually knew what it was. I want to give a big shout out for our first customers, a few of our first customers, we actually have many of them here, which adopted us at that point in time, which means you guys were early adopters. So CDP is a thing and, yeah, so I’m happy to have a category.

All right, so let’s talk about today’s topic, which is the next thing to solve. I like to think of myself as a problem solver and a truth seeker, so you’re always kind of like thinking about the market and what’s out there, what do you want to solve? And, in the last year…actually, in the last two, three years, you know, the notion of AI and machine learning has been super buzzy.

Right? Every company does AI, machine learning. Sometimes people that do seven times three call it the AI and machine learning. And who knows the difference? It’s very hard to understand what it means. Does it mean computer science? Does it just mean being smart about something?

And it made me think about this question, is there any real intelligence in marketing technology? Or what comprises real intelligence in marketing technology? And I kind of like looked, you know, deeper and, you know, with great folks in Optimove, and tried to kind of like map the current state of orchestration.

So I want to focus on something more specific in the greater realm of marketing technology and focus on the state of orchestration. You can see from the picture, which I don’t believe that that state is very, you know, is very good. I think that we all know those, you know, very convoluted journeys from hell that nobody really wants to build or maintain.

It becomes very, very difficult. But I’m going to break it down further. And mostly, I don’t know if you know this version of…this kind of like category of software, but there’s a category called BPM, Business Process Management. And essentially, if you look at most marketing technology today, an orchestration, you notice that you have a lot of automation. So almost every marketing technology allows you to set some kind workflow, some kind of a rule set that will be, after that, ran automatically.

If you set it up to run every day, every month, every minute, it’s going to do this pretty well. So automation is very much covered. You got some connectivity, some API connectivity to other systems that you can connect to and speak to. And you have some data, typically limited access to data, but you do have some data. And that’s basically how most systems look like.

In some of them, you have more data, less data. In all of them, you have automation, in some, connectivity. And if you think about what it means, it means that all the intelligence is left to the human user who needs to basically define rules and flows. Right? So the human user, if there’s anything smart about what you’re doing with your marketing technologies probably was defined by someone smart at your team or, you know, something that the team came up with.

But all the intelligence, all the intelligence is 100% injected by the human user. And I think that it does two things. One, it burdens the user. So, for some users, it’s not as easy. And second, it doesn’t leverage the capabilities of the human user to the best possible ability.

So the problems with this current approach, and you see it from the picture over there, just like in coding, in many ways, it becomes spaghetti codes. It becomes convoluted, it becomes something that’s hard to maintain or it’s not built to scale. And what do I mean when I say it’s not built to scale? So the first thing is we always talk about, you need to leverage every opportunity you’ve got to start a meaningful conversation with a customer.

So as people who look to build campaigns, we’re always looking for good conversation starters. But campaign volume and variety is very much limited to the capacity of manually-defined journeys. If you define a journey and you want to start five more conversations, you know, to either build a new journey or find the splits in the existing journey, and you say, “I need to split here, here, here, and here, and here.”

And, at a certain point, it gets to a moment where your hands are tied and it’s almost impossible to add on top of it and you get into the state of a deadlock. It’s not going to work. The second problem is that you have very high dependency in the designer of the journey. So we all know this, we’ve built this amazing plan in our marketing technology, and then, the person who built the plan leaves the company and moves on to another company.

And, as the managers of that thing, we’re stuck. It’s like, “Who’s going to run this? I don’t know what’s going on there.” Just like trying to figure out what happened in a code, in a piece of code that was built by one of the developers who left, it’s very, very difficult to maintain and to scale. And lastly, in most marketing plans, you have a ton of crossfire.

So messages, you know…customers get multiple messages from different campaigns. There’s a ton of crossfire because it’s really impossible to manage priority and exclusion. And moreover, what happens is we don’t really leverage the benefits of machine learning, or the human user that gives us all the intelligence is pretty limited.

I mean, humans are good at what humans are good and we can’t get a deep analysis from looking at multiple data points and crossing them across one another. It’s really impossible. So we don’t really leverage the benefits of machine learning, and eventually, what happens, we’re not saying the right thing at the right time via the right channel.

So the Holy Grail promise of saying to the customer the right thing at the right time via the right channel is not really happening. It happens to some degree, but mostly very little. So people would do a good win-back series, a good welcome series. They’re going to do that, but that’s the basics. Customers expect us to do this. We’re not going to excite them doing that anymore.

