Mailchimp's Head of Data Platform Product on Using AI to Enhance Email Marketing

The rise of generative AI has gotten many marketers nervous about the security of their jobs. For good reason: ChatGPT and its generative AI brethren can out-write, out-design and out-code humans when it comes to speed, volume – and often quality of output as well.

So what is the human marketer’s role in the golden age of AI? Still plenty especially when it comes to higher-level tasks, contends John Humphrey, head of data platform product at Mailchimp, an email automation platform owned by financial software giant Intuit.

Humphrey joined the AI Business podcast recently to talk about how marketers should use AI, what fundamentals remain the same, and whether or not they should fear for their jobs.

Listen to the podcast or read the edited transcript below.

AI Business: Emails have been around a long time. Are they still relevant as a marketing channel today versus, say, social media?

John Humphrey: Short answer is absolutely. It's relevant for two reasons. One, marketing channels are sort of like food − you want to maintain a well-balanced diet, you don't want to get all of your content and all your communication in one particular channel. And also, the right combination of channels is not necessarily the same for me (as) for you.

But there's one other aspect of email as a particular channel that I think is unique and worth calling out. And that is, in a world where increasingly having a known, first-party relationship with your customer is essential, email is still the best channel in which to own that relationship.

As we've seen in recent years, especially if you're relying on say social media channels, changes that can be made by other parties in short notice can suddenly undermine the relationship that you think you have with your customer.

Email in particular has a very special quality in that it is still the best way for me to own a relationship with my customer in order to maintain relevance with them. … Certainly, channels will come and go. But I think email in particular has a unique place and will continue to be so for some time to come.

AI Business: Automation capabilities have redefined the role of email marketing over the last decade. How do you see the use of AI and ML taking that to the next level? And what additional benefits can marketers expect?

Humphrey: These are incredibly exciting times for marketers. …

If you look at what has happened in the last couple of years, and in particular what has come to the forefront in the last couple of months, now AI has the ability to be generative, to create content essentially from nothing more than a prompt. That's very exciting and something that we here at Mailchimp have done a ton with in recent months.

But if we only talk about content generation, we're only talking about half the problem. In recent months, content generation has gotten faster and better, and all its possibilities are starting to pop into people's minds. For a marketer, your biggest bottleneck (has long been) how do I create enough content? Now I can create content almost at will; who do I send it to? And that's where I think getting our audience generation game as equally automated (as content creation) is going to be important.

This is a class of problems for which ML is very well suited to serve marketers in terms of identifying the right group of customers for which that newly generated content will resonate and compel them to take action.

If you’re a marketer, these are really exciting times because two of the hardest parts of your job that slow you down, now, suddenly, we have the technology to automate a lot of that and go much faster. It frees us up as marketers to think about higher-order problems like creating value for our customers.

AI Business: How do you use AI in audience generation?

Humphrey: In terms of audience generation, I would look less to AI and more to machine learning.

What we're looking for with machine learning is a scaled version of a thing that we have been doing for decades for marketing, which is trying to predict a response. If you even go back to the olden days, and logistic regression and trying to predict who is most likely to respond to a campaign, that's kind of the thing we need. The shortcoming of logistic regression is it doesn't do it (at the) scale that we need in order to keep up with as big as our businesses are and as much data as they generate.

Machine learning is essentially a scaled version of that. It’s about making sure that you have good data about your customer that you’re feeding the ML algorithms, with a clear objective as to what the customer outcome I wish to predict. That becomes the bedrock of how you get to a place where you can actually start to segment your customers at scale, and have these audiences being generated in lockstep with the content that we can now generate pretty readily as well.

AI Business: How do you access that kind of capability? Do you build or do you buy?

Humphrey: If you're a very large enterprise, you probably have a small army of data scientists and engineers who can start to build and make this stuff readily available to you. And they can do so in the confines of your one very specific use case that you − and only you − in your particular business happened to have.

I'm interested in how to mass produce this in such a way that businesses that hardly have an army of people, let alone an army of data scientists, can have access to it. And also, how do we then make it generalized enough that it works reasonably well for your business and for mine?

It's the perfect time to be thinking about this, because we are at this confluence: The way in which we store and access data is pretty good, the way in which we can generate content is pretty good, and the algorithms we have at our disposal to make sense of customers as audiences are finally good enough as well. When you bring all that together, … marketers are going to have a much better set of tools in their hands in relatively short order.

AI Business: Today, you unveiled a new product called Email Content Generator. Can you talk more about that?

Humphrey: The idea (behind Email Content Generator) is that we generate email content on behalf of marketers. It does leverage GPT technology. … And the way to think about it is if I, as a marketer, need to create a campaign, I can enter in a short prompt − say, I want to offer a 15% off bedding products – and ask it to write an email about that. Our technology will come back with three different options that you can choose among, make whatever edits you want, and then drag the text into the body of the email that you're working on creating.

