AI and the customer journey: How brands predict what you’ll do

Photo by Markus Spiske on Unsplash

If you’ve been around marketing, you’ll have heard the phrase “customer journey.” It’s what people do when they buy … or don’t buy. One marketer I talked to once told me he’d mapped more than 800 separate journeys customer took on the path to buying his product (!!).

Naturally, marketers want to optimize that trip, and Adobe has developed an AI system that finds out where those journeys break. Theoretically, that will help marketers optimized customer journeys, help brands sell more, and I guess … ensure we spend more.

In this edition of TechFirst with John Koetsier, we chat with Steve Hammond, a director at Adobe Experience Cloud, about an “Adobe sneak,” a test project with AI and customer journeys.

Subscribe to the TechFirst podcast

Watch: AI and the customer journey

Subscribe to my YouTube channel so you’ll get notified when I go live with future guests, or see the videos later.

Read: AI and the customer journey

John Koetsier: Can marketers use AI to predict what you’re going to do and whether you’re going to buy?

Welcome to TechFirst with John Koetsier. If you’ve been around marketing, you will have heard of something called the customer journey. That’s what people do when they buy something, or when they don’t buy something, that’s a journey that didn’t work. Naturally, marketers want to optimize that trip. And Adobe has developed an AI system that finds out where those journeys break.

To learn more, we’re joined by Steve Hammond, who’s a director at Adobe Experience Cloud. Steve, welcome! 

Steve Hammond: Thank you so much. Glad to be here, I appreciate the opportunity.

John Koetsier: Excellent, great to have you. Let’s start here: can AI predict what people are going to do in a customer journey flow? 

Steve Hammond: It absolutely can. It needs certain points of data input to help with that, but it is a big part of how we have interactions across journeys today. 

John Koetsier: Interesting. So I want to dig deeper into what your AI is doing, but let’s start with talking about kind of the degree of difficulty, the amount of data that it needs. I remember talking to one person, a marketer, sometime and he said he had mapped out like 800 different customer journeys that people were using. What’s the level of diversity we’re talking about?

How many different possible routes are there for people to buy a pair of shoes, or book a flight, or something like that?

Steve Hammond: Yeah, that’s a good question.

There are usually thousands, and usually that journey is based on the individual. Each of us has, you know, some special need.

We are, we’re coming through from a given browser technology, or we’re looking for a certain type of product, or we might be going down the road of going and first checking on ratings and reviews. Maybe we go check out customer service support. But there are any number, infinite number, of possible journeys that someone could take.

So it’s important that AI can help with that because the real challenge is that when you try to get a team of people to create a path, the ideal path for an individual, it’s never right because you can’t create that infinite number of paths.

You really need to be able to have some level of automation and machine learning to be able to help support that.

John Koetsier: Exactly exactly. I mean, if you’ve got sort of thousands of journeys and multiple data points in there, there’s probably, like you said, an infinite number of ways somebody could go through that. You can’t understand all those, you can’t diagnose all those, you can’t understand the challenges with those either.

So, let’s dive into your project that you’re working on from an AI perspective. What was your goal? What were you trying to do? 

Steve Hammond: Yeah, absolutely. So I think to start off with, it’s important to understand the scope of this kind of a project. This project is part of a program that we have at Adobe called Sneaks, and Adobe has this program and it’s really meant to inspire the engineers, and the innovators, and the product people across the business to come up with unique solutions to challenges. 

John Koetsier: Yep.

Steve Hammond: And we’ve been doing it now for nine years in the enterprise space and the digital experience business. And this being our ninth year, one of the projects that came out and that we revealed in March as part of Sneaks, was this idea that you can take the understanding of analytics and the paths that people are taking, and then match that back to the interaction points that people have, and then create kind of the ideal journey or path orchestrating that across different experiences.

… you can take the understanding of analytics and the paths that people are taking, and then match that back to the interaction points that people have, and then create kind of the ideal journey

And that was done, and again, kind of going back to how we got here, is there’s a set of engineers who have been researching and looking into this particular challenge for quite a while. We have different versions of, or flavors of, you know, kind of path analysis and journey orchestration, but putting the two things together was an interesting challenge.

