Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Personalized Marketing at Scale w/Jim Laurain
211 Plays1 year ago
In this episode of AI-Driven Marketer, Dan Sanchez is joined by Jim Laurain, an expert in advanced marketing platforms and personalization through AI, to delve into the intricacies of using AI-enhanced Customer Data Platforms (CDPs) for crafting personalized marketing strategies. They discuss the advantages of AI in understanding consumer behavior on a granular level, the transformation in message personalization with tools like AMP, and the real-world implications of integrating AI across various platforms like ecommerce and fintech. Jim also shares insights into overcoming the challenges in data aggregation and the future potential of AI-driven targeting in creating more effective and efficient marketing campaigns.
Timestamps:
00:00 Personalized music and video creation using AI.
03:42 Basic steps in learning AI, personalization through AI.
06:53 AI personalization tailors product offers to users.
11:56 Identify reasons for app use, personalize interaction.
14:35 Quality of sample matters more than statistical significance.
18:37 Ecommerce: popularity doesn't always equal conversions. Personalize.
20:00 Customization needs parameters to avoid chaos. Similarity is crucial for user experience.
22:52 Customize messages, build content library, timing.
29:07 Creating products through data science, initially unnoticed.
29:53 Advancing technology creates need for education.
33:59 Understanding customer behavior drives effective marketing tactics.
36:23 Seeking specific items at Lowe's can be frustrating.
40:34 Email automation saves time and simplifies decisions.
42:50 Zencastr's clipping tool simplifies clip sharing.
46:53 Use AI to engage users with injuries.
51:05 Identify churning early; data science will prevail.
52:38 Twitter data potential, AI Grok's real-time insights.
55:54 Discussing growth, personalization, and tech at MAU conference.
Recommended
Transcript
Introduction to AMP and AI Personalization
00:00:05
Speaker
Welcome back to the AI-driven marketer. I'm Dan Sanchez. Friends, call me Dan Sanchez. And I'm here with James Lorraine, who is working for a fascinating company called AMP, who's really killing it on the personalization side of AI. And James, before I had you on, I told you I was excited about this topic because I recently had an epiphany
00:00:25
Speaker
i'm thinking about like okay the future of a you know you try to like the better you can forecast the better you can make decisions now and start going in the direction of that forecast and when i started thinking about the future james i was seeing sora and i saw audio or.
00:00:42
Speaker
I forget how they pronounce it, but I call it udeo.com with the music, right? And it's making live music. And you're like, Oh my gosh, like some of the cherry pick songs are actually freaking good. Like that could be real music. And then you look at that and you're like, you know what? Like right now it takes 60 seconds to generate this. And so it will probably take a few seconds
Future of Personalized Media Experiences
00:01:00
Speaker
to generate. But I'm like, in a few years time, this stuff's going to be done in real time.
00:01:06
Speaker
And there's no reason why we're not going to be experiencing like, custom music tracks and custom little videos that are completely customized and personalized for people real time on websites or wherever the heck we're going in VR. Like it's good to feel like video game level real time stuff and I'm like, the future is just going to be highly, highly personalized.
00:01:26
Speaker
And that's, that's going to be a huge part of AI, like huge, because who doesn't want it to be more personalized? Like I used to joke with, I know I haven't even let you talk yet, but I used, I've made a couple of kids books. I'm like, wouldn't it be cool if I could like custom order a kid's book and it was customized to my kids, what I want to teach them with their names and personalities and interests baked into the book and just printed on demand. Bam, boom, fully personalized.
AMP's Approach to AI Beyond Content Generation
00:01:51
Speaker
And essentially that, as far as I know, that's what you're doing with AMP. Is that correct? Like you're personalizing apps. Yeah, this is, this is, this is actually, we're hopping right in what you've just hinted at. What you've just said is where most people are right now. What we're trying to do is get people to the next level. Here's what I mean. Right now, AI people are thinking content generation.
00:02:15
Speaker
I can make stuff. I can make a kid's book. I can make this. I can make videos in 60 seconds. I can make music I can make. And so we've had this explosion of basically the content isn't your constraint anymore. It used to be, it used to be that like, you know, I can't, I can't write enough blogs. I can't make enough product descriptions. I can't make enough videos or songs or whatever. That was my, that was my holdup. Now the problem is distribution.
00:02:40
Speaker
So I can create anything. I can create a bunch of songs. I can create a million songs right now if I want to do. What do I do with them? How is it effective once I have them? How do I determine who gets what and who likes what and who, and that's, that's what we, that's what, that's what AMP is doing basically.
AI in Education and Personalized Learning
00:02:56
Speaker
So I see it as, I mean, I don't think people are even using AI for true personalization yet. Yes, they're coming up, they're making content, but even with the kids book thing, the personalization factor is the app that makes it easy to make the kids book, right? Someone's got to make that app or that SaaS or that tool, maybe it's Amazon, I don't know. Someone's going to make that and load it in with all the questions. Essentially, it's almost like a survey in order to make said book because people don't even know what they want, so you have to guide them through that process.
00:03:25
Speaker
But I imagine it's going to get to a point where yeah, personalizations happening. It's not just content, but like the content gets personalized. Let me let me tell you an example that I'm working on currently. And I'm like, I haven't seen anybody else do this, but it's currently technically possible. So I'm launching this little AI course, you know, it's it just goes through the basics, because I have a lot of friends that are like, Dan, how do I how do I get through to where you're at? And I'm like, well, it starts with super prompts, learn how to do super prompts, then
00:03:54
Speaker
I have something called the step method where i'm using templates and examples and also to feed into the super prompts and then Chain prompts and then you start with custom gpts after that, you know But i'm like i'm feeding these lessons into my crm to drip out via an email course And of course my crm is equipped to go go ping
00:04:11
Speaker
AI or OpenAI's API. I'm like, what if I just take the transcript from each one of the videos in these emails and a few pieces of information about the student, what their job title is, what industry they're working at, maybe what their stage of learning AI is, and I send it back to OpenAI based on this lesson and based on what we know about this person.
Data-Driven Personalization Techniques
00:04:34
Speaker
Give me a short little action plan for this person.
00:04:39
Speaker
That's personalization, but it's it's way more than just the mad lib style we've been using for forever. It's actually personalized instructions. Is it kind of shallow? Yeah, but is it way better than if I had just written a generic sample?
00:04:54
Speaker
Yeah, way better because it's speaking to their world. Speaking to CMO is going to be different from email marketing specialist, right? Very different, but still in the same context of the lesson. That's kind of where it's capable of being now, but in the future, I can only imagine what that, like the video will even be customized in the future.
00:05:12
Speaker
No, and you're hinting towards exactly exactly what we do. This is really cool. So part of this is, again, around the distribution, who gets what and how do we determine who gets what and what they see and all that stuff.
Agentic AI in Content Distribution
00:05:24
Speaker
So what we have found so data, data, data is a problem for companies. They view data as a problem. There's never enough data. It's never clean data. You know, for your example, it's your email list. You have your email list. Everybody has a job title and things like that.