It strikes a nice balance between automating away some of the tedious parts of the job, but not depriving the marketer of their agency in making sure it maintains the brand voice and personality that they want to come through to be authentic and compelling to their customers.

AI Business: On the flip side, what's the benefit to the customer of marketing teams using AI and ML?

Humphrey: Looking at the arc of marketing technologies over the last couple of decades, we have gotten really good, really fast at putting stuff in your inbox and in your social media feed. We have done that faster than we have gotten good at making sure we are delivering the right message to the right person. And what's really exciting about the advent of all of this AI, is now we're finally getting those other two parts of the process caught up to our ability to end up in your inbox. It's not necessarily that you will see fewer things in your inbox, but you will see better and more relevant content in your inbox.

However, in many ways that's not even the best thing that happens if you're a customer. If our marketers are no longer spending quite so much time dwelling on copy decisions and having to slog through that, they will be doing something else instead. They will have more time to think about what are the compelling products and offers to bring to market for their customers. And that frees you up from having to do administrative tasks. Now you get to be more innovative on behalf of consumers and bring better products to life in more relevant ways.

Skills needed by modern marketers

AI Business: What skillsets will modern marketers need to be successful in the AI age?

Humphrey: More will stay the same than change. You still need the same skillsets in terms of understanding the fundamentals of marketing, understanding what motivates people to consume, understanding what makes a compelling product and offering − all of that remains intact.

As for how to make the most of the technologies that are now available to you, as you go through that process? I would say it looks like a kind of survey level understanding. So if we were to take generative AI, for example, do you need to dive all the way down into like neural nets and Transformers and how those things work? No, not particularly.

But you should have some idea of (what it means) when somebody mentions large language models, what a large language model does and what a large language model absolutely does not do. (But) if you’re to the point where you have some point of view on whether a large language model can get us to human-level intelligence, then you probably understand more than enough.

And again, you don't have to dive too far into the math or quite frankly, at all. … Just understand what these things do and do not do.

AI Business: Will AI replace marketing teams anytime soon? And should marketers start retraining in other disciplines?

Humphrey: The short answer is no, it does not replace marketing teams, for exactly the reasons we’ve been talking about – the fundamentals of what connects a product and the consumer are still intact.

Are there some roles that will be less necessary in that it will take less time to perform? Yes, there are clearly some things that are going to get automated. But if people were to step back from all of this and have the takeaway that ‘automation is coming, marketing is going away,’ that misses the point because while we will spend less time doing certain tasks, the fact that we have so much better technology and can automate so much more of what we do now means that while the percentage of work done by the human may go down, the denominator in this fraction essentially goes up. There's so much more opportunity we can go after, so even if we're doing a much smaller percentage of it than we were before, net, we're still all going to be pretty busy.

AI Business: Do you have a sense of how far along marketers are when it comes to using AI/ML? Or not using it?

Humphrey: This is a question I get a fair bit because everyone's got a sense of FOMO about where am I relative to my peers. There's a couple important things to remember here. First, these are incredibly early innings. Most of us weren't even talking about large language models, or a thing called GPT a year ago, so don't allow all this fervor about this topic to lead you into thinking that everyone's got a head start.

What it looks like to use these technologies now is nothing compared to what it's going to look like, say, a year from now, certainly five years from now or 10 years from now. We're all just starting to figure it out together. I'd also say, to the extent people are using it, it's still very skeuomorphic at this point. (We are still at the stage of) ‘let me take a task that I do today as a marketer and then let me find a way to use AI to automate that task explicitly.’

What we have not yet gotten to is reimagining the whole marketing process and how I go to market, and how to employ technology (in this effort). When we get there, that's when we'll know things are really starting to escalate. And if at that point, a marketer hasn't necessarily fundamentally revisited how they're going about engaging with consumers, they should probably feel a bit more concerned as to whether they're potentially falling behind.

Segmenting customers using AI/ML

AI Business: Can you talk about segmenting customers using AI/ML and data?

Humphrey: Personally, I know what I'm going to say is going to sound almost ridiculous at a time when everyone wants to talk about AI and ML. But the first piece of advice I would give to anyone thinking about segmentation is have customer data. So often, where segmentation efforts go to die, is I literally do not have the data that I need about our customer. And in fact, one of the things that I probably spend the most time on in my role as a product leader at Mailchimp, is how do we make Mailchimp a great platform for housing and activating your data so that you can do things such as segmentation.