And so this team worked on it and then they presented it as an option for us to showcase. And we did that, and so we were able to showcase the Sneak or the sneak peek into this concept technology that we’ve codenamed “Bon Voyage.” And it’s really a, it’s a way to look at the path, the journey that you’re trying to take and make sure that it’s a positive and good journey. 

John Koetsier: Excellent. What’s the process? I mean, just to dig a little deeper there, you must get a lot of different sort of proposals, almost like, you know, little pet projects or something like that. And what’s the process by which you decide, hey, you know what, that meets the criteria, we’re moving forward with that. Is there a certain level of success it needs to have?

Steve Hammond: It does, yeah. So we generally evaluate on what’s the value to the customer, how innovative and unique is the idea. And then we also have to add in a couple interesting elements, like, I mean, obviously feasibility — is it a functioning prototype that we can show and showcase? And then the thing that I was going to add there, is it has to be somewhat entertaining to be able to show people, you have to have a story behind it. 

John Koetsier: Yes.

Steve Hammond: There’s a lot of great technologies and a lot of really interesting ideas, but we want to be able to put forth things that people can really grasp and see, oh, I see how that could contribute to the value of my business.

John Koetsier: Sure. 

Steve Hammond: We do get hundreds of submissions and we go through this evaluative process, and it’s fun. It’s fun for the people who are a part of that every year. You know, my teams that work with the groups on this, but it’s also really fun for the engineers who have an idea and can put it out there and put it in front of a large audience. It’s a pretty unique opportunity. 

John Koetsier: So in this particular case, how much data are we looking at? How much data are you feeding the machine? 

Steve Hammond: Yeah. You know, the level of accuracy gets better with the level of data you put through the system.

So typically any kind of really advanced AI type systems are going to need millions of rows of data to be able to start to evaluate patterns and behavior.

But once you start to get that amount of data through the system, the system can start to then evaluate those patterns of behavior and look for future combinations and things that will work better. So once you kind of feed the model with a million or so rows of data — I mean, and that can take depending on the site, you know, someone could have that done in hours, other sites might take weeks. 

John Koetsier: Yes.

Steve Hammond: But generally speaking, you need about that kind of volume to feed the model and then it can start using the ongoing testing and prediction to be able to determine what other variations to consider. And then the variations at that point become very real time. 

John Koetsier: Yeah.

Steve Hammond: It can start to model based on immediate interactions, immediate success, and those kinds of things. 

John Koetsier: Interesting question comes to mind, because I was talking to the CEO of some weeks ago, and they’re building self-driving technology using AI, and they inserted what they call “priors” into their models. So something just to tell the machine that it happens.

So one example, is like object permanence, that when a car is obscured by a truck it doesn’t disappear, it doesn’t not exist until you see it again, right? 

Steve Hammond: Right, yeah. 

John Koetsier: It’s still there. Do you have the capability to put something like that into the system? Or are you worried about if you do that, you might tell it something that might actually disrupt its potential to find something new later on.

Steve Hammond: Yeah. The concept of priors is a really good, it’s a good way of referring to it. I’ve kind of referred to it in the past as like a “narrow AI,” but prior is a good word for it. I like that. So we have a huge library of really well-defined and really well-baked AI that in some ways can exist as a bit of a prior.

So an example of that would be, we have within our Photoshop application the ability to be able to evaluate all the pixels in an image, and then based on the pixels in the image we can start to look for patterns and we can obviously evaluate is this a scenic picture? Is it a city? Is it, you know, a dog, an animal, person, whatever it is.

You can start to look at those patterns of pixels and then use that to be able to make decisions. Like a good example of a decision to be made from that is a technology inside of Photoshop called Content-Aware Fill, where you can select an area and say ‘replace this with’ and it will automatically generate the content there. As a concept of a prior, this works really well because when you’re evaluating what kind of content to put in front of somebody, in a journey or an orchestration, you want to be able to look and see, you know, what is this content?

And so you can quickly evaluate based on this already established AI. 

John Koetsier: Yeah. 

Steve Hammond: You know, what is in the image? You can see what the focus area is. You can see where the text might be. You can do all these things. So for example, you have to crop it to say that for this particular journey that someone might be on, and maybe they switched to a mobile phone and it’s a smaller screen, so you need to be able to show a smaller rendering of an image. We can take into consideration the pixel patterns and say, ‘Don’t crop the important part of the image.’

John Koetsier: Interesting.

