Ep 299: LLMs and AI for "Behavior Triggers" - Conversion & Retention

I first talked about Behavior Triggers in 2019, but they were difficult or expensive to build. Now with LLMs and AI, I think they will be easier and cheaper. Let's think about some ways they could help companies convert and retain customers.

Transcript

SPEAKER_00 0:55

Welcome, low ego action heroes and phoenixes, and Analusia. Um, welcome to episode 299. Uh, we are looking back at behavior triggers. We've talked about them before many years ago. I think I started talking about behavior triggers probably eight years ago. I think they've been in one or two of my books. But admittedly, at the time it was kind of hard to build behavior triggers. Um, it wasn't easy, it wasn't inexpensive. But now that we have capabilities from AI and LLMs and machine learning and things like that, I think behavior triggers are going to be much easier to build and utilize. And I want to see your company out there building and utilizing them. So we're going to be talking about them again. Oh, I forgot to turn on my backlight. All right. I even brought out some of my old slides from yesteryear, uh, from some other era, so we can take a look at some old slides and talk about behavior triggers. So if you have any questions about them, put them live in the chat. Or um if you're watching the YouTube video later or the audio podcast, get in touch. Leave a comment, let me know what you have questions about. Don't forget to check out dcx.to. That's our page of links, and it just keeps growing. There's so many more resources and things there now, from books to articles to courses to uh programs and all kinds of things. And of course, since this is episode 299, next week will be episode 300, and I've got a big announcement uh for episode 300. So let's look at some of my old slides and talk about behavior triggers. And it looks like the top of my slide is cut off, but that's okay. I'm going to talk through it. Uh, excuse me. So, um when we think about people's behaviors on our site, we're looking for traction, adoption, conversion, uh, how much money people spend, how often they did something. That can be okay, and you can track those things. But I want to see tracking and data science move in a more uh proactive direction. And actually, I see from my slide my story, I tell a story from July 2019. And uh, that was a fresh story when I wrote this slide. So this slide is uh about to have its what seventh birthday? No, yeah. Um, so in July 2019, pre-pandemic, I tried for over an hour to buy a week, uh long weekend cruise. Um, and I tried to buy that online. But no matter what I tried, at the very last step of the checkout, I got this generic error message that just said, nothing helpful, you know, there was no help, there was no solution. I didn't know how to fix my problem. So I tried logging into my cruise account from different browsers. I tried buying the cruise again, I tried different credit cards, I couldn't pay for the cruise, I couldn't put a hold on it, I couldn't put a down payment on it, and it was just a dead end. And even the next day, I got a cart recovery email where it's like, oh, you left this in your cart, don't you want to buy it? Yeah, uh, I tried. I sure did. I spent a long time trying to buy it. Um and I started thinking when when I lived through that, I started thinking about stories usually sales or marketing might tell. About a customer like me who puts something in the cart and then seems to not buy it. We might make up stories about that person. We say, oh, customers like Debbie, she really doesn't have enough money for a cruise. Maybe she just changed her mind, or people who fit into Debbie's demographic don't always buy cruises on the first visit to the site, or oh, lots of people are abandoning their carts. They're just tire kickers and they never intended to buy a cruise. So many of us have been in meetings where somebody had some made-up story about customers and why they did or didn't do something. And it's just guesses and assumptions and imagination. But what do we really know if we haven't done excellent research and learned more about people who buy or or don't buy? But then I really got to thinking about something else. Sure, this could be a bit of a tale about research and the importance of research and not making assumptions about customers, but what if we could have saved that sale? Right? Um, could we have saved? I can't be the only person this happened to. What if we could save that sale? And I thought at the time, being 2019, when we weren't really talking about AI, I imagined an algorithm that would proactively notice customers who were struggling, getting error messages, trying to check out multiple times and not being able to, and then put some sort of process. Again, we're assuming this is on a website or maybe in an app, put some sort of process in place to help them just in time. And Donna Lucia says, Why making stuff up instead of investigating or trying it for themselves in case there are actual bugs? Yeah, agree. Uh, sometimes uh I I had a fight with a company the other day when they said, Oh, we're sorry our app's not working. We're gonna send you an APK to sideload. And I said, I'm not your unpaid QA. I don't want to sideload your APK. You've got a QA team, right? Use them. So continuing with the behavior triggers idea, um, I want to give you an example of behavior triggers. It's it's something that I made up, so if you look for it online, you're just gonna find me. But the idea is that we're gonna try to help people accomplish a task before they become sad, sad metrics or people we're making up stories about. We wanna help them. Oh, thanks for the follow. We wanna help them before we make uh more incorrect guesses and assumptions about their behavior. So here's some sample behavior triggers that we could use based on the cruise ship story I was just telling you. So, what if the second purchase failure and the second time I got an error message triggered a live chat from someone who could see through their support interface what I was working on, honor the