This is a problem I think about a ton. The first thing is to collect the data. As for how you go about doing segmentation? Well, that varies a lot. Certainly, big businesses and smaller businesses have very different technical problems. And it would be easy to kind of get lost in the sauce on that. But I'd say, remember, at its core, what segmentation is trying to do is trying to help you get more personal with your customers so that you can have relevant conversations with them.

And the truth is, we as marketers can probably very quickly rattle off the … number of business personas that exist in your customer base. Okay, that's cool. You know what your five customer personas are. How do I go about making sure that for every last customer in my database, I can line them up with the right persona, so that I can then have the right conversation with them?

I make this point because regardless of what algorithm you use, regardless of what data you use, coming into the process with extreme clarity about who my customer is and what I'm trying to get to in terms of my personas, is the easiest way to make sure that you stay on track and successfully deliver a segmentation that is impactful with respect to your ability to talk to customers.

AI Business: Can you name some use cases that succeeded and some that failed?

Humphrey: Audience generation use cases have been around for a while. One example comes from one of our customers that recently activated a segmentation split of their own customer base − which accounts were created with a workplace email versus those that were created with a personal email address. They then sync to the delivery of their marketing campaigns for either weekdays between 9 a.m. and 5 p.m., or during the weekend, depending on the email address used. This is an example of (benefiting from) knowing the persona of your customer and bringing that intuition to life at scale − you can drive better outcomes.

As for generative AI use cases, while today is exciting because I'm talking about things such as generating an entire email, this is not the first thing that we have put into the wild. And not all of those attempts have been as successful as this one. It’s early innings, and oftentimes we have done a good job of with concepts like headline generation — and no one discovered them and used them in the grand scheme of things. We were generating headlines before ChatGPT blew up. But was anyone talking about headlines in Mailchimp? No, because we didn't do a great job of creating awareness and driving people to realize that it's there in a way that they could create and extract value from it.

For all the improvements in technology, and there are many, the reality is you still need to deliver a compelling product to a customer in a way that they understand and trust, or else they’re not going to use it. … Fundamentals of good marketing simply have not changed.

AI Business: Do you think physical direct mail is dead? Or is there still a place for that in marketing?

Humphrey: If I were to judge by the contents of my personal mailbox over the weekend, it is far from dead. Not just because it shows up in my in my mailbox, but because I know that at least one person in my house is reading the catalogs that arrive with quite rapt attention, and it's ultimately leading to conversion.

It kind of goes back to our whole premise of, ‘Is email itself a channel that might be on its decline?’ And the answer is no, because it turns out, we all consume information differently. We all appreciate different surfaces. As we think about how do we leverage these technologies as marketers, it's really important to remember that we should meet our customers on whatever surface they want to be on. And the principles of what is good marketing are the same across, and frankly, how we leverage these technologies are very similar.

Yes, there's some last mile differences in terms of how do I print something, versus how do I put it in an email, versus how do I render it on a social platform, but everything else that happens in the middle is fundamentally the same. And so I think it just becomes a matter of different surfaces work at different times for different people, and we need to be there.

Decades after the advent of the electronic message, and years after social media ad platforms have become huge, catalogs still show up in the mail. That's not going to change in the foreseeable future.

AI Business: What do you think the future will look like with AI/ML, data and marketing?

Humphrey: The future is going to look very different. In the early days − we already have some clues − it's going to look very much like Copilot (a generative AI app that writes code). We will start with blank slates, whether they be in terms of audience creation, or in terms of content creation. And from there, we will get some assistance, like ‘Hey, have you considered this piece of content?’ Or ‘have you considered this audience?’ and we will build up.

Over time, what will change is as the technology gets better and as we become better users of the technology, it will start to look more and more like autopilot.

(For example,) the AI creates a campaign that it thinks would be good for your business, given its objective, and tailored to an audience that would benefit from seeing it. Then you as the customer, as the marketer, review that, put your own spin on it, maybe even provide feedback to the machine – ‘I think that this discount is too steep, or this audience is too narrow’ − and it goes and iterates, based on your feedback. Then it comes back (with the changes). Eventually you will have something that you are comfortable sending to your customers.

The mental model I would use here is like having a smart intern. You would never necessarily let them hit send on a campaign without taking at least some sort of peek at it. But by and large, they're going to get a good first iteration.

This then begs the question of if we're not doing the sorts of things we were doing before, what will we be doing? And the answer is a lot. One, there will be far more campaigns and far more products than we are capable of supporting now.

But the far bigger change is we now get to up-level ourselves to think about higher-order problems. What are the customer pain points that we're not even thinking about addressing right now? And how do we bring to life products and offers that address those pain points for them? How do we think about our business and how it's performing? For these reasons, and many more, I'm very excited about where this is all going.