00:05:39
Speaker
But really, you probably have more data than you think you do or realize you do. You also have information about did they click? Did they not click? When did they click? When did they open? How much did they read? Did they scroll down this far? There's a ton of data here. So what we've found is you can feed that data.
00:05:56
Speaker
back into the AI and not just create like the custom content. Cause again, everyone's heads in content right now, not just the custom content for the course, but also when should I send this email? How often should I send this email? Should I send this email or this other email?
00:06:11
Speaker
Like there, there are a lot of decisions there that can be made. And that's, that's what we're doing at a brand scale. Now think of it for a brand you have, especially. So we'll just say B to C think of like a clothing retailer. I have every time you've opened an app, if you have an app, every time you open an app, did you open emails? What did you click? Which piece of information did you click? What did, what products did you view? Did you add to cart? Did you add to wishlist? Did you remove, did you use a promo code, a gift, a promo code, a gift card that the amount of data you have is massive.
00:06:41
Speaker
So what most companies do now, to your point, is they build kind of a static user journey. You know, I send an email day one, welcome and whatever. And then I might put in a recommendation, a product recommendation or something like that. And the view of AI personalization or just personalization in general, to your point is the template. Like, hi, first name, you know, and here's a product for you. And then I fill that with a recommender system.
00:07:06
Speaker
What our system does is say, we have way too much information to be playing that game. What we should do is we should take all of that customer data and feed it into AI, similar to what you're saying with your email, you know, the same thing you're doing with your email with subscribers, and say what, you know, for example, what should we send to users? Should we send an SMS? Should we send a push, an email? You know, if we send that channel, when should we send it?
00:07:29
Speaker
What should we put in it? And we're building that user experience. And it's even decisions like, um, and we're getting into the agentic part, which we should talk about, but decisions like if they abandon a cart, should I send an abandoned cart email or should I not? Cause
Ethical Challenges in AI Decision-Making
00:07:44
Speaker
it's always been the de facto that like, yes, you should. And it should be sent within 30 minutes and all that until they get trained for it. And then they're take just taking advantage of the system, you know?
00:07:53
Speaker
Yeah, I just saw that. If you cancel your Grammarly account, they send you an email for 50% off. So it's like I told everyone on our team, go cancel your Grammarly and then sign back up, because you get 50% off. But people learn to that and then you exploit the loop, right? So what we're working on on the AI side,
00:08:12
Speaker
is agentic AI. So agentic AI is AI that has agency. So in other words, it's able to make decisions. There's kind of two parts of this. One is it's able to make decisions on its own. So when you're creating content, essentially it is making decisions.
00:08:29
Speaker
Right? Because it's, if you prompt it for something, it is determining what it should feed back. And people are used to this with LLMs. So I ask it to write my, you know, science paper for school and it makes decisions about what words to put on the page. It's already doing that. We're just allowing it to make decisions now, like again, email or SMS.
00:08:47
Speaker
8 a.m. or 6 p.m. You know, like, there's a lot of decisions on what should I send, who should I send, what should be in the content and all that stuff. We're allowing it to make those decisions. The second piece is, have you looked into, like, agentic AI at all as far as, like, multi-agent systems and all that? I've explored it. Everybody that I've seen talk about, like, multi-agent systems, I'm like, this is bogus. There's no way this works yet.
00:09:09
Speaker
I'm like, there's no way this is the working thing. But I think what you guys are doing in a tighter environment probably actually works. Because everybody else, I'm like, nah, it's not going to order food for me and
AI in Marketing and E-commerce
00:09:21
Speaker
get all this done. At least not in any way I want.
00:09:24
Speaker
Yeah, there's a lot of ways you can approach it. The one example that most people seem to get is you prompt an LLM and it gives you a return and you're like, nah, it's not what I want. I kind of want it more like this. And you go back and forth with it a bunch. And it's like, could you create the agent that knows what you want and the other one that does the prompt and then have the two communicate to refine it before it gets back to you?
00:09:45
Speaker
So that what you get is closer to the result you want so it's still on a micro Yeah, I mean you're still essentially just breaking apart a larger project in the micro micro steps Yeah, I mean that's what's gonna be teased to do that kind of stuff
00:09:57
Speaker
Yeah. And that's what's interesting is to me, it's like the fantasy football analogy. Like every player isn't good at everything. So it's like, I want specialized players who can do special things and then come back to me. Cause what you see when you start to build prompts is if I prompt, if I make my prompt to, if I'm asking for too much, what I get back is garbage. So I need to like build a team of focus things.
00:10:17
Speaker
So how we do it on the customer service side and the customer engagement side is we build an agent per user and really we could break that down finer. So I broke one down earlier where you could do one that's a pricing analyst and you can have a pricing analyst per product that you have. You know, what are the competitors offer for this product? What are similar products? What are, you know, fine-tuned the pricing for this product?
00:10:41
Speaker
Right? Because I'm not limited by, once you do that, you replicate it across your products. I'm not limited by like physical people in chairs. For us, it's for each user. So I want a user who understands you really well. So like I'm using, I freaking love Rocket Money right now. Like that's my new addiction is I'm checking my Rocket Money app like 15 times a day.
00:11:00
Speaker
I don't even know what that is. It's a fintech, so it connects to your bank account. It'll tell you subscriptions you have, how much you spent this month, does it compare to last month? Like a new Mint Mobile, or not Mint Mobile. It's a new Mint. Yeah, yeah, yeah. Oh, yeah. And I think what Mint went under, right? So they're capturing a bunch of people who left Mint. Yeah, so exactly. And there's a bunch of them similar, but this is the one that I found. But it probably actually works the way Mint was supposed to, or the way we were hoping it would work.
00:11:24
Speaker
Yeah, no. And it was one of those where like, two, I love like, how long do you have to wait before you get your aha moment? You know, like before somebody really time to value, like before somebody really sees value. And it's like the minute I got it and downloaded it, it caught a couple of subscriptions that like I had been meaning to cancel forever. So it like paid for itself in the first like five minutes.
00:11:46
Speaker
But why do people use, one example we use is why do people use an app like Rocket Money? Some people it's to save, some people it's to feel in control of their finances. There's a lot of reasons why people use an app like this. And then there's also, some people are saving for, why save? Is it for your kids for college? Is it for vacation? And when you can suss that out and figure out exactly why somebody is using something, exactly why somebody's engaging,
00:12:13
Speaker
then speaking to that is so much more powerful. Because now I can talk to that person with the thing that they care about. So what we do is like a one-to-one agent. So this agent would watch me, would watch the actions that I do, would serve up recommendations, see if I click them or not click them, and then learn from it and use that for new action. So if you send me something that's like, for example, right now they have something on there for car insurance. I'm sure they have a partner that's car insurance, save money on car insurance, it's right in there. It looks like the app, it doesn't look like an ad.