Steve Hammond: You know, leave the focus area as the focus, and then that way you’ve combined data for what to put in front of somebody with an image that’s appropriately sized and scoped for the situation.

John Koetsier: Very, very interesting, and not something I would have thought of as well, just using AI to make sure the pictures look right in a different environment. Let’s talk about some of the common challenges with customer journeys. What are some of the things that you’re finding are typically interrupting them?

Steve Hammond: Yeah. Well, I know it’s, I think we all experience some of these things and oftentimes it can be a series of going and showing interest on a website. Maybe even go through and say, ‘I want to buy this product,’ then you put it in your shopping cart and maybe you have even logged in, but then maybe a half hour later you get some kind of an email notification for something completely different or that’s really disconnected from the previous experience. 

John Koetsier: Yes. 

Steve Hammond: And that is a concept of a journey that needs to be better connected that hasn’t been. And so having that real-time understanding of what just happened right now, you know, the historical information is important, but what just happened right now. So if I do send you an email right now, it’s relevant and in the moment. That’s a really common example.

Another one would be maybe across interactions that are not always just technology, but you know, with humans as well, where you’ve been engaging with a brand for a specific product. Maybe you have to return that product, you call into a call center and now even though you may have a history with that company in the online forum, now you call into a call center and everything’s like starting from fresh, you know?

And it’s no fun. 

John Koetsier: Nobody has that experience. That’s never happened to anybody, come on, haha. 

Steve Hammond: Yeah, who are you? What are you looking for? So those are good examples, I think, of very common everyday challenges that can be overcome with proper journey orchestration. 

John Koetsier: It’s kind of interesting actually, because if you think about it, customer journeys can be something that can last 30 seconds, an instant buy with a one-click buying process. It can also be something that can last months and years, not just for products that are, you know, the major purchases of our lives, like automobiles or something like that.

But I was browsing Facebook the other day and I see this pair of shoes and it’s amazing, it’s incredible. I love it, it’s really cool, and the price point is just right, and guess what? I don’t need them right now. And so I kind of have this policy of maybe not buying something that I don’t need just ’cause I like it. I mean I don’t know, maybe that’s crazy to have that idea. 

Steve Hammond: It’s pretty wise. 

John Koetsier: Haha, exactly. But you know, I kind of want that remind-me-in-five-months button, and that’s maybe a very specific customer journey that nobody’s figured out yet, but maybe your AI will.

Steve Hammond:  I mean, just to look at that example for a second.

John Koetsier: Yeah.

Steve Hammond: I mean that is absolutely a perfect case for really good, a good relationship with the customer. I mean, if you constantly go back to that site you’ve obviously been sharing your information with that brand and you can have an opportunity for that kind of reminder.

There are a lot of other sites that struggle to have a personalized individual relationship with their customers because their site isn’t set up in a way where they have that first-party data and the user saying, ‘I’m willing to share my information with you,’ and in a way where you can then say as a brand, ‘Okay, thank you for sharing your information. I’m going to try to make a better experience for you next time you come back.’

John Koetsier: Yeah.

Steve Hammond:  By, you know, privacy-controlled shared data, in a way that is mutually beneficial. But yes, I think given the fact that if you had that underlying data structure in place, you could absolutely solve that problem with a good journey orchestration.

John Koetsier: Interesting, interesting. So here’s another example, I was on the other day, I bought some shirts — shockingly, we’re not in the stores very much these days — so I came to the site and yeah, I signed up for the newsletter because it said there was 15% off, so what the heck, I did that. Then I created an account like immediately after, next click type of thing, still asks for my email address again.

Is your software going to kind of detect little issues like that? It’s a bit of friction in the customer journey, right? It’s not a big deal, I just entered it again, but it’s kind of like, I just told you that, don’t you know that? Come on. Is your software going to detect those sources of friction and kind of fix them? 

Steve Hammond: It can. I think that what you described is primarily an implementation challenge in terms of recognizing the steps of a journey and then making sure you have that continuity between the steps and the journey.

But to your point, I mean, it’s that foundation of understanding who the individual is and then carrying that through the experience. And so I would say that an AI-based journey orchestration can absolutely help with that.

If you were applying AI to go back and look for the best possible path then it could do that, if you could see the breakdown. For example, why did someone not continue on to the next step in this process? And then you go back and, you know, you were asking them twice for their email address when you should still have that. 

