#2295 Ryan Carson: How AI does my marketing for me

0:00Why isn't every email customized for the customer so it's completely personal? So I'm going to share my screen and just kind of walk you through visually how this thing works. Ryan Carson is a three-time founder and he's building Untangle, which is AI for divorce.

0:18Presented by Zapier, the AI automation company. Ryan, before you started building your own email service provider, did you even consider that you can use other people's software? uh you know i've been running startups for like 25 years right and email drips have been just a constant part of the business and everybody uses them and i you know was setting up an email drip for untangle um my new startup and then i thought what am i doing why don't i have ai write the email drips uh and ended up you mean like the series of emails yeah yeah and why why isn't every email absolutely customized for the customer that so it's completely personal like we can do this now and so i build a workflow and i'm very happy with it i see so theoretically somebody who is going through a rough divorce might get a different message on your platform and someone who's going through an easier one might get one that's that's softer someone who's been married for 10 20 years might get one that's different from someone who's just kind of got married quickly and is trying to untangle their marriage.

1:26Got it. And you're saying, I don't want to have to write 50 different drip campaigns, but all of these different people are completely different. I think I could do this in AI. Exactly. Did it work? It did. Give me an example. So I want to kind of walk your audience through how— Before you show me what you're doing, give me an example of what it has been able to do for your business. Okay. So every day I have a cron job that basically looks at key marketing stats, and then it passes those stats to Opus 4.5, and then it generates an email with recommendations on what is the one thing I should do to improve my marketing funnel today.

2:10So this happens in the background, right? And I read the email yesterday. And that's a whole other fun thing I could talk about is how do you automate insight every day? So I got the email and it said, Ryan, you have these five people that have clicked three different times on your email drips. You should probably do something about that. And I was like, good idea, Opus45. Thank you, AMP. And I basically went and said, let's build a custom email drip for those people. right and so what it did is it collects okay these are the you know five ten people that click a bunch of times let's put them on a an automatic email drip that is completely customized to them that's highly engaged and like paul graham says do things that don't scale right and so the email said very simply hey i noticed you've been clicking on our emails like i'm the founder of untangle i'm ryan and i would love to to meet with you or help you in any way i can um and that email went out and i immediately got a reply from one of those customers that was clicking on her emails but hadn't signed up to the free trial and she said i would love to meet i was like okay and so i was like do you want to meet on a video call do you want me to call you like i'm not going to try to sell you anything what do you want she said let's do a video call and we had a video call about an hour ago and I learned all about her situation and in the end she's like I'm going to sign up for the free trial and so she signed up so it's like email automation AI you know all this is beginning to work it was really fun all right let's see how you put it together okay so I'm gonna share my screen and just kind of walk you through visually how this thing works.

4:05Okay. And of course I had AMP build this for, for this show. Cause I was like, I want to, I want to be able to explain this well for Andrew and his audience. So what are we talking about here? All right. We're talking about AI personalized drip emails. The idea is you, you have customers or leads that you want to do something, you know, you, you send them emails over time. We call this email drips or sequences, but in the past, you know, you would use a tool like outreach and you would pre-write each of those emails, right? I want email one sent on day two. This is what it says, insert their, their name.

4:41I want email number two sent on day three, you know, here's the template, very primitive in the world of AI, right? So what I did, um, is build a system that uses AI to write each of those emails. So the first thing, step one is the lead enters the system, right? So whatever your funnel is, you know, for me, it's a child support calculator. That's a common way that people want to enter into the divorce process. They're trying to figure out what their child support is. So the lead uses the calculator or chat, right? Okay. So then what happens? So then we create the drip campaign plus the context.

5:20So So the agent goes in and says, okay, we're going to enter in this email address into this campaign, and we're going to insert the context for that customer. And this is the key, right? So we live in a world now when you collect these leads, you have information on them, right? Whatever they gave you, plus you can actually use tools like Parallel.ai to go out and scrape the internet to find more information. you can get a lot of context on that customer. So you store that context right in your database. And then you put that in your email drip campaign table. So person comes in, generates a lead, you store their context and you put them in a table to say that they're an email drip.

