Today on the Demand Gen Pod, Episode 15, Ryan from the Demand Gen Pod discusses his skepticism about relying on AI for email marketing decisions. He highlights the importance of designing frameworks and making decisions in a timely manner that align with the sales journey. Although AI has its limitations, it can be helpful when used in a guided and natural way. Ryan explores various ways to leverage AI in email marketing, such as personalization through customer data analysis and tailored content generation using tools like Chat GPT. He also mentions using AI algorithms to segment audiences based on behavior and preferences, as well as automating tasks like audience segmentation, content creation, and send time optimization. Additionally, he discusses how AI can improve A/B testing by analyzing results and suggesting optimizations for future campaigns. Lastly, Ryan touches on how AI can assist with copywriting by generating subject lines and content suggestions. The speaker expresses skepticism about relying too heavily on AI tools without actively tracking and monitoring them. They acknowledge the power of AI in improving workflows, but caution against letting it run free without close supervision. The speaker suggests using AI to generate multiple subject lines for testing and optimization, which can significantly increase productivity. They also emphasize the importance of including brand voice, balancing automation with personalization, and avoiding bias when using natural language processing and machine learning programs. The speaker shares their experience with Chat GPT, highlighting the need for variation in prompts to avoid generating identical responses. They discuss how they used different prompts to write TikTok descriptions for their podcast to achieve more diverse results. The transcript also mentions how AI can be helpful in improving email deliverability by identifying potential issues before sending emails. It suggests integrating an email validation tool or leveraging AI algorithms to detect red flags or spam triggers in email content. Ethical considerations such as data privacy and security are discussed, particularly regarding HIPAA compliance and obtaining user consent when using AI in email marketing. The speaker emphasizes the importance of complying with regulations and securing data properly. Other emerging trends mentioned include hyper-personalization in marketing, AI-powered chatbots, and enhanced customer journey mapping for better customer experiences. Overall, while there are opportunities for leveraging AI to improve customer experiences, there are challenges that need careful consideration due to evolving technology. The speaker admits being skeptical about fully embracing AI but acknowledges its benefits when used within controlled parameters. They invite listeners to share their thoughts via email at hello@demandgenpod.com.
Summary notes from Episode 15:
AI is used in the podcast for tasks and automations. AI lacks naturalness in
certain areas, like writing descriptions. Automating a step-by-step process
can improve repeatability. Using webhooks and tools like Zapier can connect
customer data to AI. AI can generate personalized email paragraphs based on
customer data. Dynamic content and personalized recommendations can be created
using AI. AI can automate tasks like audience segmentation and send time
optimization. AI reduces manual work and improves efficiency. AI can analyze
A/B testing results and suggest optimizations. AI tools can help generate
subject lines and content for emails.
Full Transcript:
00:00
Welcome to demand Gen pod. My name is Ryan. Today we’re talking about AI and email marketing. And to be honest, I’m actually not fully on board with this yet. I actually use AI. Even in this podcast, we use AI to accomplish some of the tasks and the automations that we run. And I’m going to do an entire episode. I think about how we actually build out all of our episodes because I think it’s really interesting. But in terms of email marketing and relying on AI to make decisions for us that we may not necessarily have any part in is a struggle for me. So a lot of the things that we do is to design frameworks that contacts will flow through.
00:42
And so we get to design the decisions that are made and when they’re made, and then the contacts role is to help us to make those decisions in a timely manner that aligns to their own sales journey. Right. And I will say that AI does lots of really cool things, and it’s pretty impressive, but it’s far from perfect and really far from perfect. And even in developing the show, in the places that we have used AI, for example, writing descriptions and things like that, we’ve found areas where it really lacks. And what we have also found is that it needs a lot of handholding along the way. So you can’t really just say, go do this end result. You really need to say do step one. Now that step one is done, do step two. Now that step two is done, do step three.
01:32
And that might be a five step process to get to your end result. I think that leveraging AI in that way can be very helpful. And if you can automate that process so that it’s repeatable, that’s even better. The important part though, is that it’s done in a way that feels natural. And natural does not equate to us asking for the end result upfront. Natural equates to us guiding it along the way. With that said, let’s get started. All right, so AI, listen, on the surface, AI really does kind of change email marketing, and it enables theoretically very personalized and targeted campaigns. And some advantages of AI and email include increased engagement, improved conversion rate, and better customer experience. Theoretically. But then again, all of those things can be accomplished by our own individual work and doing a good job. So there’s that too.
02:43
But let’s talk about some aipowered personalization so you can leverage AI by analyzing customer data and delivering some tailored content for each customer. I think that is kind of like this big weird cloud, like, oh, okay, yeah, sure. I could do that. So how could you literally do that? Let’s talk about it for a second. You could leverage something like a webhook. Okay, we talked about webhooks in another episode. You could leverage something like a webhook and a tool like Zapier, relatively low code, no code, to connect your customer data to something like Chat GPT. And we’re going to talk about the ethics of all of this later. But let’s just assume, put the ethics aside for right now. So, for example, you could have data fields. Let’s say that we know the last product that they purchased.