00:12:42
Speaker
But I'm sure they have a partner that they try to drive business to for car insurance. OK. If you see that I'm not engaging with that, should you not replace that with something else?
00:12:51
Speaker
You know, like right now it's just static on there, but like I, I pay way too much for car insurance, but I live in the middle of nowhere where they're a deer and I hit them occasionally. And I really like my car and I'm not going to switch. So it's like, if you learn that because you test and you prompt and you see, and now I changed the product for you and hope you interact with this and not this thing and okay, I can learn that at a user level now.
00:13:13
Speaker
Whereas previously our products were very fixed. You know, we had like fixed categories and fixed things like that. So that's, that's where we see the future comes in this, like I said, intelligent distribution beyond just content creation of like, I have infinite content. I don't have to put an auto insurance ad there. I can put anything there. What do I put there? That's almost like multivariate testing down to the individual level. The problem with multivariate testing is that you need so much data in order to prove that one's better the other, but AI.
00:13:43
Speaker
I'm guessing your AMP is using what it's learned across the board from people. So it's using some of that data. But it's also using just some common sense, like, hey, they've seen it three times. I don't need more than that to prove statistically they're probably not going to click on it. Because for this type of ad, I know a bunch of other people, if they didn't click on it in the first three times, the likelihood they're going to click on it 20 times is unlikely. So I'm just going to switch it to something else now.
00:14:11
Speaker
Um, do you, do you give, is that how it works? So, so this is actually really funny. So I work with pretty much all data scientists and I assume, especially when they first hired me, it was like all data scientists. And you would assume that there would be all over statistical significance and they're not like they could care less about, I shouldn't say that. Well, maybe I should about statistical significance. Here's the example though. Here's the example. If I want to know what my wife asks wants for dinner, I can go ask my entire neighborhood. Right.
00:14:41
Speaker
And beyond the whole neighborhood, I could go ask my whole city. I could ask the entire world what my wife would like for dinner. Like how much statistical significance do you want? But I could just go ask her and the answer's probably more accurate.
00:14:54
Speaker
So really it's about the quality of your control group, the quality of your sample, I should say, not statistical significance. And really the rate at which you have to make decisions to personalize like this statistical significance doesn't make sense. So right now it's like, should we do a green button or blue button? And we put up green button for some and we put a blue button for some. We run this test for a long time and then we say, okay, X amount of people have seen green button or blue button. The blue button has a 6% lift. So we're going to go with it.
00:15:20
Speaker
And there's a number of problems with that. One is we've shown people the bad example forever, right? Cause we're running this test. So we have to have one is the loser. So you've just shown the loser to a bunch of people for a long time. Secondly, like just because you didn't know it was a loser. So it was a great system. Yeah. Well, I mean, I've seen some pretty big games with CRO before across marketing channels, but of course you can only do it on certain things in certain ways in order to get the gains from it because, but like deeper down in the funnel where I wish it would work, there just was never enough.
00:15:49
Speaker
Traffic down deeper in the funnel for it for you to be able to do it even though you could be killing yourself down there You know, but not so traffic. Yeah, and the second that well Yeah, and the second thing though is if it does six percent better like is that is that really better? Are there people who like the other version like who you're not serving? Yeah, because you have to pick one or another so what we do is we have a whole technical paper on it, but we do a one-to-one control group and
00:16:14
Speaker
matching. So the point is when you have enough data, you can find people who are similar enough. Because again, if I'm not just collecting points, when you do that test, that basic AB test, you have like no data on visitors pretty much. Sometimes that's a problem. Cause it's like, sometimes am I testing mobile users or web users? Like I've seen things swing because you built something, but you built it for web, but is your testing it on mobile users too? Like by accident.
00:16:39
Speaker
So when you have all of this data, you send the agents out and say, find me a duplicate user. And it depends on what you're testing. If I'm testing something like what time I send a message, I want two people with incredibly similar time usage patterns. If I'm testing like a specific product, I want people who have the same purchase history. Like, you know, you've bought the same products. If I'm testing, right? It depends what you're testing, what you want. You grab two users and you test one, not the other.
00:17:05
Speaker
I messaged one, not the other. Did it work? Did they respond? Did they not respond? And then I'm continuing to update a probabilistic model for each individual user as I'm going. So it's like, you know, I want to find people with the same proximate probability that they will respond if I send something and then I send this and did they respond or did they not? And I test the other one.
00:17:24
Speaker
So we're doing this one-to-one control matching, which is how we don't need millions and millions of users. We just need to make sure that in your inventory, if you want to call it that, there are users who exhibit similar patterns for the thing that we're testing. Can you do this on websites without people logging in or is this only working in a login environment?
00:17:44
Speaker
be, we need it over us. There are two things you can learn. Like, um, so we, we talk about cold start problem. Are you familiar with the cold start problem? Like, what do you recommend somebody who's never done anything before ever? You know, that's where most recommender systems stall.
Building Personalized User Journeys
00:17:58
Speaker
You just see defaults until you do something. Once you do something, I can send personalize the most popular things across the board. Right. Yeah. So most people do. Yes. So other things that was like, are the most popular things, the highest converting things very often not.
00:18:12
Speaker
Yeah, I mean it kind of works out. That's what YouTube does every time I go to YouTube on a new browser or something That's never been touched before it's always the most Random stuff and it's like mr. Beast some other it's always whatever the heck is trending right now But in in multiple categories because it doesn't know so just throwing crap on the wall That's the most popular with everybody else which is usually nothing I would watch except for mr. Beast cuz
00:18:35
Speaker
Who doesn't watch Mr. Beast? Well, so yeah. Well, it depends for now, if you're looking at e-commerce, for example, popularity may not equal conversions. So my point is, do I want to show you the most viewed or do I want to show you the most bought or do I want to show you the highest likelihood of buy versus view? Like fewest views, highest buys. Like I don't most people don't see this, but the people who do buy. Like there's a lot of complexity in that. But my point is by collecting all this data,
00:19:04
Speaker
we are able to personalize more right from the get-go. So we can do that because we can say things like, you do know things about people who first visit your website, even if they don't log in. And we can match you with a similar person because of, maybe it's just the time you visit. First time visit, X time.
00:19:22
Speaker
You know, or maybe it's location or what, whatever little breadcrumbs we have, allow us to personalize more than just going, Hey, this is generally the most popular thing or most of our stuff. This is why we focus app based and why we started at base is because the data you have on an app is so much more than what's available via web. And do you work mostly with apps or do you work with e-commerce too?
00:19:46
Speaker
Mostly with apps, but I mean like e-commerce apps. So we have e-commerce and food delivery and FinTech and subscription and all that stuff. Is Rocket Money one of your clients then?
