Time to recalibrate: How AI is transforming Email Marketing

Faruk Aydin 0:34
Hello everyone, this is Faruk , I'm so happy to join you here at Inbox Expo today. Thanks to Nely and Andrew to make it possible for us, we can at least meet virtually and I find this event as a great opportunity to learn from the experts and network with the fellow email marketers. For those, I have the chance to meet for the first time, maybe a little introduction. I've been in the b2b sales world for a decade now. But I found myself in email marketing about six years ago. In recent years, I tried to understand AI use cases in marketing and share it with our community. on personal and I live in Gdansk in northern Poland. I love scuba diving, and I can't wait for this lockdown to be over to have my next diving trip. at my company, i h we make AI accessible for email marketers, you can ping me at our virtual booth after the session to learn more. Some people may think that AI is a kind of futuristic concept, but we are interacting with it countless times. Even without noticing it. It decides if your incoming emails are going to the spam folder. If you get a loan from your bank, or if that guy on your dating app is a good match for you. All those things are not magic. They are all powered by AI technologies, such as machine learning, computer vision, deep learning, speech, or image recognition, and more. Today, I'll refer AI as an umbrella term for those technologies. When you do a Google search for artificial intelligence, you will find many different definitions. Usually, I like to talk about AI as the technologies and algorithms to make machines smart AI stays at the highest level. And what makes machine smart is machine learning, which is basically computers figuring out things from data without being explicitly programmed. And there is a further subset of algorithms or techniques called Deep Learning, which enables computers to solve more complex problems. If you look at AI on a global scale, reports from the most reputable business organisations all agree that the economic impact of AI is mind boggling. PwC research shows that the global GDP could be up to 14% higher in 2030. As a result of AI. This is equivalent of an additional $16 trillion, which makes AI the biggest commercial opportunity in today's fast changing economy. The covid 19 pandemic has accelerated AI powered digital transformation across different businesses. In 2021 alone, Gartner projects AI augmentation to enable 6.2 billion hours of productivity, which means saving more than 700,000 years. This is exactly the reason why I named this session as time to recalibrate. on marketing and sales. The annual impact of AI is estimated to be around $6 trillion. In today's marketing, the most applicable component of AI is machine learning. What makes machine learning stand out and useful for us is that you no longer need to programme every single thing but it can learn over time is it faced with more data, more cases and more examples. Ai use in marketing is relatively new compared to some others. industries such as financial services, high tech or production industry production sector. In the marketing technology landscape today when we have a look at it, you'll see about like 8000 companies, about 1000 of them claim to solve marketer's problems using AI. As you can see more money and more investment is flowing into AI solutions. In line with the growth in computing power.

AI is now becoming increasingly accessible to the average marketer. According to the 2021 ai index report, which was published just three weeks ago. across all industries, ai adoption in marketing and sales is in top three among different business functions. Apparently, marketers enjoy AI capabilities when it comes to the revenues. McKinsey survey shows that marketing and sales professionals benefit the highest average revenue increase through AI adoption. Here there are some of the AI use cases in marketing. I'm sure at least one of them is already implemented in your marketing programme. as few months, we have certain needs and limitations. We work eight hours a day, while machines can process and optimise continuously. Also, we can't even wake up without drinking a cup of coffee in the morning. Moreover, we don't like making decisions each time we make a decision. There's always a consequence of it that we may not foresee to get the highest benefit and the lowest cost. We develop simple models, but we often fail because we have our biases. We want to believe that having choices or having more options is freedom. But they just bring us more confusion in our inefficient decision making cycles. On the other hand, machines are programmed to make the best decisions according to our goals, and they can help us become fully data driven. In our modern times. There, there's a customer data available. Like and it is just growing across different number of channels and devices. And it is much more than any marketer could ever collect and process. At the same time marketers rely on traditional automation techniques which are mainly manual. This actually is the main reason that marketers don't AI using AI technology enables marketing teams to save time for things that actually matter. And they can focus on working on creativity, strategy and more revenue generating ideas. After this short introduction, let's talk how AI can help email marketers and let's check out some of the use cases of AI in email marketing with some examples and case studies that may look at the successful implementations of AI in email marketing. We noticed if you notice that those companies have their customer data from different sources integrated and available for the AI system. To learn from a typical AI workflow would look like this. Once you identify the business problem, you need to input all data in the system. Only after that AI can provide predictions which would be used to make decisions later.

