AI use cases in email marketing to consider in 2022

Rickey White 0:00
Then, who will present today on AI and a use case in email? So, welcome to the stage for

Faruk Aydin 0:13
it's very exciting to be here at this great learning and networking opportunity. I'm a big fan of email expert events, I attended almost all the events before all the editions. And I don't remember which one was it, I guess it was 2020. And Andrew asked us, guys, if I organized a kind of event, would you come over? And I was like, Yes, I'm being I'm going to be there. So I'm very happy to be here for two main reasons. The first one is sometimes people ask me, what do you do, I mean, those people who are not in email marketing, right, and I get so excited, and I tell them about my job, what I do like two minutes, and then they want to change the topic, and they don't talk to me anymore. So here, we have got about like 100, plus, email marketers, and we have even more people online watching us. The other reason is, well, I'm coming from Poland. So as you may guess, it is like minus seven Celsius over there. And there's like some smell. So we have this beautiful weather in Valencia. It's like heaven for me. However, those days when we had the conversation with Andrew, unfortunately, it was really tough times for email marketers. So we were hit by the pandemic, and we didn't know what to do people get, like kind of scared, email marketers, they were trying to figure out what kind of response they would give. And later we got at the beginning here, we got some kind of, you know, confusion. And later, we figured out some ways we adapted the situation. And I must say that. Yeah, so I must say that 2021 Actually, one more time prove that email marketing is not that. So a lot of brands, they, they saw that email channel is actually very important. And the brands, who defined the email channel, very critical for their marketing efforts are actually more than 40%. So we had a great year, I would say, and all these challenges we handled, we used our innovation, and we improvised sometimes. And we had pretty good times and at some point, we're like, okay, GDPR checked, and later pandemic, check, you know, we were victorious. And we got like pretty, you know, comfortable for a while. And I think Apple hurt us. And they brought MPV Well, still, I believe that 2022 will be a great year for email marketing. I believe that more budgets will be allocated for email marketing, more email, marketers will be hired, and more money will flow into agencies and freelancers. Today, I would like to point out a couple of challenges that I believe we may see next year, obviously, there will be more of them, I hope not as bad as the pandemic or anything like repeating. And later on, I will have some suggestions from my point of view, how to handle them. And later, we'll talk about a couple of AI features to help us with that. And finally, for those who are interested, I will talk about my framework about how to get started with AI. I'm aiming for about like a 20 minutes session here. So at the admin, we can have five minutes for q&a session if we can make it. So let's get started. So before I get my flight to Valencia, it was Friday,

I saw the news. And in United States the inflation rate was published. And with a 6.8% inflation rate, actually, we have the highest prices in the last four decades, basically since I was born. And the case here is actually the inflation rate even surpassed the growth on average wage. So for that reason, we will see that people will think twice when they would like to buy some products or services, and they will check their wallets. In every penny they spent they will be really careful with that. So it will be a little bit of a different year. And at first you may think well, is it coming a kind of a crisis or suddenly this not What I'm trying to say, I don't mean that there will be a recession, I don't mean that will be a crisis, I just want to mention that it is going to be a big impact on customer behavior. And as you can see here, the projection by EA marketer Tabak in this October 2021, the online sales will even rise much more than retail over there. So, I believe that this online sales will go, you know, you'll have a better grant. However, it will be harder for brands to find new customers convince those new customers or retain the current monster they have. And you may think maybe, okay, so this is like a kind of a budget issue. So we can just drop some prices, we can offer something, you know, lower prices to them. But unfortunately, as you know, many brands are struggling right now with supply chain issues. So they may not be able to cut the prices. So for that reason, that's not the way to go. And also, consumers, they're not really driven by the price discounts, they don't make rational decisions, when they see like price tag that they actually value the experience. So their purchase behavior is powered by the the experience that they get from the brand. So, as we talked about the price action, it's not a barrier at all. And according to some studies, consumers are willing to pay even like a 20% premium, if they receive a good experience a good shopping experience. So for that reason, as email marketers, we have another thing to compete next year, not only the other brands, or ourselves, but the macro economical circumstances. And we need to provide a good inbox experience for our subscribers. So I know that you guys might be able to get hungry before the lunch break. So I don't want to spike your insulin. But there's a very good example from Chipotle. Recently, they announced double digit sales growth. And they just focused on delivering the value of experience to their customers. So what they did was, they changed their delivery model and the pricing structure, they offered customization options for their customers, and they put some of those options online on and they added an additional 17% premium on top of that, and people were actually buying that. So you know, that's a great example, in my opinion, when it comes to, you know, offering the value, not, you know, changing the price, even increasing the price. So the next one when we're talking about, you know, premium experience, the next one, the next challenge that we have is the male privacy protection, as you may have noticed, subscribers are getting more informed and aware of their privacy. And Apple actually just listened to its users, you know, they, I believe they did the right thing. And in my opinion, the other mailbox providers will be following Apple's move. So in this case, I expect brands to embrace privacy even more and use it as a kind of a tool for competition, they want to stand out, because they offer that privacy.