00:19:56
Speaker
Not officially. Not at liberty to say anything quite yet. Oh, man. That's exciting. If you're customizing a lot on the app, there has to be parameters of some kind. Otherwise, it could be throwing all kinds of... Your app can end up looking like spaghetti if everything's a variable. How do you control your app from becoming spaghetti? Do you have to have defined areas of what can go where? And then it just kind of decides within those categories of sections of the app in order to keep
00:20:25
Speaker
there because it has to be somewhat there has to be some similarities between how you're using your friend using it otherwise when you recommend it your friend downloads and it's not the same experience then that friend will stop referring so there has to be some similarities between other using it.
00:20:41
Speaker
It is actually really funny. So there is a, one of our CEOs has a, I mean, you can do this with pretty much anybody though. It's funny. You go open the same, you go open the same app as a friend of yours and they look identical. Even if it's an app that you've used for a while.
00:20:56
Speaker
Like, except for the YouTubes, the Spotify, like the big players. Take your average one. Lately, I've been playing with like food and grocery delivery. You open it and have your friend open it. And it doesn't even matter if your order history is different. It's like the same. Except for like maybe one recommender slider might be different. My wife was looking at Thrivecart and it had Mediterranean diet. And I'm like, oh, have you like, I don't.
00:21:19
Speaker
I like Mediterranean food, but we usually don't cook it. We don't do it. I say, have you done this? Why is this recommended for you? And she's like, I don't know. Maybe based on my history, blah, blah, blah. I went and downloaded it to Mediterranean diet. I'm like, it's just the flavor of the month. You know, so half the time what's recommended isn't even really recommended. So the parameters, this is what's very interesting to me.
00:21:39
Speaker
When GPT came out, we added an option that would let you, so we were more heavily focused on messaging at the moment. We added an option that would let you generate all the messages that you would want to send somebody. And we even put permission controls in the end. So like you will approve every variant of every message that goes out. So you don't have to be worried about, you know, if we're going to bogus send out, you know, either gibberish or something offensive or something. Nobody uses it.
00:22:07
Speaker
Nobody uses it. So the constraints that we have, it's like self-inflicted by users. They want to write all their own content. They want to create it. I imagine it'll change in the future as trust with AI improves and stuff. But right now how we work is the agent watches the user and then it determines when should we send a message based on usage patterns with an app? When do you log in? When do you look? What do you tend to click? When do you tend to click? All that stuff. It figures out when.
00:22:36
Speaker
Then it figures out what channel. Should I SMS? Should I push? Should I whatever? You know, these are all decisions that humans have made with flowcharts before. It doesn't make sense at an individual user level. My wife clicks SMS like marketing SMS messages. Any brand that sends me an SMS, I uninstall immediately. Right. We're two very different people. Don't treat us the same way. Also stop sending them to my wife because then she buys things. But.
00:22:57
Speaker
Then once we determine the channel, so we have the timing, we have the channel, then we determine what message should you see. Or for example, in the product experience, if you log in, what should you see? Right now our safeguards, our guide rails, are that the brands are generating all those options.
00:23:13
Speaker
So the advantage is you basically build a content library and it can even be like your blog posts and summaries to your blog posts or product descriptions or the way you talk about your product or what should I even just send you period messages about for food delivery apps like for breakfast or lunch or dinner. You can generate those but instead of like fixing them in space, so like here is the message that I send to everybody on Tuesday at 8.30.
00:23:39
Speaker
The agent says, okay, I think this user responds to, you know, Hey, this, this guy, this guy is a cheapskate. Like he loves to save money. He doesn't want to spend money on anything. So when I talk about, Hey, you should buy lunch. I should talk about the value of buying lunch and how it's really saving you money instead of.
00:23:57
Speaker
just getting crap out of your pantry and blah, blah, blah, blah, right? As opposed to somebody who cares about health as opposed to, so the agent then dives into that content library, pulls the content that the brand has made that is available and then populates it. So that's how we prevent one way. We prevent it from, you know, your product experience being.
00:24:16
Speaker
completely uncontrolled compared to the person next to you. We're not even close to people being able to push it to the point where it can really change a lot across the app yet. Your users are just plugging in into little sections of their app and no one's like,
00:24:27
Speaker
giving it full throttle control. I know you're not in the course of being worried about yet. Yeah, it's interesting though because you can make a huge impact by affecting little things. I mean, we've seen this, right? That's what AB tests show. Yeah. Except for what I've learned is, it's cool. I've tried to write this up in a post, but it's hard to write it up in a way that like,
00:24:47
Speaker
makes sense as opposed to talking through it. An AB test can give you a step change. I was here. I went up to here, right? Because I like, here's my conversion rate and then here, but then it stops. It doesn't like, it's, it's a step change. I was at 8%. Now I'm here.
00:25:00
Speaker
Right. And now I do another AB test and I hope I get to here, but I probably will end up here. Maybe I'll end up here, you know, like I'm basically like walking upstairs and the thing that stops me from the next step is me. I have to write another AB test. I have to do the next thing I have to test and then I'll move up to the next step. What AI in tweaking little things allows you to do is really create an exponential curve. That's improving over time.
00:25:24
Speaker
Because I'm, I'm testing, I'm testing, I'm testing, I'm testing and the individual user level incrementally to improve it. So the difference is, like I said, if we did a food, if you did a, if you did a workout app, people care about working out and they're passionate about working out. Why?
00:25:39
Speaker
Well, some people it's because beach season is almost here. Like, so I need to look good. Some people it's because they deal with chronic illness. Some people it's because I need mental clarity for work and I just, I feel more, you know, if I'm working out and I feel sharper, there are so many reasons why someone works out. And if you can tap that reason, like all of a sudden they're incredibly engaged.
00:26:03
Speaker
Because, oh yeah, oh yeah, oh yeah, I do have that wedding coming up. You know, I do need to look good for that wedding or high school reunion or whatever. And then the kicker is though, once that event is over, if it is event driven, then somebody like tunes out, right? Well, I did the wedding, that was fun. Like, so how do I keep that user engaged? Now they've changed, right? They're no longer the event driven person. They're the, hey, wasn't that great? Didn't you notice your job performance was better? So even by tweaking little features of an app,
00:26:31
Speaker
little headlines or what blogs we recommend or what thing to align with those user values. Huge impact. Huge, huge. So you don't, it's kind of funny. You don't need like every single element of everything to be incredibly personalized. Although I agree with you. I think it's going that way. Um, at this point it's funny, those, if I can catch those softer user preferences and pivot to those, like you already see massive, massive improvement.
00:27:00
Speaker
What's the minimum viable audience you need on an app? Let's say like a weight tracking app of some kind. Like how small of an audience do you need for this to be viable? Or how big does it have to be?
00:27:14
Speaker
Yeah, I mean, you just need enough redundant users that you feel like you can make an educated decision, you know, that the AI feels like you can make an educated decision on a user because that control group we talked about. So the smallest apps that we work with are like 5,000 to 7,000 users, which is pretty small. And monthly active? Yeah, and then our larger are, you know, 20 to 50 million monthly active users, 100 million monthly active users.