Ai helps us create a single customer profile where we can unify all attributes of a customer in one place. They can be activity based or demographic attributes. And this profile is continuously updated in real time. based on customer activities. It is also possible to sync or enhance customer profiles with other systems or databases. Usually, the traditional segmentation is a time intensive process. And by the time a segment is ready to set, it might be outdated and we may see those customers lose their interest in our brand or they may even purchase a product One of our competitors, scoring models can predict customers next action based on past behaviours. For instance, if we would like to predict which new subscribers would purchase a product, we need examples of new user profiles from the past historical data about their activities, and data about outcomes such as an order was placed or not. Lastly, scoring system scoring model is trained on past examples, we can apply on real customers to predictive feature actions, usually assigned scores are between zero to 100. And using these scores, we can run highly targeted campaigns for certain groups. For example, if you say that those who have a score of 80 and more they're high, they're likely to convert, so we can send them certain campaigns to encourage them. Recently, we implemented a scoring mode scoring model for a large publishing company, their list was above 7 million subscribers. It was one of the largest subscriber bases we've worked with. The implementation took us about a month to structure the data, create the model and validate it, then we started to create the segment's for them. By using this scoring model. Using the same creatives email copy, we identified the top 20% of subscribers with this algorithm. And we got twice higher open and click rates. All Brands strive for meeting customers expectations, so they will stay for a longer time and they can become a loyal customers to us. So in this case, customer lifetime value concept becomes very important. CLV is considered as one of the top marketing metrics, especially in ecommerce, and retail industries. I think it's extremely valuable for all businesses. First of all, it allows marketers to see which acquisition channels deliver the customers with the highest value. This way we can prioritise those channels and spend our marketing budget wisely, we can identify and apply the right strategies for each customer. Either they are newcomers or frequent shoppers. For instance, you can work on retaining existing customers through email marketing, especially in this current circumstances. Retention could be a key strategic gardener states that acquiring new customers may cost 6.7 times more than retaining your current customers. In order to train a mathematical model of CLV. We need historical data associated with your customers and they orders if there is also older value and time details that could be very helpful as well. So let's say we have all this data and we would like to work with it. So how does it work actually, then the prediction model is created and the data is introduced to the AI system, there is a learning stage. The grey dots, you can see here represent past transactions. The model

has a holdout period. In this period it is tested and after the validation. Finally, the model is good to go. For the same client that I mentioned earlier, we built a CLV prediction model as well. The accuracy rate was between 92% to 95%. Within the last six months period, our client could identify the most valuable subscribers prioritise the most profitable channels where they acquire those subscribers and optimise their email marketing strategy. At the end, our clients saw a revenue increase of 8.7% from email in this period of time. Alright, that's great. We have scoring models in place, we can predict CLV what's next, to make those models serve for our strategies. We'll need a personalised and relevant messages. With personalization, we are looking to build a strong relationship with customers provide them a unique experience. And ultimately, we want to convert them into loyal customers. The most common AI use case is predictive recommendations when it comes to personalization. We see recommendations in our lives every day. I think we're at a point that we even expect brands to show us compelling and personalised content that we might be interested in. So to make these recommendations work properly, we need to have certain data such as list of products or content, customer attributes, and pass interaction data. It Ah, we use SVD singular value decomposition, which is a very common unsupervised machine learning algorithm. To build custom recommendations, we use SVD. And let's maybe have a look at this and how it works. As you can see here, each row represents a user and each column represents a piece of content. And the AI system rates each content piece based on the user's interaction with the content. And then later on, ai decides the next best email to send or which content piece to include in email message.

One of our clients had challenges with sending relevant content and they want to increase subscriber engagement. For them, we created a model based on SVD. We compared it with the results they had with their CTR optimised strategy. Basically, they were sending the most clicked emails to the other subscribers. Our recommendations model increased clicks by 24.2% compared to the non optimised group. So I believe everyone here knows Netflix, so I wanted to give an example of it. And Netflix has about 80% of fever activity. Influenced by its recommendation algorithms, the company estimates that they save around $1 billion dollars annually by preventing cancel subscriptions. The company's goal is to achieve the maximum happiness per dollar spent for each use. But how have they managed to get such high accuracy with their recommendations, they track any data to maximise the user experience they create, like 1000s of clusters derived from subscribers reading habits, habits, even from day one. So there are different types of content recommendation deems that we can apply for customers in different stages of customer journey. When we acquire new subscribers, it's a great opportunity for us to build a long lasting relationship and convince them that we have interesting and quality content for them. However, we only have limited data from the sign up, which is not enough for further personalization. We suggest recommending new subscribers and sending them in, like showing them maybe trending content are the most popular content that we know our existing customers like very much. Everyone also likes new and fresh content, so you may include your recent content or newest products in your recommendations. Having recommendations of seasonal products may also trigger interest for a new commerce. You can also show them what content or which products will be available soon. For the active users. The main goal is keeping the interest high and engagement strong. At this stage, we have enough data to run one to one personalization and the recommendations you may show can be very effective. So you can show them the price drops for the items they've used in the past. And you can maybe drive urgency with items which are back in stock and other ideas I'm be showing them those items which were bought together, which will increase your average order value and also give your customers some inspiration.