John Koetsier: Yeah. 

Steve Hammond: But that’s also, that’s a combination of AI supporting a good implementation, but then also what we talked about earlier, is that idea of really understanding the individual based on a common data platform.

Having that ability to be able to know your customer through every stage consistently and not having that be, you know, this system knows the individual this way, this system knows the individual this way, and you can’t put the two together.

John Koetsier: Yeah.

Steve Hammond: So having a common data structure is a critical step in being able to solve that problem. 

John Koetsier: Yeah. Talk a little bit about the data that informs the insights that you’re generating. Are you using aggregated data from previous visitors and shoppers and buyers? Also some insights that are live from what people are doing right now? And is there third-party data involved as well? 

Steve Hammond: Yeah, so this is a really important question, because our data is a definition of who we are in an online experience. And so one of the first things that we did when we were building these technologies out was to ensure that we allowed for and really provided the best possible way to be able to maintain the highest possible privacy standards.

John Koetsier: Mm-hmm.

Steve Hammond: So that if we are looking at historical information, that that historical information is information that you as an individual have chosen to share with the brand.

John Koetsier: Right. 

Steve Hammond: And you have the right to be able to retract if you want to. And so that first-party is what we call that information, your information shared with the brand.

First-party, that’s a critical piece to be able to help make that work. But outside of that, you know, you asked about aggregated data or third-party data, that can also be important for certain brands.

nd if an individual opts in to sharing their information and making that available, and they feel comfortable doing that, then a brand can leverage that to be able to help create a connection between, for example, an ad that’s produced or shared in a site that’s not hosted by that brand. 

John Koetsier: Mm-hmm.

Steve Hammond: For example, if it happens to be on Google, or Facebook, or Instagram, or whatever, and then that individual clicks through from that back to the site, that continuity kind of depends on the ability for that sharing of information across those sites.

And so if the individual allows for that, then that can be used to create a better experience.

John Koetsier: Right, right. And I mean, we’re going through a massive change in the marketing and advertising world right now, right? I mean, the latest of course is Apple killing the IDFA, which is a mobile tracking identifier in iOS 14 coming out in September. They’re not killing it, but they’re really deprecating it, making it harder to get.

And so that’s kind of, I guess an economist would say a secular move, right? A move across the entire industry, all industries really, towards more privacy. How’s that impacting your AI, the amount of data that you can collect for that? Is it making marketing harder? 

Steve Hammond: I think it’s an important step, you know, giving the user control of their data is important. And so, and we support that from Adobe’s perspective. The idea of the ID for advertising on the iOS 14, as I understand it — I’m not an expert on this — but my understanding of it is that it’s an opt-in option per application. 

John Koetsier: Yeah.

Steve Hammond: So it’s one extra step, which means that as a user, I get to choose if my ads can be shared through from a website into a mobile application, for example.

Now what that means for modeling and for AI, is that in the case where a brand is working with an individual in a first-party data scenario, the individual has said, ‘Here’s my information, I trust you as a brand to share this information.’

I create a login, whatever it might be, that information can be fully AI driven 100%. Also within the application, interactions between even anonymized information that completely abstracts the individual can be used to create the best possible experience within the application. 

John Koetsier: Mm-hmm.

Steve Hammond: Across applications, or across domains, across experiences, that opt-in for advertising plays a role in terms of whether you can have continuity between seeing an ad offsite and experiencing something related to that in an application. 

John Koetsier: Yes. 

Steve Hammond: But again, if the individual has a close relationship with a brand that’s trusted, then there are other ways by which that brand can create that continuity across channels. So it will have an impact. We don’t know exactly what that impact will look like.

We absolutely support the importance of privacy of individual data, and our systems are designed to be able to provide the best possible experience whether that individual opts in or not, given what information they want to share. 

John Koetsier: Right, right. That makes sense. Do you anticipate, and I know that you’re more on the marketing side than the ad tech or advertising side, but do you anticipate a move towards more contextual versus behavioral targeting?

Steve Hammond: Yeah. You know, it’s an interesting mix, I think. Because context is critical, you know, that’s what describes what you are doing right now. And it gives, it makes it far more relevant if you have context. Behavior is obviously important as well, and I think they’re very related.

John Koetsier: Yes.