Okay. So a different approach is the innately intelligent approach, is an approach in which we’re going to combine the marketer and the machine. So marketers do what humans do best, while the system adds an extra layer of this innate intelligence using machine learning and AI. And please look at this, to my right, there’s a little bit of weighing between the two.

And if we look at the rule-based marketing, which is the way I see the current state, the marketer defines all the logic. Everything is done by the marketer. And in innately intelligent marketing, the marketer actually sets a framework. So the marketer uses his human creativity to say, “This is where you’re going to play.” Right? This is going to be where you’re going to play, and then, the machine will optimize by looking at multiple data points that the marketer cannot crunch in his head alone.

So think about the framework. So a great example of this framework and this notion of being innately intelligent is something that already happens in Optimove, the Self-Optimized Campaign, the SOC. And in the SOC, what we’re doing is we’re essentially running an A/B test, or in this case, an A/B/C/D test.

But think about what would happen in a normal scenario. The marketer would choose a population, let’s say, all of my active customers and all of my reactivated customers, and then, I’m going to choose my, in this case, four variations of some kind of a mix between message and channels.

So I have A/B/C/D, and A can be some kind of a blog, some kind of a piece of content, you know, deployed via Facebook and Google, and B can be some kind of a discount that I’m going to send via email and push. So it’s a mix of channel and message. But ultimately, what’s going to happen, I’m going to run this, then, the marketer is going to analyze the results and say, “Hey, here’s the winner.”

So A is the winner. And now, what you’re going to do, you’re probably going to deploy A for the entire population because that’s the winner. And that approach doesn’t leverage machine learning and doesn’t leverage the fact that there’s a big cost of generalization that we don’t necessarily need to pay. So what we’re doing with the SOC, we’re saying, “There is a winner per micro-segment. So we’re going to do this smart mix and match, not for the entire population, but we’re going to run optimization on multiple groups of customers, on smaller groups of customers.”

And, in every one of these squares here, you can see that each micro-segment, or each cluster, can have its own winner. So even the losing action would win a few of the micro-segments. And that’s a pretty powerful approach. If you think about what happens online when you search on Google and you see three ads, you’re not seeing the same three ads as I’m seeing because they’re essentially doing something very similar.

Right? So they’re showing, they’re adapting the ads to each micro-segment that they’re crunching on their own behind the scenes. And what’s beautiful here is that we do need the marketer to come up with A, B, and C, like the variance of the action. We do need the marketer to come up with the overall population. So the marketer is really crucial for experiment design.

The machine cannot design an experiment. This is something that humans do a lot better because we need to bring in the business understanding. We need to bring in our understanding of the world to design a very good experiment. Once we design that experiment, the optimization can happen by the machine.

This is the best connection and there’s no competition here. Because really, the human ability and, let’s say, machine learning ability is so vastly different that there’s never competition, which, sometimes, we like to see it like that in, you know, in articles and things that want to address our fears or things like that.

But it’s not even remotely true. So that’s the Self-Optimized Campaign, and today, we want to take it up a notch. So I’m announcing today the Self-Optimized Journey. It’s not yet in Optimove, so it’s not yet ready, but we’ve taken a huge step, meaning that in a lab, we’ve completed a lot of research and we now know exactly how to productize it into Optimove.

And I want to share with you today how it’s going to look like. So what is the SOJ, the Self-Optimized Journey? So, given a marketer-defined framework, the SOJ is an intelligent agent designed to encourage customer migration towards lifetime-value maximization. So there’s a lot of tion and in, but let’s talk about customer migration.

Customer migration is basically when customers move from different states. Right? So if I’m a new customer, and then I become a one-timer, and then I churn… Right? So I’m moving between a few states. Or, if I’m a new customer, then I become active, and then, VIP. And then, I change my affinity, so I was a VIP who liked these products, now I’m a VIP who liked these, and these, and these products.

So whenever there’s a big change in customer behavior, there is a migration happening, so they move between state to state. We want to manage that process and, moreover, we want to influence that process. As CRM marketers, our entire goal with every…sorry, with every message that we send is basically to change customer behavior. We want to influence this migration, we want to make sure that if something is at 40%, we want to take it down to 30%, if it’s better to take it down, because that’s a churn migration, or we want to take it up if it’s a positive migration.

So that’s what we want to do, obviously, but there’s so many alternatives and so many variations. So how can that become scaleable and manageable? So now, I’m going to use an illustration. And please, forgive me, I’m going to geek out a little bit, so I’m going to do a few slides of geeking out. So bear with me and be nice. But I’m going to illustrate the SOJ on the Optimove calendar.