And this will be maybe a little bit harder for brands to tackle that they would like to maybe So focus on creating some dynamic, interactive email. So we saw some presentations about the dynamic emails, and for email, for instance, a great suggestion for that. And they want to use those dynamic and interactive emails to gather those customer data, and feed those data that they have to the customer profiles. And they would like to make those profiles richer. And the main focus here is looking for the engagement. And they may use some kind of, you know, tactics like innovative tactics, like customer false or gamification in their emails. And the last challenge is about data. So as marketers we heavily relied on third party cookies for years, and it will be a little bit harder next year to have relevant and personalized experience for our customers because Google is going to phase out the third party cookies and as email marketers, we I know that most of you already started you know finding some solutions for that. And what I noticed is, brands are shifting towards zero party and first party data So they would like to get some data about the preferences of their customers. And they'd like to figure some more more ways of engagement and having those interests and making different profiles research. So at the end of today, it all boils down to gathering quality data from your subscribers, managing that data, and providing relevant and personalized email experience for our subscribers, and somehow making the whole thing in a way automated, because you'd like to allocate the time for creativity, for strategic, or for something more, like revenue generating ideas. So that's why I think AI can step in here. And we can talk about some use cases about AI. In the previous inbox, x four editions, I talked about, like many different cases from the acquisition all the way to the end of the funnel. But today, I would like to focus on just three categories. So I'll start with who to send to and later what to sign. And finally, we'll talk about when to send those emails. So modern marketers mostly use right now one to one communication model. So I actually in at this event, I haven't met many people who are like sending bulk, emails, bulk segmentation, and they're not doing this kind of stuff, they really focus on providing this like one to one experience. And predictive analyzes is one of these models in artificial intelligence that you can use is actually very good for that. So it is possible for us to predict the next action that subscribers can take. And also it is possible for us to see insights that AI gives gives us and later we can, you know, use those insights and take an action to provide that experience to our subscribers. All of these apps, absolutely, they focus on the past behaviors, you know, all the data that we got about subscribers habits, their interest, and so on. The next one is optimizing the customer journey. So all the predict all those analyzes that you get, get all the predictions that we get, we can use them to optimize the customer journey. And we can know at what stage what kind of messages we can send. So there are like two questions we need to ask here. And AI can provide the answers for so what kind of content really affects your customers at which stage of the customer journey? And at which stage are your customers are leaving so AI can answer those questions. AI can tell you, if there's any churn detection, for example of the if they already they're inactive. So it is possible to give all these details or insights to another model that can help us is customer lifetime value prediction.

So, to predict the customer lifetime value, you need to have historical data about your subscribers or customers and their past purchases. And if you have the timestamp and also the order value, it will be even better. So there are like three very basic stages of training these kinds of models. At the beginning, you need to introduce the data to the AI system. So there will be like a learning stage. And later on, you need to have this holdout period. So it is the place it is the stage where we train the AI model. And after the validation after testing, we can apply the model for the next customers and get the insights. So if you know for instance, your EPC or if you know your average order value, you can see let's say in the next quarter or next three quarters, how much revenue you can get from email channel. So let's talk about what to send. The most popular content production algorithm in general is recommendations and recommendations. Models, they help us to answer two questions. First of all, what type of recommendations deems that you should be using and another one is what's caught on peace or what products that you should be showing to your customers in the message. So when it comes to the recommendation themes, obviously there are various recommendation themes that you can use. It depends on where they are at the customer journey. So the first one here we can look at the new subscribers so you acquire a new subscriber. And in this case, you'll it's a great time for you to tell them that you have a good contact. It's a great opportunity for way to build a long lasting relationship. And also you don't have enough data to send them one to one communication. So for that reason, you need to get some intelligence from them. So what you do is you can maybe send them most popular or trending content, so you see their actions and get the feedback from them enrich that profile. For active subscribers, you have enough data to use the recommendations because the subscribers are the customers they already engaged with you in the past. So that's a great time to use a command is now finally, if you notice any kind of churn or inactivity, we can send some feedback messages. So regarding the second question that we're asking to AI, like, what we need to show in the message, it is fairly easy if the customer already purchased the product in the past, so you can use purchase history and demographics, put them into the recommendation algorithm, and the algorithm gives you likelihood score of purchase. And then according to those scores, the product is feeding the message body and fast is itself a place over there. But if the customer didn't purchase, in this case, we need more data. So what we need to do is we need to use purchase data, and also some marketing response. For instance, if that customer added that product into the wish list, if it was anytime in the shopping cart, or if the customer clicked on that, and never purchase. So if you need to know all those things, later on product attributes, and seasonality is also added. And algorithm again gives you the likelihood this case is a little bit harder, it requires a little bit training. But it is possible especially for new products, let's say you have get a new product that you can start to commanding. Yeah, so this is a liquor company from Australia. And they have about like 2000 different email templates to send out. And as you can see, there are like four different messages, or templates over here. So the first one is, you know, if you like it just keep going on, you can why you don't need to change that. A second one is like suggesting some change. So sipping suggestions for is like more personalized. And later on the AI algorithms suggest the trends or tells the other people is similar to your profile, they liked this thing. So it's also another encouragement. And finally, this is a discovery suggestion for a discovery.