6:10And you enrich it with information from parallel. Exactly. You can. So the neat thing is like for our lead funnels, We get a lot of valuable data about that customer without having to do that. But you could use Parallel. You could use the XAPI. You could use LinkedIn. There's a lot you can do to gather enriching data for that customer. So that is step two. All right, step three. So we've got them in the funnel. We've created the campaign with the context. We've put them in the database table. And now what we do is we trigger an hourly cron job. So the idea is, are there any folks that are up for an email?

6:53Let's go ahead and grab those and cue those to schedule. This is not rocket science. Pretty straightforward. All right. So after the hourly cron job, we're going to. All right. So a couple things. We have to normalize user profile. So you have to think about, are you sending the same email to multiple people? Are they on a different email drip? How are we thinking about this specific email to make sure that you're not spamming anybody? Pretty straightforward. And then this is the key, right? So what you do is you take that user, you normalize their data, and then you take the context and then you feed it.

7:41You just make a call to your favorite LLM. At the moment, I'm primarily using Opus 4.5. And then you have a system prompt that says, here's the data that you have on this customer. Here's the goal of this email. Here are the rules. And you have to also do things like you have to specify the feature set of your app. You do not want the LLM hallucinating features for your app. That's always terrible. and then you write this custom email that it feels like it's written by a human. I mean, it's really pretty astonishing. So you queue that up and then we're on to step six. Then you use resend. So there's resend, there's send grid, there's a lot of email sending tools these days.

8:31Resend is just great. I love it. I'm not being paid to say that. Everyone's using resend. Why resend over the others? You know, it's funny. is I think they crushed it on AEO, Answer Engine Optimization. Because when I was talking to AMP about this, you know, like, hey, I want to send emails, you know, programmatically. It was like, why not resend? That's what happened to me. You're talking about AMP Code, the company. Yeah, AMP suggested, you know, ClogCode probably does the same thing. It suggested resend. Okay, got it. That's funny. I thought that was pretty great. And then using Resend or your favorite email sending tool, you want to check for a couple of things.

9:16Did they open it? Did they click it? Did it bounce? What's common today is often not to use open because it'll degrade your sending score. So we just use click and bounce. So we measure clicks, and then we just check if any emails bounce. So it's pretty standard practice. To know whether you should send the next message or to know how effective the message was? Both. And when you get click data, you can do things like I did, but you can layer another step on top, right? I see. So, you know, we had a lead come in. They used our child support calculator. Great. The daily report showed me that she was engaging by clicking on multiple drips, right?

10:04and then I realized oh I should add another layer of engagement on top which is another automated email drip which encourages them to schedule a meeting with me and then boom I got a free trial out of that today so and then of course after that email sends then you update the campaign to say it was sent and then you load in the next one to send pretty straightforward And then finally, you have a nurture loop, which basically I have saying this, once someone reaches the end of an email drip, and typically email drips are, you know, five to seven emails. Typically, they're two to three days apart.

10:46So you get to the end of that. And I can't remember who is talking about this on a podcast. I think it was Wade from Zapier. He was on Claire Vaux's show. and he said, you know, AI allows you to touch every single lead you've ever had infinitely, right? In the past, all of us humans, we get exhausted, right? Like, oh my God, we have 30 ,000 leads. We haven't emailed them in a long time. Like, what do we say? How do we make it personalized? But we live in the world of AI now. And so every single person that finishes that email drip should be put on a long-term monthly nurture drip right and that's also customized yeah absolutely customized so so the beautiful thing is customized to them customized to what you've already said to them customized to what they've clicked right like this is all very valuable data um so all you do is you feed into the llm every time all the context on the previous email drips the previous clicks and you can generate very helpful long-term nurture loops.

11:56I'll stop sharing now. The beauty though is that you can deliver real value. So everybody hates marketing email drips, right? Because they're usually spammy. They're not actually helpful. But if you're doing two things right, if you're doing your AEO strategy content correctly, which means you've created hundreds, if not thousands of articles that actually answer people's questions, then what you can do is in each of those nurture loops, you can say, you can feed to the LM, say here's everything that we've said to this person. Here's the data we have on this person. Look through our content library and find relevant articles that will answer their questions and then suggest those articles to help them.

12:45And you just get this beautiful feedback loop. So you don't have to have it make it up on its own. It's using your database for reference. Okay. Is there anything else you're going to show or is that it? That's the primary loop. But the thing I want to point out that's very important here is that probably everybody listening to this show is thinking about building an agent, right? You know, building an agent, either an agent is your product or an agent is inside your product, right? So for Untangle, we have Grace. She's our AI divorce agent. She's inside of Untangle. Now, she has specific tools.