price that I was seeing, and help convert me into a paying customer. Doesn't have to be creepy. Um, and the fact that I was trying to check out looks like I'm a serious customer and I'm trying to buy this. I'm not kicking tires. I tried to pay. And the second checkout attempt is generous of me because I could have given up by then. So save the sale. When things go bad for me a second time, step in while we have momentum. Um, once I leave, I might buy from a competitor and you've double lost. So, you know, I imagined this as like a little thing that could pop up in the bottom right, like where live chat usually is, and it could say, Hey, looks like you're having some trouble checking out. Um, would you like some help? And then my idea was you would be connected to a live human. Now, nowadays someone might say, Okay, connect them to an AI chatbot first and only give them a human if they really need it. But uh, you know, that's a that's a fight we can always have. This would need to be tested to see what kinds of problems people are having. Can the AI chatbot solve them? If so, maybe that's a a good first line. If the AI chatbot generally can't solve these problems, then just get people to a human. Um, and what about the idea that I was logging in from multiple browsers? The system could catch that also. It could say, wait a minute, this person keeps putting the same thing in their cart, they keep showing up from different browsers, they're having trouble. We could pop up that little chat and say, hi, would you like some help? And of course they can say no, they can decline it. We'll only pop it up once and then we'll get out of their way. Um, but that could be another behavior that could trigger this proactive chat to try to save the sale. What about the system noticing that I attempted multiple channels? Part of the story was that I didn't, which I forgot to tell you, was that I didn't just log into the website and try to buy from there. I actually also called the company and I said, can you sell me this cruise? I can't seem to buy it online. And they said, We see this cruise in our system, but not for the price that you're getting. We see it for a significantly higher price, and we can't give you the price you're seeing on the website. But it sure looks like I'm trying to buy this cruise. I'm a serious customer, I'm not a tire kicker. If I'm if I'm seeing some sort of weird price that half expired in part of your system, just do you want to make the sale? Do you want to have zero dollars from me or yesterday's price? Now that the price went up, you decide. But that could be a save that you could uh a sale that you could save, especially if you notice I am seriously trying to buy this. Uh, what else could we look for? So thinking outside of the cruise situation, what about um uh a behavior trigger that maybe doesn't lead to a chat pop-up? Now imagine separate from the cruise story, somebody logs into their account in your company's system and they say, I want a full data export. I want to download all of my data. Should that trigger some sort of action? Sometimes we don't ask people if they're happy with us or anything like that until they're already canceling or deleting their account. Could we save this earlier? Sure, it's possible that someone's just downloading their data just for a backup, but I would say fairly often people are downloading their data because they intend to leave. And so this could be an opportunity that could trigger a customer support person, a sales representative, someone to maybe send this person an email because maybe the chat pop-up is a little too creepy and we don't want people to feel like they're being watched. We want it to be helpful but not creepy, and I think it's gonna take some experimenting to find that line. But it would be interesting to um to write an email to somebody and say, hey, we notice you downloaded your data. Um, we're curious if we can help with anything, or maybe you just happen to send them an NPS survey and you don't mention the data download. Hi, you know, anything we can help you with? Here's our NPS survey. How can we help you? So that could be another example of a um uh behavior trigger. Another one, again, outside of the cruise story, could be what if somebody had auto renewal turned on for a subscription or something like that, and they log in or they're in their account anyway, and they go and they turn off auto renewal. That is a bit of a sign that they might not want to stay. And sometimes we don't communicate with that person until the renewed date has passed and they didn't renew, and now they're gone, and we're sending them. Oh, how could we have improved emails? Too late, too, too late, sometimes months too late. So that could be an opportunity to send an NPS survey and get some feedback, or send a salesperson, have a salesperson, no, don't send a person, have a salesperson contact uh someone and reach out, just find a healthy way to reach out. And I'll give you another example of this. I actually saw this in action a few months ago when I had tried Claude for PowerPoint and it was a disaster. I gave it one or two chances and it was terrible, and I didn't uninstall it, but I stopped using it. And I think they had some sort of behavior trigger of their own type set up because about, I guess, a month or two later, I got a survey saying, we'd like to know why you stopped using Claude for PowerPoint. So they obviously knew I had used it, and then obviously knew I had not used it. And that's a great time to get feedback. Maybe you're not necessarily saving a sale, maybe you're just learning what went wrong for somebody, but it's your opportunity to be more proactive. And now that we have AI and LLMs and other tools that can be watching for these behaviors and watching for these triggers, I think these are probably much less expensive and easier to build and operate than when I first came up with this in mid-2019. Uh, couple more slides. So again, someone might say, ah, Deb, you're asking us to build a whole thing just to see if we could save a cruise or save a subscription. I don't