So when we train train the recommendation models, there are two ways to start. The first one is called Start, the other one is warm start. So in the cold start, we sent those random messages to everyone. And then we get the feedback from them. It's a bit harder at the very beginning. But we noticed like based on our experience, it is much better because in the warm start, we already have some information about subscribers about customers, and we base those emails on initial data. But later on, we noticed that the algorithm keeps in this very narrow corridor and you cannot really get a broader data. So I would suggest if you think long term about your products and revenue, cold start is much better. So as email marketers, we love to test our messages before, you know we send it and multivariate testing is a great way of doing that. Because a bit in the A B testing, as you know that you can, let's say you want to test subject lines. So subject line, there are two subject lines that you have, the rest of the message remains the same. So you don't really know what really catches the subscriber or customer. So in this case, multivariate testing can help you to test subject line preheader content or recommendation CTA all of them at once. The only drawback here is that you need to have a larger sample size. So when so you may ask me, okay, so where is AI here, you know, is already very high word, right? So we can look at the two different examples of the traditional AV testing. Let's say you have three different tests that you're sending out. Once the winning message reaches to the statistical significance, the messages standard, but you lose all those engagement, all those clicks over there more in multivariate testing. It happens in a way that gradually and automatically the Vinick message is sent so you don't leave the money on the table. So the last one, you guys are all very aware with send time optimization. So everybody has of course their own preferences when they would like to engage their emails, and so far centime optimization as you know very well on email vendors, it is open triggered, but after Apple's move is MPP I believe vendors will come up with some different metrics, more focused on engagement, more focused on maybe different channels or conversion or clicks. So what we do with AI, we get the intelligence from different channels from website from app, and also from email. So we put them into a pool, and we give some kind of like time slot to send those emails, it is a great way to optimize because as humans, we cannot know when to send those emails to our subscribers at very large scale. So this is a very simple example. If you see it, you can see the frequency, the email frequency is really, you know, there's a sweet spot over there. So it's really hard to find it out. And Michigan's actually can't do that. So over there, you can reach the optimal frequency. And if you send too many emails, you know, your subscribers, they may unsubscribe, it may overwhelm them. But if you send too few emails, in this case, what happens is you they will forget about your brand. So the measures will find that optimal frequency. And another case study here we have is about the operations and how optimized standing tabs can help. Let's say you want to launch a campaign, and you have got your customer team or call center, getting all the calls, but there's a limited capacity. So what happens, you know, are there bad client experience or you know, old, retired employees, so we don't want to recommend that. So what happens is you can use the optimization, and the messages are sent according to the times that your customers are engaged.

So, you know, I've got a bonus use case over here, everybody's favorite topic deliverability, I would like to show you one of the algorithms that we use for deliberative. So whenever we get the data from our customers, we put them into the system, and our system is a customer like profile centric system. So all the attributes are were there. And we have got a reinforcement learning model is one of the algorithms that we use, and sending resources details are posted over their ISP, Santa Cruz volumes per ISP, and so on. And the modem may decide not to send those emails because in some cases, maybe we don't have a good IP reputation. So what happens is, we may want to just cancel that sent out after we fix the problem, if we decide to send the emails, so we need to exchange all this information with ISP, so we get all the details. And let's say the email is deployed. In this case, there are a couple of options. Email is blocked. For instance, when it is blocked, we paneled, like we paralyze the model, in this case, and the same happens when email is placed in the spam folder. And later on. In the good case, when emails actually in the inbox, we have two different options over there. The first one is, well, you may learn the theory, but there's no engagement. And the second option is there is engagement. So this is the one that we reward the model. And we try to follow all these steps. And we will start to teach the model to go all these ways. So this is how we work with that. And finally, I would like to talk about the Getting Started with email, AI in email marketing. So it's not like you will just get an AI solution. And overnight you will have wonders. So you definitely need to figure out a couple of things. And AI is just one of those tactics, you need to have a strong email marketing strategy and unity implemented, and you can use AI and other stuff. So the most important thing at the very beginning is identifying your origin problems, what you'd like to solve, and later, you'd like to out have the projections for the output. So what makes you successful, what makes this model or AI implementation successful, and I suggest starting small and then data later on, you can scale it up. And finally, it is important to integrate all your data sources to the AI system, it will help you to learn and it's not going to happen overnight. As I said, You need to give it a time because AI is a self learning system. So that's the time allocated for me today guys. I'll be around here at the event and also I believe, our viewers, they can send us some messages as well. I'm happy to interact, you can get connected with me over there. And thank you so much for your time today. Have a great rest of the show.