If your customers previously purchased some items, maybe you can suggest them to buy them one more time. So it could be also another idea for you to go with recommendations. Well, nothing lasts forever, some of your subscribers may lose their interest, and you may notice that they are not really engaging with your content. At this stage, it is crucial to win them back before they churn, it is always a better idea to show them new content recommendations rather than the old ones. You can also play into FOMO, which is a very popular term recently, by letting them know about the last chance to leave a content or buy a product in the categories that they like. sending them some exclusive offers will help win them back. If you have some statistics, you can include them as well since people always look to others for validation before they make a decision to buy an item. In email marketing, we always say tested before you send it because even some small elements in your layout may have the individual effect on email metrics. multivariate testing or MBT is a method of email testing which allows marketers to test different combinations of variables at once. In the traditional AV testing, we can compare different versions of a single variable. For instance, using MBT we can test multiple subject lines, images CTAs, at the same time, against a different combination of these elements. On the other hand, with AV testing up can only test two subject lines for for the same email while all the other elements remain the same. For this presentation, we did an experiment using five different email arrivals. And we prepared a JSON feed with 108 email message combinations out of those 13 components. After one week and sending 600,000 emails, we saw an uplift in the open rates by 22% and increase the in the click rates by 26%. Although the AI features we discussed are commonly used by many marketing teams, still plenty of marketers think that AI is a black box. The reason is that they do not understand the concepts around AI and the benefits it can offer them. Obviously, it is not easy for everyone to get a grasp of what exactly artificial intelligence or machine learning is, especially for those who do not have any technical background. Luckily, today, there are so many resources available for marketers to learn from. And they can improve themselves about these concepts and find ways to implement them in their business. In the early times of AI use in marketing, large enterprise level companies introduce certain features leveraging artificial intelligence. Those were the times when the first adopters had to cover quite high price tags. For this reason, there is a perception in the market that AI is an expensive solution to implement. However, it's no longer the case. And the outcome that marketing teams get from AI systems is much more valuable than the cost. They are essentially associated with the implementation time is also another hurdle. And I would say it depends on the company. It may be between a few days and several months. The most important factors here is the size of the company, the number of data sources and the necessary integrations that should be done. The most AI companies offer very convenient solutions and we noticed that the implementation times are getting shorter as these technologies advance

For those who'd like to learn more about AI in marketing, I'll get a few suggestions. In my opinion, it's not only important for your business but also for your for yourself as well because in the near future, I believe artificial intelligence proficiency will be a must have on any managers CV. So marketing AI Institute is a great place to go. They have got a lot of great resources, videos, blog posts, and they recently started a podcast and I definitely recommend checking them out. It gives you a very good idea about the whole ecosystem and HubSpot Academy. They they're doing it very well and it's a completely free course. So I strongly encourage you on our website, i h technologies.com. You can find some blog posts, white papers, we are eating videos as well. So you can just check them out elements of AI by University of Helsinki I completed this one you can get a certification as well. And my friends, my contacts, they give a very good feedback about it. And you that stay has introduction to artificial intelligence course on on the website, and it is by Stanford University, I think it is worth to check out. So if you'd like to deep dive on the development side of AI, maybe you would like to check out Microsoft AI school, Google AI or IBM AI learning. Well, okay, so you learn about AI, you know the use cases. But does it mean that your marketing team should go ahead and get an AI system tomorrow, powered on and you'll have wonders? Not really. Before you consider going all in a AI in your email marketing programme, there are a few things that you may need to consider. marketing teams may encounter different problems depending on their size, structure and goals. Some may want to increase their sales, others may be after reducing costs. It is very you know like unique for each organisation than implementing AI to your email programme you may consider to start with the most urgent business need that you have. Having AI inaction also doesn't mean that you're successful. In order to measure your success of the AI system that you have got, you need to clearly define the expected outcomes. AI system that you use is just a tool, you still need to set your marketing strategic, and then the AI can help. You may consider starting with a single use case as a pilot project. It will allow you to understand better how AI works. And when you have some successful results. From the initial use case, you will feel like more confident you'll have more support from your upper management. And finally you can allocate more resources for the next AI use case that you'd like to invest. AI can only work with the data that you fit into system, you need to integrate your marketing stack with the AI system. This will this will be also an opportunity to review your current workflows and you don't overlap current business processes or create any any gaps. After you implement AI in your email marketing programme, you may not see overwhelming results overnight. AI is a self learning system and you may want to give it some time. Okay. Although AI can run many things on its own, it still needs a human touch. We need to be able to the strategies first, and then the AI tools or algorithms can do the job. We as humans, we are gifted at doing things which are not programmable. Artificial Intelligence is no match for human creativity, empathy, improvisation, social and emotional intelligence that we use and reflect to our lives every day in business. If we would like to run meaning teams, we should invest in people, not just data and new tools.

If you'd like to discuss how you can apply some of the use cases we talked about today, you feel free to reach me via email, or you can ping me at Ah, booth in the expo hall. Thank you for your time, I hope you had at least one thing that you could pick from this session and apply. I'll go ahead and check the questions that you may have. So I'll just have a look at it.

David, folder of Will this result in an opt in landscape for email? Finally, I do not really get in which part of the presentation David you refer to? I'd love to chat with you and ask you personally. And then maybe we can just figure this out together.

So I'm waiting for some other questions if you have. Alright, I think there is none on the application. And I'm looking at my mobile.

So once again, thank you. So if you have any questions, feel free to reach out to me. And finally, I would like to mention that on LinkedIn. I share almost daily updates, trends and news about artificial intelligence and marketing. So if you'd like to catch up with the recent trends, and go ahead and add me on LinkedIn, and let's stay in touch Thank you and enjoy the show.