13:25They're very specific to our vertical. Divorce statute search, case law search, child support guideline search. These are very specific to divorce in Connecticut. But Grace writes all of our AEO content. because she has the right tooling to write legally accurate content. And so everybody needs to take that mentality and say, oh, I'm creating a highly specialized vertical agent for whatever field you're in. Say it's floor tiling or snow removal. You're going to have an agent that has specific skills. That should be the agent that writes your content. That should be the agent that writes your emails.

14:09This is all the same entity. okay i'm gonna ask you to bring that back up on the screen with the whole flow chart so we can talk about it by the way you mentioned zapier they're they're sponsoring this uh podcast are you using zapier what are you using them for i love that that's awesome um i've just been a massive fan of wade and the team and zapier for years i mean and i think gosh zapier has been around for probably what 15 years or something i feel like honestly between you and me that that's that's kind of a challenge for them because we have 15 years of expectations from them as the company that connects any two other pieces of software and we forget that they are so ai driven i'll give you an example this morning i was thinking i would like to set up claude co-work to connect to grain and then from grain pull out my transcripts then use that to organize an email for me to tell the team about what I've been doing that might be relevant to them.

15:10Cause I forget to update people. And I'm in my own little world here in, in my home here in Austin working and I couldn't connect Claude to grain. So I said, okay, great. That's where Zapier comes in. They actually will have a connector into Claude that connects to all their other tools. And I started working with that. And then I realized the hell am I even doing? I now have to have Claude connect to Zapier, connect to grain to do this weird out thing. I went right to Zapier and I said, can you do this for me? and Zapier just pulled it together and created an email in Gmail for me. And even I, someone who's now been deep in their world, forget about their connection.

15:45Yeah, I need to do that more. And it's funny because I was in SF last week and I walked by a Zapier sign. And I thought, gosh, I need to use their automations more because I'm probably making a lot of things hard for me personally. So I'm a big fan of them and the team. And I think the way Wade thinks about things is clever. I mean, he's converting the entire company to AI first and really using the tools. So that's where that connection came in. All right. So a lead comes in. That's this top left. You start sending them drip. What's the tool? What's the software that you use to send all this out?

16:24I know what's actually manually sending it out. It's resend. But what have you built internally that's doing all this? so gosh this is the beauty of the world we live in right now so i've you i use amp to build all of this uh infrastructure right so literally all i did is i opened up amp and untangle is a next js app right so it's all on typescript um and for those of you listening and watching that don't feel like you're deeply technical don't have a computer science degree then it's okay like you You live in a world now where you have AMP and CloudCode and all these tools that are the best engineer you can imagine and they can build.

17:04You just have to talk to the agent and ask it to help you. Now, I have a computer science degree, and I've been coding for 25 years. Yes. So I have a slight advantage in that. But all I did was open up AMP and said, I want to build an email drip. Can you do that now to show us? Yeah, sure. Sure. Yeah. What are you using to access AMP? I use Ghosty primarily, which is, let me switch back to this screen. What's the advantage of using AMP? I know you tend not to talk about AMP, even though you're connected very closely to them. But what's the advantage of using them? Okay. Okay, so AMP is an amazing coding agent.

17:55So there's a lot of choices people have here. They can use, if they like a visual environment like an IDE, they can use VS Code or they can use Cursor, and they have their own agents. But in VS Code, for instance, you can use AMP if you like that. I am more of a command line interface type of guy. Like I like the cleanness of this. So what I'm going to do is start up AMP. I'm going to make this a little bigger. And so this is my coding agent, right? So what I do is I hop in here and I just start asking AMP to do things, right? So we've got, you know, usually I do actually use Whisperflow for this.

18:37So I'll say something like, have a look at our email drip system and tell me how it works from a technical perspective and then create a markdown file so I can share that. All right. So AMP's going to go off and do that. I'm actually secretly showing you a new feature. If you see deep here, this is a mode. We're actually using GPT-5.2 instead of Opus 4.5. So a little bit of inside baseball there. So what's happening is it's cranking. It's thinking about stuff, and it's going to end up giving me what I want. Here you can see it's creating a task. and it's going to complete that task and do what it needs to do.