00:27:41
Speaker
So not what we've seen is it just takes longer if you have a smaller group because there's less learning, but you really have to look at current state. Cause again, the current state of like, I just blast everybody one thing. It really doesn't take much to improve that. What I love about what you guys are doing is that you're bringing like Netflix level customization down to like smaller players. And we all want that as users of all these apps. We all want that.
00:28:07
Speaker
We're tired of seeing stupid messaging. We're tired of seeing opening up a dashboard for a tool and you being like, like how many dashboards are actually relevant? Almost none of them, right? Because they just, they're not, unless you can customize it to what you want, they're generally bogus dashboards. I mean, when's the last time you saw WordPress's dashboard? Never. Because you never go to that page. No one does. No one ever has. It's a bogus dashboard. But a lot of apps are like that, where it's like, they just random stuff on there. It's not, doesn't matter to you.
00:28:37
Speaker
So that's that's exciting. I'm excited to see where that that goes. And where you guys are at with it now, because I thought personally, I thought this level of thing was going to be like maybe a few years away, but you guys are already crushing it, but bring it out there live. So yeah, my next question for you is the head of growth for this company. I'm sure you're experimenting with a lot in order to drive amps audience in order to get more conversions, drive people through the pipeline. Tell me a little bit about what your experiments look like internally as the head of growth there.
00:29:07
Speaker
For us, it's actually really funny. We're finding that we've created things that are products in and of themselves, but we didn't realize it as we created it.
Marketing Challenges for AI Solutions
00:29:17
Speaker
So one thing that's been very unique is I've worked for a lot of like traditional industries where the market exists.
00:29:23
Speaker
This is one where when we stepped in, there was, there was not really anything like it. And it's funny because we were, you know, we were pre GPT and the data scientists brought me on and were teaching me all this, all of the new kind of things I wasn't familiar with. Right. And it's data science is like a whole nother language, but I found it to be fascinating, quite fun. So their marketing teams were very, very gun shy about AI.
00:29:49
Speaker
until, you know, like GPT hit and it became like accessible. Oh, this is a thing, you know? So before that, it was like total trailblazing. You know, a lot of it has been educating users for like, what do we do? And where does, where do we fit? Like, it's like, you show up with an automobile and all the people who are, I said, automobile, I feel old. You show, you show up with a car and all the people who like shoe, the horses are like,
00:30:13
Speaker
What do we do now? You know, and you're like, I'm not getting rid of like an industry here. You have just different work. So we've talked about, like, for example, you brought up Netflix, but how Netflix has taggers, annotators, people who do tax product taxonomy and how product taxonomy, how do we classify certain things is so much more valuable in an AI world than it was pre AI. So a lot of the challenges that we face.
00:30:37
Speaker
are like, how do we help people understand what this new world looks like? Which is like this conversation that we're having, right? As opposed to the world that they were used to and how is there a place for you in it? Um, and things like that. So that's what category does it sit in? What category does amp take up? Yeah. So it's, it's funny because at first we,
00:31:00
Speaker
thought we were more of an engagement platform. And why we thought we were more of an engagement platform is because typically in modern companies, that's where you build your flowchart, you know, of like day one, send this, think of any, like even people who are just used to like sending an ESP, right? You're just sending out emails. You segment your audience and you say day one, send this email, day three, send this email, day five, send this email.
00:31:24
Speaker
So we thought, well, we're in that gate because we do take that information and we do send the message. We don't actually send it. We connect to whatever service you're using that sends it, but we make the determination of what should be sent to whom. We put ourselves in that bucket. And then we learned to actually, it's not the right bucket.
00:31:42
Speaker
Because again, there are engagement platforms already that do the sending. We don't do the deliverability, the physical sending. Semi recently we found out we are actually in the customer data platform space. So are you familiar with the CDP? Vaguely. I mean, I've definitely heard of it. I'm bumping around in tech. Yeah. Yeah. So, so for those that aren't.
00:32:02
Speaker
What a CDP is, is basically a place to aggregate all your customer data. So you have a data warehouse where all of your data goes, but then from the marketing side, you want to have specific data about customers. And then you can run queries to say, okay, I want to send a campaign out to everyone who's visited in the past week, but hasn't bought anything in the past week.
00:32:20
Speaker
right that's really common what should i do should send them all discount or something so i go to my cdp and i build that audience give me everybody who hasn't visited you know hasn't bought in the past two weeks but has visited great and then i use that to send a campaign. What we find out is that we do that but at the individual user level.
00:32:37
Speaker
And then we will say for this user, send this message because yes, they haven't bought, but they typically buy when they have a discount as opposed to send this message because this user hasn't bought, but they typically care about quality, you know, or like high reviews or whatever. And so we've found that we fit in the, we brand ourselves as an agentic CDP. So we fit in the CDP bucket, but whereas traditional CDPs allow you to query.
00:33:01
Speaker
We see the query as one of the biggest constraints. You can only query one thing at a time. When you're writing a query to find an audience, you're missing out on all the other queries that you could write. And then we're making the determination of what to send and feeding that to the engagement platform. I've always had a hard time with the whole premise around segmentation. Like I understood it, why it was important, but in execution, I'm like,
00:33:24
Speaker
I've worked across multiple marketing teams now. Nobody does this because it's a freaking pain to like try to figure out what even these segments want. And if you're doing it regularly, you're doing it haphazardly with no end goal as far as what you're doing to like think about the journey of how somebody comes in and what they're doing and if they've lapsed what happens then.
Integrating AI for User Experience Personalization
00:33:43
Speaker
So I've generally been one who's like customize the heck out of drip campaigns, right? I was a huge Infusionsoft user before they started to suck.
00:33:50
Speaker
Where I had this very complex onboarding with all these variables of what they would get or not get and then what would happen at each stage and if they disengaged at a stage, what would happen, where they would go.
00:34:02
Speaker
That always made more sense then that was kind of like the inbound marketing dream that HubSpot built hardly anybody does it that way but it always made more sense to me than segmenting because you build it but it's like I kind of see where you you're going where you guys are going with amp is like the whole new level from that like that was good but It's even better if you could just personalize it to the individual instead of mapping it to the journey
00:34:25
Speaker
Because different people, as my friend Ashley Foss says on LinkedIn all the time, she's always like, dude, nobody goes through a straight funnel. But I'm like, yeah, but it's better than sending out random segments, reacting to some down, like you didn't make enough revenue. So you're going to go pull that random segment where you know you can get some revenue, because that's just reactionary. Before the journey, it was at least more predictable, because you could run people there through there as cohorts and even split test it and do some cool stuff. But
00:34:52
Speaker
With AI now, I'm like, oh my gosh, like this is, it's like a whole new game. How hard is it to build though? How hard is the AI to build? Yeah. Cause build it like for me to build all these journeys. Gosh, it took a lot of time. You build one, but it was worth it because you'd write a five email sequence and maybe you had a few different paths that people could get in there and where it would go. Great. And it was worth putting a lot of time and effort into it because that, that journey might last for a couple of years, you know?