03:37
Let’s say that we know what company they work for or that they haven’t purchased a product yet, I suppose. And maybe we’ll just start there. Okay, so you could take that data, send it from your marketing automation tool or your CRM over to, say, zapier, connect it to Chat GPT, and start a conversation and say, write me an email to person’s name, period. On Xdate, they purchased this product, period. They work for this company, period. Right. So on and so forth, whatever. Keep it really simple. Simple is better with something like Chat GPT and then make a request. So write me an email to encourage this person to buy another product or buy this product. Okay? Knowing that we know these things about them and it will spit out a result for you.
04:35
Now, what might come out of Chat GPT is not necessarily something that is directly usable. You might need to format it in some way. But again, a tool like Zapier can manage all of that for you, which is really impressive. But you could effectively take customer data, each person’s data, and you could apply it through a conversation with AI to return an email, or even if it’s just a paragraph inside of an email. So maybe you have the course kind of breakout framework of an email, high, first name yada, yada. And then the second paragraph is this dedicated paragraph, and maybe that’s what the AI generates. So this is how you could literally implement this now, right? Just using a tool that allows you to send webhooks, fire webhooks from it.
05:22
The kind of like lowest cost email marketing tool that I’m aware of that does that is activecampaign. You can do it in the middle of a campaign. So, like, you would fire off a webhook, say you wait a day, you don’t have to wait a full day, but you wait some period of time to get that data back, and then you can send an email based on the result. So that’s certainly one way that you can literally do it right now to have AI enhance personalization by analyzing customer data and then replying to you with some sort of tailored paragraph literally for that one person. And then you just populate that into a field and you submit it and put it into the email, and then everybody would get a different paragraph every single time. Totally doable. Really pretty easy, actually.
06:05
Some more complex things, AI algorithms can segment audiences based on things like behavior and demographics and preferences. And another really good way, and this is kind of leaning on the first point that I had, is that some really good ways that you can leverage it are with dynamic content. So we can create this dynamic content for each individual personalize recommendations based on what they’ve purchased before, and you could even adapt the email design itself if you really wanted to create custom images. Although I think that the images with AI are the farthest from success of anything else, but you could theoretically. We actually tried to do that really early on in the podcast. We tried to use AI to generate blog post images for us, and it was a disaster, did a horrible job. So we decided not to keep going with it.
07:00
Some of the things that you can do, maybe outside of marketing automation, but just leveraging your data could be things like predictive analytics and recommendation engines. So you can leverage predictive analytics and those can improve your email marketing by predicting customer behavior and then optimizing campaigns so it can look at customer data. And these models are also built into CRM tools as well. But you can look at customer data or have a tool look at customer data and then predict which people based on behavior and timing are most likely to be willing to buy now or tomorrow or next month or whatever, right? And those recommendation engines suggest personalized product recommendations based on customer preferences and behavior. So you’ve probably seen this.
07:42
These are getting used everywhere, particularly in b to c where you might have purchased something, and it says, well, if you liked that product, then you’ll like this product. And a lot of that can also be managed by tools that are AI generated, where you teach at your library. And then once it understands your library of products, then it can make recommendations based on previous purchases and when those things were purchased. I think that you can also see this in timing based nurture where what is a bit, I guess it’s behavioral, but it’s behavioral time based, where you’ve got a product that’s purchased that gets used and has to be repurchased.
08:20
And so if you can calculate how frequently somebody needs to go and buy that product, then you can target them around the time that they would be buying it, just top of mind, brand focused, so that they remember that when they need to go buy that toothpaste again, it’s kind of a bad example, but that’s all right, you need to go buy that toothpaste again, that they get Colgate and not Tom’s or whatever it is. So beyond that, you can also leverage it for things like abandoned cart emails and then for upsell and Crosssell programs. So I normally think of upsell crosssell as more like b to b rather than b to c, but I mean, certainly relevant in both areas. And you can use AI in both of these places too.
09:00
What about automation and workflow so you can automate some tasks like audience segmentation. We talked about content creation and also send time optimization. This is also managed by AI, and this is implemented in lots of different tools already. Send time optimization looks at the time that emails were actually opened rather than when emails were sent. So for the people who have been engaging, it looks at the time, so they open the email and then it actually sends each email after that you want to have done this way because there are some emails that maybe you actually want to send at 09:00 a.m. Along with a press release. Right. But for nurture emails, you could set it up where it will send it at the time when the tool thinks that particular individual is most likely to open that email.