So let’s think about a client that we have. His name is John Smith. And John Smith, tomorrow morning, is going to be eligible for three different customer states. John Smith is a new customer so, tomorrow morning, he’s going to be a new customer and it’s time for his welcome series.

As part of his welcome series, after seven days, there’s a message to be sent to John. It happens to be that John is also going to be 40 years old tomorrow morning. So we also have a birthday campaign waiting to happen. And it happens to be that because we’ve analyzed John’s behavioral patterns, everything he bought in his first purchase, and he bought items and he behaved in a way that we are almost sure this is a one and done type of a client.

He’s not going to come back. There’s a very high risk of churn with John Smith specifically. After seven days, we know to say that he’s probably not going to come back. So there’s three competing alternatives that basically want to message John. And think about it, whenever we say something, we’re not saying something else, so there’s always the alternative cost that’s happening.

And also, more importantly, there’s always the option of doing nothing, which is something that we typically discount as people, but doing nothing sometimes is a good strategy also in life. So we have the three options of messages and we have the option to do nothing. And think about an optimization process that does this analysis for John every day.

Every single day, we’re going to do it again and again and again, and analyze John. And what are we going to look at? We’re going to look at all of John’s transactional data. Right? So everything he ever purchased and transacted with. We’re going to look at his behavioral data in terms of browsing behavior on the web and mobile. We’re going to look at all the demographic data we have.

So if we know his gender, age, level of affluence, kids in household, country, geo, whatever. We’re going to look at the specific states. So how did my birthday, the overall birthday segments, all the different campaigns that ran for birthday, how are those doing?

How are the campaigns for risk of churn doing as a whole? And how has this specific campaign that I’m going to run tomorrow morning, how has that specific campaign been doing, not generally, but for customers like John? So putting all this analysis to place, I can basically predict, right, predict which one of these actions will be most valuable in terms of immediate uplift.

Right? So I’m going to see that on average probably, and as an expected value, because there’s a good chance that John’s not going to respond, but as an expected value, I’m going to get two dollars of uplift from the first campaign, three dollars and five dollars. So this is kind of like trying to define the priority of what John’s going to get using all the data and using machine learning.

However, I’m talking about the Self-Optimized Journey, so where is the journey? Right? There’s no journey here. The journey comes into effect because we realize that when we make a choice here, it’s not only about the immediate uplift that we’re going to see. It’s also about the fact that we’re going to set John on a specific path. So if John sees one of these messages, there’s a better chance of him being eligible to see something else in two days time or three days time.

So basically, there’s all of these different journeys with different probabilities of happening or not happening and each one of these journeys would change and would impact where John would be in one day time, two days time, three days time, or four day time. And, as according to that, what is he going to get or what is he going to be eligible to?

Because again, we’re going to take the same decision of trying to understand which message he should be getting. And there’s kind of like this route that’s going to happen and there’s the expected value of the route. And after we take into consideration the future journey that we’re going to put John on, all of a sudden, the optimization changes and maybe what I want to do is I want to choose to keep John on my normal welcome series.

And he’s going to get this, which will lead to another thing and 70%, and another thing, and, overall, his accumulated uplift will be $21, and there is this… And I’m just showing a few things for illustration, but, obviously, you guys understand that mathematically, this explodes, there’s multiple journeys and you need to kind of like always choose the best policy that will put John in the best-expected route that he can take.

So to sum this up, innately intelligent marketing is only possible when you combine human creativity and machine optimization. So again, this notion of machines doing the optimization after the marketer has defined the framework, has designed the experiment. We do the experiment design, the machines do the optimizations.

And going back to this notion of CDP, you cannot do this if you don’t have a strong data foundation. If you don’t have a rich customer database with loads of raw data and if you haven’t processed this data to make business sense, you’re not going to be able to achieve the self-optimized journeys or innately intelligent marketing in general. And lastly, I want to talk about something which is a little bit personal, but actually, all I’ve been saying now is one word.

I’ve been saying Optimove. So we have been obsessing over this one problem of what’s the best thing to do for every customer, every time? What’s the best marketing action for every customer, every time, or the best move? And that’s basically Optimove. So throughout the years, we’re just always kind of like taking a different approach at the problem.

As technology advances, as our resources and capabilities advance, we’re able to solve this one problem of truly becoming what we were meant to become, which is Optimove. Thank you very much everybody.