00:35:20
Speaker
And it was worth it. It wasn't a one-time thing. It was an asset. But how long, but it took a long time to build it out for every segment for all the contingencies that they fall out of this part. Here's how we're going to re-engage them here. If they're in this segment, like how we re-engage them there, depending on what journey, part of the journey they're on took a long time to build from years to build it all out. How long does it take to like really get AMP included in all the different sections that you would recommend for customers? Yeah. People, people laugh, but it's like a couple of weeks.
00:35:49
Speaker
integrations, integrations cake. But here's, here's, here's the difference though. It's like, we feel like, or we think we know the right thing to send even for like a first email onboarding. Hi, welcome. Take a look around here, all the features we offer, right? But it's like people pick our product or people pick an app or whatever, for very different reasons. Like why do I use the Lowe's app? Cause I do a lot of handy work.
00:36:13
Speaker
Why do I use the Lowe's app? Because I want to know what aisle something is in. I never buy it on the app. I want to feel it in the store. I just want to know what aisle everything is in because when I get to the store, I like where's the specific tool and where the heck is it? Like that is the only reason.
00:36:28
Speaker
that I use the Lowe's app because I want to figure out and it's in the store. I only use it when I'm physically standing in the store to go, where the heck is that? Like that's the only time I use it. The reason why people use things is so drastically different that building a welcome campaign or even a re-engagement campaign, like it's so specific to the person that to me, I get like paralysis, like decision paralysis. When you can send somebody everything, what do you send them?
00:36:52
Speaker
So what we find is it's, it's fun. If there's a clicking point with new customers, that is so much fun to watch. And I don't know if you've had this with your other jobs, but when they start using AMP, you're like, okay, listen, you have to break them in that mindset of like day one, this day to this day three, this, no, it's not the sequential thing. It is what do you want to tell somebody? Why would they like your app? So like we worked with a really small climate friendly cooking app. It was a cooking app. The goal of the app, even though they're not like,
00:37:20
Speaker
overt about it. What I mean is they're not like, you know, every meal doesn't have a carbon count or anything like that. But it's just like, this is really good food and the recipes that we have are climate friendly. Why would somebody use that app? Some people it's because of climate friendly options. Some people it's because it's majority vegetarian. So it's like maybe they want some more variety in like their vegetarian offerings.
00:37:44
Speaker
Maybe it's people who want to see what vegetables are seasonal. Maybe it's people who just want to be more healthy and they know that this food is generally more healthy. Like you have a whole spectrum of people. So when we start working with people on messaging and product, we're like, the goal is find those people. And this is a good, you asked about things I use GPT for all the time. The first thing I do is I say, give me 25 reasons. I asked GPT, give me a list of 25 reasons why people use whatever app. And if the app is too small, I give it a competitor.
00:38:14
Speaker
You know, but it's like literally just do that. And it'll give me literally 25 reasons why people use every single app. And if you compare that to what a product current, what a company's marketing department currently runs for their marketing, they speak to like one, maybe two of those. You know, they assume that.
00:38:31
Speaker
The reason that you use, you know, I'll just say rocket money, for example, is because you want to save money. Yeah, but maybe, but maybe also not. Right. Some people, it's that feeling of control, like we talked about. I want to be in control of my finance, safety, security. It's this, is that. Or maybe it's like investing. I want to track how my investments are doing, not my savings. You know, some people it's like I have a bunch of debt. Some people have no debt. Right. So anyway, I pull those lists of 25 and then we start writing messages for all those 25.
00:39:02
Speaker
like 25 different personas. And then what's interesting is like sometimes what you would think of as a win back, like so a customer churned, I should send them a discount.
00:39:10
Speaker
Well, maybe that's the exact wrong thing to do for a customer. I should send them a product recommendation. So in other words, you're not with AMP, you're building those different messages and building those different experiences and all that different messaging, but you're not having to attach it in a rigid flowchart. And that's where all the time is taken up, is determining day one, this is what I'm going to send day two. And there's all this pressure of like, I have to write the perfect headline. I have to write the perfect subject line. I have to write the perfect message.
00:39:39
Speaker
You don't cuz any message you write is perfect for someone it's finding the someone that is perfect for so you also see a lot less people are less intimidated about having to be perfect because the AI for we have had a lot of examples where the AIs like that message is a dud
00:39:56
Speaker
And you can see it tries to send it to a few users and like nobody ever clicks it. So it'll test it occasionally, but for the large part it gets binned. And it just never uses that piece of content anymore. So what we have is, you know, usually people, like I said, customers hit that clicking point where they're like,
00:40:13
Speaker
I could do a campaign for this now, or this now, or this now, and creating it doesn't mean where do I slot it. Do I slot it on day five? Do I slot it in for these type of users? Do I pull it? No, you just dump it in the library and let the AI tell you, hey, this is really good for, you know, this specific users. And that's just informational. Like, oh, that's cool. But you just continue to let it do its thing.
00:40:34
Speaker
So yeah, so it does, it takes a fraction of the time. We've had so many people surprised at how quick it is because a lot of where you spend your time is like, should I send it at 8 AM or 6 PM? Or should there be a four hour delay between these two or how long, you know, and then AB testing of should I, should I send it 8 AM or 6 PM? Like I'll set up an AB test. Should the color, should the button be green or blue or what? You know, a lot of that you don't have to do anymore. You just have to create like new ideas for campaigns, which is honestly the fun part.
00:41:05
Speaker
Other than AMP itself, what are some of your favorite AI tools to use, maybe besides chat GPT too?
00:41:12
Speaker
I was going to say GPT Dolly. Um, I use a lot of the, a lot of the big and basic ones. Um, on my side, I haven't, I'm trying to think of some of the best ones. Riverside. I know where, what I, what are we using? This is on Zancaster. This is on Zancaster Riverside. I am so impressed with like, as far as generating clips for podcasts and summaries and I'll say, hopefully you're not being sponsored by Zancaster, but that.
00:41:38
Speaker
So I was thinking pre AI, like we, we tried to do a podcast and it was so much work. Like I know I'm preaching to the choir, but it's like, you know, pulling clips and trying to find the best and then the transcripts and then this. So I was using like VEED to make the transcript for me. So I'd have to export everything from Riverside to VEED and get the transcript and all that stuff. And Riverside, you know, most, most things when you're like, here's something of interest. Like I want to pull a clip of interest. Go find me something of interest. Like they're garbage. Riverside.
00:42:06
Speaker
does a phenomenal job.
00:42:09
Speaker
That's good to hear. Last I checked Riverside stuff and it was only three months ago. I thought its clips were bad. That's like, but you know, these things, they get better with time. So yeah. Interesting lesson though. It's, it's like, you don't have to be perfect. Like with a lot of AI stuff, and this is a message for Riverside, whoever too, you don't have to be perfect. You just have to allow customization. Cause if I watch that clip and I'm like, wow, it's not great, but what was just before it was, it's always like changing just the slight tweaks at the front of the end of it. Usually.