09:47
And so that is what the sometimes optimization is, things like that. I think sometimes optimization is a really great tool. It’s been around for a while, actually, and it’s been in tools like Eliqua for several years. And I think that it’s really special for nurture. Obviously, when you’re doing single sends, it just depends on the send. If it’s time sensitive, then you can’t leverage it. But it is really neat to think that you could send 100,000 emails and it would, in theory, send. Let’s just say that 20% of your database of that send has engaged recently, that it could pick up that time and send 20,000 of those at the most optimized time for each individual person. We certainly could never do that as individuals. So really wonderful that a tool can do that.
10:34
And then that really translates into improved efficiency, theoretically reducing manual work. AI certainly reduces manual work for us on the podcast, and then it also really helps us to get the podcast done quickly. And that’s another really big thing that has been very helpful. And it’s not really just AI. It’s also a lot of automation. We use Zapier basically for the entire podcast, except for basically when I hit between the time of me hitting record and stop recording. That is, I guess, theoretically manual, but we have automations built into that. If you’re watching on YouTube, you’ll notice the camera changes back and forth. It’s still only one camera, but there’s an automation to push it to zoom in and then zoom back out again at random time intervals so that it’s more interesting.
11:24
I think it probably did it in the middle of me saying that there are lots of different ways that we can kind of do that. The other thing that’s kind of nice is that you could leverage AI to help to optimize resource allocation and to enable you to focus on strategy and creativity. And that can be very helpful. What about some continuous improvement opportunities so we can improve a b testing by analyzing results and suggesting optimizations for future campaigns? AI can help us do that. Techniques like machine learning and data analysis to uncover patterns and insights. There are lots of tools that already do this, and they can be really helpful, especially when you have large data sets, smaller data sets, not so great. Large data sets. Excellent, because there’s lots of things to learn from, right?
12:11
And if we can leverage a tool to help us to uncover different patterns and insights, then that saves us hours of manual labor and thinking power. And we can be focusing on other things. So as long as we are having our tools, rather than using our tools, so that we don’t have to do anything, if we’re using our tools so we can focus on doing something else, then I think that’s a really big win. Maybe on a Friday, maybe on a Friday. I’m not really doing anything. Just let the tool do the work and we walk away. And then the other nice thing is that some of those insights, they can really drive some continuous improvement and ideally better results in email marketing. So there’s that too. All right, something that I think is particularly interesting. What about copywriting and optimization?
12:55
So we touched on this in the beginning, and AI can definitely help you in copywriting. There are AI tools that will let you to generate compelling subject lines and content. So you can just type in what you’re trying to achieve and then it will give you, spit you out a bunch of subject lines. There are even tools that do that and then connect through an API so that you could theoretically send unique subject lines for each person. There’s that too. And I think that can be really neat. Again, just there’s a lot of faith that’s getting put in a tool that you’re not actively tracking and watching. And I don’t have that much faith in AI at the moment. I think it’s really powerful and can really help workflows, and we use it again to help our own workflows.
13:41
But with that said, I also think that just letting a tool run free is really dangerous when you’re not tracking it and watching it really closely. And by mean really closely, I mean, if you generate a subject line that’s different for every single send, that you would know what that subject line result is before you send the email and it starts to defeat the purpose a little bit. However, if you flip that and you were to say, let’s use AI to generate 100 subject lines so that we can look at what we feel is the best one and maybe test a few of them, then great, I think that’s wonderful. And if it can generate 100 subject lines in 30 seconds, then it would take us probably a day to do the exact same thing.
14:22
And we might not even do quite as good of a job, at least half, let’s say half as good or whatever, then that’s a huge win. That’s a 50% uptick in productivity, and I think that’s perfect. The other thing is that some of these natural language processing and machine learning programs, I mean, they’re getting better and they are. And I think over time, whatever that time is, I don’t really know, will improve. But again, some considerations that you really need to think about when you’re preparing these tools and starting to rely on them is ensuring that you include your brand voice, that it can mimic the brand’s voice, that it can balance automation with personalization, and also that you’re avoiding bias. So that comes down to how you request the prompts to AI.
15:15
You sometimes might want to have variation in the prompts that you give it. Even if it’s a sentence that’s different, it will spit something else back out. One of the things that I found with Chat GPT is that if you ask it basically the same question, even with slightly different data, so say like the key points in our example of an email. If we said write an email for person’s name, they work at this company, the company does this is in this industry, and then try to get them to sell the product right. Few extra sentences are going to be in there. If you keep basically that entire prompt, but you only change out the name, the company in the industry the result is going to be borderline identical every single time, with the exception of those few things.