Steve Hammond: You have behavior influence context as well, but oftentimes when you think of context, you think of environmental data like what time of day is it, what’s your IP, what’s your location, for example. And what part of the application or site are you in? Are you in a customer support section? If you’re in a customer support section, maybe you’re filing a, you know, maybe a grievance against a particular product or whatever it might be, you don’t want to then ask them to advocate for your product.

John Koetsier: Yes.

Steve Hammond: So, you know, that context can become critically important. So, I think they go hand in hand. I don’t think you could ever have a situation where an individual’s behavior doesn’t somehow play a role in the context.

John Koetsier: Mm-hmm.

Steve Hammond:  But I think that you can’t have context without behavior. So …

John Koetsier: Yeah, yeah.

Steve Hammond:  I think they go hand in hand.

John Koetsier: That is interesting really. I mean, because for somebody if they were just going on contextual clues, might target me with that pair of shoes. But if they were the brand and I had a first-party relationship with the brand, and they knew, you know what, this guy just bought those shoes like 30 days ago, or the last model, or something like that, and he doesn’t buy shoes more often than once every, I dunno, 18 months or 12 months or something like that, then they might know okay, he’s just, you know, he’s just window shopping.

It’s no big deal or something like that. But if it’s just — so that would be behavioral with contextual — but if it was just contextual, well, hey, looking at the shoes, obviously he wants to buy them, you know, here’s the offer, right? So it changes how you react as a brand. 

Steve Hammond: It does, and also can save you a lot of money if you do it properly.

Because if you keep putting that ad in front of somebody over and over again, and like you said, you just bought it or something very similar. I had this, I just recently bought a car, and it’s funny because I went to a website to look at the car options and I ended up buying the car.

And still it’s been like two months now, I still see ads for even that exact same car, and I just keep thinking how much of a waste of money is that for that poor brand? 

John Koetsier: Yes, yes, yes, yes. And we have such a fractured ecosystem for that right now, honestly, with so many different, you know, marketing separate from advertising, and customer database is still separated from all that stuff. It’s really challenging to do. I once had the same thing with a pair of boots that I bought and literally saw the ads for, you know, literally three, four months thereafter.

And here’s the negative thing. Here’s the thing, we can laugh about that and it is kind of funny in a sense, but A: there’s some wasted money, but B: as a brand, you kind of look dumb.

Steve Hammond: Yeah, exactly. 

John Koetsier: You kind of look dumb.

Steve Hammond: That’s right.

John Koetsier: Because if somebody looks at that, especially a consumer who doesn’t know all the intricacies of how ad tech works or something like that, and you see that [and you think] I bought that car, don’t you know? I bought that car two months ago, how do you not know that and why are you still targeting me?

And by the way, we’ll lose suppression lists in iOS 14 because of IDFA, so …

Steve Hammond: Ah interesting.

John Koetsier: We may get more of that. 

Steve Hammond: Yeah, that’s interesting. But yeah, I mean, the key there in this specific example of an ad, is to be able to understand that behavior at the point of purchase, and sometimes that’s hard to connect. 

John Koetsier: Yes.

Steve Hammond: You know, an example of buying a car, the dealership that I went to is a used car dealership. And I don’t know that they have the sophistication to be able to go back and link my ad to the fact that I made a purchase. 

John Koetsier: Yeah.

Steve Hammond: And so that becomes a really challenging thing in certain situations. 

John Koetsier: Exactly, B2B, B2B2C, B2C2B … there’s all these challenges, right? The dealership, the data going to the brand, a third-party agency running it, it’s challenging. 

Steve Hammond: Yeah.

John Koetsier: Absolutely. Well, Steve, I want to thank you for this time and you know what, I’m going to close, and I think I’m going to get the video that I didn’t get at the beginning.

Steve Hammond: Nice.

John Koetsier: See if that works. 

Steve Hammond: Hope it works.

John Koetsier: Thank you so much for the time. 

Steve Hammond: You’re welcome. Thank you so much. 

John Koetsier: Excellent. Well, hey, thank you for joining us on TechFirst, really appreciate it. My name is John Koetsier. I appreciate you being along for the ride. Whatever platform you’re watching on, hey, like, subscribe, share, comment, all the above. If you’re on the podcast and you like this, please rate it, review it, that’d be a massive help. Until next time, this is John Koetsier with TechFirst.