00:42:39
Speaker
Yeah, if you can't, then it's trash. But if you allow people to like tweak it a little bit based on what they like, then it's good. The other one, just for fun, is like Opus clips. I don't know if you've used them. Oh yeah, I've used Opus. I mean, for clips, Zencaster has its own clipping tool, which is my favorite. But it's not my favorite because of the AI. It does a good job of picking the clips. The thing Zencaster did that was better than everybody else is it allowed me to one-click publish it.
00:43:04
Speaker
Because if you have one good clip, well, usually after an episode like this, I'll probably have like five, seven clips. Great. It's freaking pain to go and take each clip and schedule it across at least three or four or five platforms. But Zencaster is just like publish now or auto post, which kind of schedules it out, kind of buffers it.
00:43:22
Speaker
I'm like for that alone I like zen casters cuz I could just be like archive I can edit or I can put I can schedule it put it in the queue and then I forget about it just puts it everywhere it also pre-rates the copy for it too I was gonna ask is that decent or is it kind of crap
00:43:38
Speaker
the pre-written copy. It's at least customized for the clip, but no, it's not what I would write. It's generic vanilla chat GPT generated based on the little tiny transcript from the clip. But you know what? I don't care because who freaking reads that on Instagram, YouTube shorts, or TikTok? Nobody. But it needs to be there.
00:43:57
Speaker
Yeah, yeah. No, but that's what's interesting for me is it's like all the content, somebody recommended, what was it? It was one for emails and they're like, make your emails better. And I'm like, yeah, kind of garbage, kind of like LinkedIn's like, try rewrite with AI. I feel like in some ways the content creation tools exist have like ruined actual AI for a lot of people. Yeah, I've tried it. I've tried AI, it doesn't work. And I'm like,
00:44:21
Speaker
You have not. No, you just click the rewrite my post. You gotta learn how to prompt better. That's why I'm making the little course that I'm making right now to help people like you just haven't.
AI's Role in Content Distribution
00:44:32
Speaker
It's there's a skill to it and people who spend more time on it get better results with AI.
00:44:37
Speaker
Yeah. Yeah. But to your point too, it's not just generation though. This is what's really interesting to me. Content generation is certainly a piece, but distribution is what matters. Cause if you had something that instead of auto schedule posts was like you knew who the audience was that was most interested and what time to show it to them, right? You're leaning on YouTube's AI and Instagram's and everyone else's to make those content determinations for you. Right. But if you had your own audience, if you had your own,
00:45:04
Speaker
Dan Ches empire, you would want to create that for yourself instead of like, here's one thing. Cause another point is we see this with companies all the time, content reuse, something you recorded like a year ago might be exactly what somebody needs right now.
00:45:17
Speaker
Yep. Gosh, that, that disturbs me more than anything else. Like the fact that I've written some killer content two years ago and it's just as good today, but nobody even remembers it and it needs to come back. That destroys me. So we put, this is one of my favorite applications. So, and it's such a quick win and I just love it. So we'll load in blogs, a company's blogs into their messaging.
00:45:43
Speaker
So this works really, really well. It works for everybody, but for subscription apps, like the workout app I brought up. So they have, so as an example, one workout app has like exercises, like how to do certain exercises. And then they have blogs on like different body parts, you know, so like some of them are exercise based, like how to do a squat. Some of them are like leg day, you know, like a body day, but then they have other ones on like nutrition.
00:46:09
Speaker
These are all, to your point, really good, valuable blogs. Nutrition hasn't changed, right? Except for if you want to get tinfoil hat, we can talk about why the food pyramid looked like it did back in the day, because we were told to eat bread nonstop. Anyway. Well, I've heard those stories, too. And then I live next to Battle Creek, which is Kellogg's home base, so I know all of it, too. But then there are also ones on pain management.
00:46:33
Speaker
Right. And there are ones on like training for marathons. My point is they have all these different, they have all this great content. So what we say is loaded into messaging, write messages that link to those blogs for different users for different reasons. And their blog traffic takes off.
00:46:48
Speaker
Because I'm reusing these blogs that are like, I don't care if it's four years old, it's still applicable. Because when you're looking at a subscription app, you're like, how do I keep people engaged, right? So if you're a workout app and somebody gets injured, they stop using their workout app. So you could build in some logic that, hey, if we have a user and they were active and then go dormant for whatever, then send injury blogs.
00:47:09
Speaker
But that's like, that takes work to set up that whole infrastructure. Instead, RAI might learn, Hey, you know, we've seen this pattern in other users before, and now I'm seeing it with this user. So therefore start sending injury blog and we start to see traffic tick up on all these different blog types and stuff. And then people get re-engaged because again, if you hurt yourself, you might think, well, it can't work out. But then if I can give you a tip on like how to get back to the gym faster, because you know, here's how you deal with shoulder recovery surgery. Boom. Everybody wins.
00:47:38
Speaker
Man, I can't wait for that to be available to small players like me. I've made so much content. Blog posts, I've got hundreds of podcasts, like hundreds of podcasts. I mean, not even just on my own podcast, but I could pull them all in because they're all out there. They're all danches.
00:47:56
Speaker
They're all team dentures. Yeah, exactly. There's no way to currently do that. I have a membership area where I put all my courses into my best content, my best playbooks, but there's no way to even drip it out via email or text message or whatever the heck I got people subscribed on. There's no way. That's not currently possible with what I have, but I wish it were.
00:48:19
Speaker
There's so many quick wins too. Just like redundancy elimination. Like if I'm tracking per user, then I know, okay, I've already sent this and they didn't respond at an individual level. Cause right now you might say, Hey, this was great. And I wrote it two years ago and I'm gonna send it again. And somebody might've been like, Oh, I've seen that thing. You know, which is always our fear of like alienating one person. It's going to happen, but either way. Yeah.
00:48:39
Speaker
Yeah, to me, that's like the quickest win that is so much fun to do is taking, you know, every apps. Are you, are you using that for your own marketing then? You're eating your own dog food and using your own tool.
00:48:50
Speaker
I wish so much, I wish so much that we could do it. I said for prospecting, we need to turn it into a prospecting tool. Cause it's like so good at like customer engagement and re-engagement for these big apps with content. I'm like, we need to apply the same thing to like outgoing sales, emails and stuff. But there are already players in that market. So, so we get to live. I just mean there's enough people who are advancing in that area.
00:49:14
Speaker
Like in sales prospecting that they'll probably be there. Our niche is for like cold outreach. I'm thinking about warm outreach just for newsletters. People like in your audience. I mean like audience plus is kind of working on this but.
00:49:29
Speaker
Not like that. Not like how you guys are. I think you'll start to see it. Yeah, it's reinforcement. So it's reinforcement learning and it's contextual banded algorithm. I think it'll get much more common. So here's what was interesting to me. So I have an electrical engineering background. Long story.