16:04
That’s what I found with Chat GPT. So one of the ways that you can combat that is simply by giving it the opportunity to have various prompts. So maybe you have four different prompts and those prompts, the paragraphs, four different paragraphs that are asking the same question in different ways with the same data, right. And then Chat GPT will provide a different result. We used this exact technique with Chat GPT and Zapier to write TikTok descriptions for the podcast. So we found that for every single podcast description that we asked it to write for TikTok, and then we automate all of this, that it would spit out basically the exact same thing, the exact same sentence and structure. And it was very frustrating to read, because then you’re thinking, I can do this, I can swap out a few words.
16:55
So what we ended up doing was we ended up creating, I think, four or five different prompts. And one might be writer description for a TikTok video for a marketing podcast. One might be writer description for the demand gen podcast for TikTok. One didn’t include TikTok at all, and it spit out different paragraphs based on that prompt changing, even if it was just a little bit. And that made a big difference. And that’s what we ended up going with to help us to write those descriptions. So something else that you can keep in mind is that if the input is consistent, the output is going to be consistent, and that can be a really good thing, and that can be a bad thing as well. So what about things like detecting bad data? So particularly around deliverability and email spam filtering?
17:44
I think that one way that AI can really be helpful is improving deliverability by identifying potential issues and optimizing email sending before it even happens. I’m surprised to a degree that something like an email validation tool is not built into more. And this isn’t really AI related, but maybe there’s a way that AI could be leveraged in it built into more pre sense, or when you go to build a segment that the marketing automation tools are not automatically running, that I guess, with your permission, I suppose, through some sort of email checker to ensure that the email is valid or that’s not process on CRM tools, I’m kind of surprised about that. But when you develop those algorithms, they can detect things like red flags and invalid email addresses, spam triggers, even spam triggers like in the email subject lines, those are getting used.
18:38
I’ve seen in a couple of different tools, email marketing tools, I’ve seen that get used as well, where if you put, say, free in all capital letters, that sometimes it will warn you to say, hey, heads up, this looks pretty bad. The email on acid does that. It’s an email checking tool, like a testing tool. And litmus, one of those two, I can’t remember, they certainly do it. So they’ll say, like, we kind of think that your email is a little spammy, and I’m assuming they use AI to accomplish that task. And then finally, some challenges to include adaptive spam filters and also ensuring that AI doesn’t compromise security. And this is kind of where we can get into these ethical considerations.
19:22
And data privacy is that if you are using particularly HIPAA compliant data, pretty much all of this is out sending anything outside of your tool. It has to go through a HIPAA compliant path, and then it has to go to a HIPAA compliant source, and then it has to go back through a HIPAA compliant path. Right. So all those things need to be considered. But HIPAA aside, because that’s a really easy one where you just kind of say no, just think about the transparency in AI usage and ensuring that the AI doesn’t actually harm anybody who’s receiving the email. Right. So again, that is partially coming down to the fact that you’re sending somebody else’s data without their knowledge, necessarily, or I should say most likely, even if you make it really clear in a data privacy policy, right.
20:07
Nobody reads them, that you’re leveraging in AI, that it’s still going out and it’s getting to a tool that we don’t really know who has access on the other side of it and what they could see or what they could take or whatever. But data privacy is really important, and businesses have to make sure that they’re complying with regulations when using AI and email marketing. So you need to make sure that your data privacy policy is applying that consent or requesting that consent. And you can certainly do that, obtain users consent, you can provide opt outs and do your absolute best to secure data. But if there are doubts on how you could be securing data, then you should not be doing anything with AI and consumer data or customer data.
20:48
And I think that at the end of the day, that’s something that we are just going to have to come to the realization with, that AI may be more helpful inside of these tools or outside of these tools, something like Zapier to assist us in tasks rather than assist us in writing language. Unless you’re taking things that are not necessarily person specific, not using their name, you can put their name in later, it doesn’t really matter. Not using anything about their phone number or location or anything like their email address, not sending any of that personal data to an AI tool is probably a really good path to take. So some other emerging trends that I think are going to be happening in marketing and already are happening in marketing, that hyper personalization, I think we’re going that direction.
21:41
And then aipowered chat bots, that’s certainly already happening. And I’m hoping some enhanced customer journey mapping to be able to maybe shape programs in a way that best suits each individual person. So I think that’s our next step too. Overall, lots of opportunities lie in leveraging AI for better customer experiences, but there are some challenges that we need to keep up with all this evolving technology. So I’d love to hear your thoughts about this. I think it’s really controversial. I think that some people love it, some people hate it, some people are really skeptical. I think I probably fall into the really skeptical, but I think when used in really specific, finite, controlled ways that it can be pretty impressive in terms of customer data. I’m just not quite on board with it.
22:37
You can see that it gets used in sales emails and stuff. I’ve received sales emails that look because they’re close but they’re not there. It’s just kind of obvious in some way. So I think that there’s that too, but I don’t know. Tell me what you think. You can reach out to me at hello@demandgenpod.com? My name is Ryan. You’ve been listening to the demand gen pod and we’ll see you next week.