19:25Now, what I actually do is I actually run three of these agents at a time. So I actually have three amps open at any one time, and I'm usually working on three different parts of the app at once. In three different ghosty windows or all in the same one, right? Yeah, three different ghosty windows. And the reason why is because I can just get so much done. like now I have and sometimes I feel like three is not enough that I have all this agent power and I'm actually not shipping stuff quickly enough yeah I picture you having three of them working while they're working you go on Twitter and you send a message out yep okay a little a little life hack that's kind of fun so recently I had a a x article get, I think it was 1.8 million views on X.

20:19And it was about this new Ralph loop where people are using a loop to code. But the way I actually wrote that article was I set up Ralph locally using AMP and I said, AMP, let's get this going. Help me set it up. And then at the end of that, I was like, why don't I just write an article about what we just did? And so in the same thread, I said to AMP, why don't you go ahead and write a markdown article that I can share about how we set up Ralph. Um, I took that markdown, I edited it a bit and then I, uh, I, um, went into X and I posted and it decided to go kind of hyper, huge, hyper, uh, crazy.

21:02So that's AMP. I use it every day. I love it. Obviously I work at AMP, so I'm going to say that, but I really do think it's, it's one of the best coding agents in the world. Okay. And so you're just going into it and you're saying, this is my vision. I want you to start creating it. How are you thinking about the way that it's getting it to create? How are you breaking down the tasks? How are you giving it instructions that's different from others? So I think this is the crux of this situation. This is the most important question that all of us are dealing with. We live in a world now where effectively code is free.

21:35It's at least very cheap, right? So what we build is now becoming more important than how fast we build it. And I think this is honestly creating a crisis, right? Where all of us who can build are saying, what should I build? Am I building the right things? You know, So I use Fizzy, which is 37Signals' new app for my backlog. I add things to my backlog. And then when I realize, okay, these are some features I want to build out, what I do then is I create a product requirement doc PRD. And then I take that PRD and I break it down to user stories. And then I use the Ralph loop to build that feature.

22:31and sometimes I'll do up to three at a time and then I'll come back and look and say, okay, is this the right thing? But I'm just trying to be vulnerable in the sense that everybody's dealing with this question of are we building the right things? Are we fast enough? Are we going to get outbuilt by everybody else now? What is the moat now? And it's disconcerting. Isn't are you building the right thing about are you getting customers? Yeah, it should be. But it's not. I do feel like there's less excitement around how do I get customers? Like your post about Ralph went insanely viral. And then Greg Eisenberg did an interview with you where you walked him through how you do it.

23:17The thing that I don't see much is like what you just showed me earlier, which is here's how I'm thinking about drip campaigns. Here's how I'm customizing it. um how i'm thinking about getting new customers in the door that kind of stuff doesn't seem to be getting as much attention is it because no one no one's paying attention to it or it's just not flashy enough for us to talk about uh i think because marketing is the real work and people don't like to talk about it um you know actually your go-to-market strategy and your acquisition and your funnel and your cac and your ltv you know these are actually the real business things that we all need to be thinking about.

23:56They're a little less sexy and a little less exciting. So I'll tell you this. I think the loop that I, you know, briefly mentioned is actually probably the most valuable thing I could talk about today, which is every day, create, just go to AMP, go to CloudCo, go to your favorite agent and say, I want an email every day with key, with one to two key actions I should take to acquire more customers at our desired CAC, right? And you build out a little script. It's very simple. It's a cron job. It runs at midnight. And you basically feed the right information into the model. And then you say to the model, write me an email with your thoughts on our stats and what we should do today to increase our signups, to increase our engagement, to increase our numbers.

24:50And actually that loop has led me, you know, very quickly to optimizing the business, you know, in a, in a real way versus like what features should I build? What are you giving it? What kind of access are you giving it to help make that decision? So, um, what I basically do is, uh, I think about the, a couple of things. Okay. So what's happening with our, our lead magnets? Um, how many, uh, visits did the page get? and we use GA4, which has an API to get that, right? How many visits do we get? How many visits do we get on the lead magnet? Is the lead magnet the calculator that you and I were just talking about?