00:49:47
Speaker
I did medical device engineering for 10 years. I won't go there. If you look at who's built Martek, like look at who builds almost any Martek company and it is computer scientists, computer engineers, computer scientists. And once you see it, you're like, you can't unsee it. Cause it's like, I had computer science courses. It looks like a computer science interface. If this, then that right. That is.
00:50:08
Speaker
straight out of the playbook and that's how all of them are built. So now we have this like wave of data science. Data science has always been there, but it's been like this, like nice to have like, okay, here's a cute little project. Can you work on this data science team? Now it's starting to be more dominant that people are like, wow, there's okay. There's some serious depth of things we can do in data science here that are well beyond if this, then that you're starting to see the next round of tools are tools and example founded by three data scientists, PhD, anthropologist, PhD, neuroscientists, data scientists.
00:50:38
Speaker
Um, they don't build things like computer engineers. So it's like the interface is absolutely totally different. Um, the, the way that it acts is totally different. So I think if a sales prospecting tool, or I've even seen this for like, I was looking at HR, cause we do churn prediction and churn prevention, but instead of like the difference, there's always this like cliff and it's like churn churn prevention. All right. Here's the users who are churning where the cliff is is like, okay, so send them all this one message.
00:51:07
Speaker
You know what I mean? Like, no, like you want to know if somebody's churning and churning is different for each user. Some users churn after their first visit, you know, and you can tell because like, oh, they've done this action, this tends to be churning. Anyway, you're going to start to see these tools pop up in other areas, I'm sure.
00:51:23
Speaker
that are built by data scientists with a data science methodology and mentality that are so much different than the ones that are built by computer engineers today. So I think you're going to start to see it everywhere. It's going to become a lot more normal. And it won't look like the flow chart because the flow chart was the best we could do at the time with the computer engineering background, you know, creating, creating these things. It's going to look different for every industry. Yeah.
00:51:48
Speaker
My mind was instantly like, Oh, HR. Oh shoot. You can track employee retention and all that kind of stuff. But then my mind went from there really quick to like the government using it. Oh my God crap. We're going to be in like minority report status.
00:52:00
Speaker
Yeah. Oh, I can't go there. That's not what this shows about. I'm all about the more practical stuff. Like what can we do now rather than theorize about the future? But I do. Yeah. HR is an interesting one because I've looked into it too. And the data that they have, this is the problem you see is the data that they have is all over the place. It's all in different systems. It's not correlated to anything. And it's it's but it is this is a data science problem. Right. This is why things are happening the way that they are. And data science has taken over in a lot of ways because it's step one is
00:52:30
Speaker
you need to actually have data that you can feed into a model that can then produce findings and results for you.
00:52:38
Speaker
Yep. It's like all the data was always there. I mean, remember the talking about the fire hose back when Twitter was like a thing, like the API was open to everybody. Everybody's like, we don't know what to do with all this information. I'm actually really excited about Grok, the AI, because of what it will be able to do with that information sometime in the future. I just tested it out last week and paid for the premium for a month, just to, just to test Grok. Because I'm like, theoretically, like it has the best real time insights into what the heck's going on over the world, but it's ability to find the right thing.
00:53:08
Speaker
Not very good. Yeah. For me, it's funny. I'm pretty confident
Future Impact and Broader Applications of AI
00:53:12
Speaker
it will be someday. So I'm paying attention to that. For me, it's like, just watch though. Watch though. The agency is where we're going to see the next shift. So I went, I went to a data science conference a couple, maybe like a month or two ago, just cause it was near me in Michigan and nothing is near me. So I'm like, I'm going to go to this. The output, the holy grail is a dashboard. So like all the data centers are all talking to me. What do we do? We build a dashboard.
00:53:37
Speaker
like for management, like literally. And I'm like, wow, this is going to change so fast. Cause again, our chief data scientist always says, you're never looking at a minority. You're looking at the, or sorry, at a majority. You're looking at the largest minority. So anytime you do an AB test, anytime you do anything and you're like,
00:53:53
Speaker
This is what everybody loves. This is our most successful option. It doesn't mean it's the majority of your customers, users, whatever, who like it. It means that it's the biggest minority that you can measure actually like it. So any dashboard you produce, any findings, anything is always at the aggregate level. It's either at the aggregate level, which is like useful, a lot of missing, right? Because if I say this is great for 8% of people, that means I'm missing 92. Or it's at the individual level where we as humans can't
00:54:20
Speaker
I can't do anything with that. I can't do anything for one person, it just doesn't scale. So what you start to see is now if I let AI take over, then the individual matters and the AI can act on the individual level and just report to me the aggregate, sure. But the dashboard is just a helpful feature to see where trends are going as opposed to like where I'm making like super huge broad sweeping decisions that affect everybody. Dashboards give you the illusion of control.
00:54:50
Speaker
Yes, but only I've made useful dashboards for myself and other people like looking at them, but I'm like, you don't even know what you're looking for. I do because I built it. I know what I'm looking for and why I built it. But generally, most of the time I see executives looking at dashboards. I'm like, you don't know what you're looking for. Yeah, you just like the report on it. And most of them are like these like like stupid, stupid like widgets that I'm like, that tells you nothing because you need the context of the number that came before it.
00:55:20
Speaker
and the history of it and even to be able to make a relative guess if that number is good or not. So, oh well, it is what it is. But I'm excited about where it's going with AI and being able to turn it forward. James, thank you so much. It's been a fun, nerdy conversation to dive into about personalization, man. I've learned so much about
00:55:35
Speaker
Where it's at now where where this could be going in the future? I'm just hoping what you guys are doing like trickles down a little farther It comes out a little bit more cuz I'm excited for it to hit all these apps because it's it's coming And what you need you have anything going on that you want to promote? Like a webinar or a video that people should watch to learn more about AMP
00:55:54
Speaker
Yeah, I can say we just spoke. It was a lot of fun. We just spoke at MAU, which is mobile apps unlocked, which is a big conference. And even if you don't have an app, we dove into Spotify's how Spotify does their growth loop, right? They have 6,000 genres. So how they.
00:56:09
Speaker
Why do they have 6000 genres? Why does that make sense? And we dive into Netflix's tag and kind of all that stuff and then show examples of personalization through chat bots and recommender systems and all that stuff. So it's a nerdy is me and our chief data scientist, nerdy, but accessible. So I can, I can send you the link for that. But that was, we got great feedback. Most time you go to conference and you hear talks that are like so high level.
00:56:32
Speaker
that are like retention is important. You shouldn't let customers churn. Great. We tried to get in depth with a lot of examples and data, um, and got really good feedback. So I'll drop a link to the YouTube video for that and the show notes, but I have, I saw that video. I know it's on your homepage, so I'll make sure to link to the homepage as well. Thanks for joining me. Yeah. Thanks for having me.