25:30Yep. And we have several of them, but that's one of them. So how many visits did you get? How many leads did you get? So a simple conversion number on that, right? And then in your lead magnet, ours happens to be a questionnaire. So it's got questions. Then we also feed into it the questions and the drop-off rate between each question. That's all in the database, right? It's all coming in from GA? Mostly. So there's GA4, which we use. And then in our own database, we have things like, which question are you on in the lead magnet? And we store all that in the database. So I think the bigger lesson here is you need context.

26:16Like the agent needs context to give you valuable data. So what you need to do is go into AMP or Cloud Code and say, okay, we've got this lead magnet. What information do you need in order to help us to optimize that? Got it. And then AMP says, oh, well, I need to know what the conversion rate is from question one to question two to question three to question four. And you go, oh, okay, build it. And then it goes, great. and then it goes and puts that data into the database. And then what you do is you go back to the cron job and you say, okay, now grab that data every day. So what you're doing actually is like context gathering.

26:58And then you suddenly have a agent that has all the context it needs and basically is a VP of marketing or VP of sales for 15 cents a day. What about this? the idea for a calculator came from you, from your years of experience. Can it actually give you that kind of idea? Let me think. I'm trying to remember how I got the idea for a calculator. I think I talked to AMP about it. I basically said. Yeah, what would you say? I'd say, AMP, we need to capture leads. what should we do? And I'm pretty sure it sort of said, well, we could calculate child support, we could calculate alimony, we could create a roadmap for divorce.

27:54And I was like, okay, let's explore that. And it turns out calculating child support is extremely complex and difficult and you have to have all the right tools, but we've built those because that's what a Tangle does. Um, so I think the lesson here is go talk to your agent, um, and then iterate and then measure and then iterate and measure, um, as fast as you can. So then it's coming up with an idea like that. You're saying, yes, I think I could do that. That is where I want to spend my attention. You create the calculator. You then, you then get some amount of data from it and you say, what do you need to do in order to optimize this?

28:27It starts telling you what you should give it access to you give it access to that and then you say to it i want every day an email that tells me how things are and are you also asking what are your recommendations yep yeah i basically say what is the one or two things we must do today um because as you know like analysis paralysis is a real thing right and you can get all this data but the truth is you just need to make one decision today to increase your number. I'm thinking the next step is not just to get the email, but actually have AMP implement the suggestion. And then I wake up to a PR that I review, where it's actually done the thing that it suggested, and then I get to review that instead.

29:21What's one type of insight or what's one specific insight that it gave you that a human being wouldn't have had? I think these are all insights that a human being would have had, but I would have had to pay a VP of marketing, you know,$250 ,000 a year to have that insight. What type of insight is it getting you? Because you've been a phenomenal marketer your whole freaking life. Like, I remember, actually, I totally forgot that you ran DropSend. I remember living in LA at the time, and people were talking about how you were building in public and you were being so open and vulnerable about it that it just was shocking and everyone paid attention then they would use drop send to share files it was basically right dropbox before dropbox and i totally forgot about it until i heard you on a podcast talk about that so you know this stuff what's in what i just want to get the type of insights that you would get from it that maybe wouldn't have occurred to you with all your experience that's kind of you to say that and i've forgotten that we did uh um bare naked app which is how we built.

30:22Think about that too. I wasn't sure if my memory was good on that. First of all, it was in public at a time when nobody was doing it and you really showed people how to do it in public. And then you called them bare naked, which also was attention getting on top of attention getting. Okay. So that's why I'm asking you about this. It worked. It did work. But as you know, most of marketing running a company is not sexy at all and and all that amp is doing for me is really reminding me of of fairly basic obvious things that i need to do that i haven't done yet and that's actually very valuable right so when amp basically read this daily email and said oh you have you know a number of customers that have clicked more than three times on these email trips you should do something about it.

31:12That's not rocket science, but it's an actionable, valuable insight that I would have had to have or I would have had to pay someone to have. And you're very kind to say I'm an experienced marketer, but just because you have experience doesn't mean that you see opportunities consistently. And all you're doing is increasing your batting average, right? It's like, it's more likely that I'll see a good suggestion if I'm getting a smart suggestion once a day, right? And so I think that's what's really happening. And it's extremely, I just feel, I can't believe this is real. I think Ilia was saying this on his Dworkesh episode, like, isn't it amazing this is real?