know if that's worth it. Oh, maybe these people are just edge cases. And I say do the math. Let's take the cruise example. Let's imagine that um the let's imagine five people per day. That's a very small number, and I'm one of them. Let's imagine five people per day had the problem that I had that day buying the cruise, getting that error message. Um, based on what the cruise cost and the idea that a lot of times cruise companies make a noticeable amount of money upselling people stuff when they're on the boat. So there's that initial purchase and then additional spending later. Hypothetically, if that were a 3,225 cruise plus later spending, restaurants, souvenirs, excursions, drinks, whatever, and it's five people per day, that could be six million dollars per year. If behavior triggers cost you half a million dollars, would it be worth it? Now that's my slide from a long time ago from 2019. This might be a less expensive project now. Maybe it's a $100,000 project. Would this be worth it to save $6 million per year and reducing the negative word of mouth? Because while I'm not naming the cruise company this time, I could have said, oh, this terrible website from this particular cruise company, don't bother with them. And then I'm spreading negative word of mouth. Um, or it could be the idea of law losing future sales. If I'd gone on that cruise and I felt like, wow, what a great company, what a great boat. I want to go on another vacation someday with these people again. That's now lost as well. So there are definitely these kind of upstream and downstream effects that sometimes we can't calculate, or maybe we can. Maybe the cruise company knows exactly what percentage of people tend to take a future cruise with them and what is the lifetime value of a customer. So you can know. Sorry, you can know if fixing something for, and I'm just making up this number, five people per day is worth it. Because chances are it is. What what at first might seem like a small loss. Ah, we didn't sell five cruises today. Well, over the course of a year can really ripple out to money that is worth saving, and customers we want to win. Um, this was another example that I had, and again, this was years ago. Um, I had been spending a lot of money on my American Express card because I was doing a lot of business travel. I used to speak at 50 or more conferences a year, and sometimes I had to pay for it myself, and sometimes the conference reimbursed me. But I had a couple of years where I was spending $75,000 a year on my American Express card. Needless to say, no, that is no longer the case. Um, but I had so many problems with American Express not backing me when there were uh fraudulent charges or people who promised something and delivered something different, and they never backed me. So finally I said, Well, that's it. I I don't uh I think I'm gonna cancel. And you might think, well, $75,000, that's not very much, Debbie. They probably care about people spending millions. I'm sure they do, but I also know that they cared about people spending $75,000 because in the app there was a little meter that calculated my spending, specifically up to $75,000, because at that point they were going to give me extra perks for being a big spender of some sort. So this must be some sort of number that American Express does care about. And when I wrote this slide a couple of years ago, um, I had some numbers. I found that in 2023 there were over 141 million American Express cards worldwide, and the average customer only spends $24,000 a year. So someone spending $75,000 a year is three times that normal, uh, that average customer spend. Um, and so and American Express is also making their money from those merchant fees. Um, so this means that maybe someone spending $75,000 a year is worth a minimum of $3,000 in merchant fees and interest and annual fees and other stuff. So we can play with some math. So you might say, ah, so what? Someone canceled their American Express card, they were only spending this much, why should we care? It's just an isolated incident. Sure, but at what point should your company care about isolated incidents? When do you want to know that something is a trend before it's a trend? So we should care about these things, especially if it if it uh is becoming a trend or we are seeing people spend less or canceling cards or downgrading cards, this should matter. Um, moving more through this slide. So what if 0.001% of American Express customers downgraded or canceled? That's 1,412 customers, and that would be $4.2 million lost in a year, plus possibly more if people were paying over time and American Express doesn't get that interest. So that could be money worth saving, and behavior triggers could catch patterns. How early can we catch customers who start to reduce their spending? Can we notice behaviors that tend to lead to someone downgrading a card or canceling a card? Maybe AI can watch behaviors and start to train a model that understands ah, when people stop doing this or they do less of this, they tend to cancel within X months. I would imagine we can model that. And it's a model that we'll have to always update and improve, but we can probably know. And so, what do we want to do about this? Is it worth getting this customer back? How early can we catch this pattern? When Debbie starts spending less but hasn't canceled her card yet or downgraded her card yet? Can we catch her early and find out what's wrong and see if we can make a difference for her? Is it worth it? You know, hard for me to say, but something, a conversation that we could definitely be having inside of our teams. And again, when will American Express figure out that I'm an unhappy customer if I Nothing. And once you've lost me, you might not be able to get me back. What would they have to offer me to get me to keep spending that much? Well, for me, nothing. Too late. My problem was around their service. And now I don't trust them. So I'm spending now $500 a year on American Express, and I moved all of my spending to another card. My change in spending started in July 2024. But then I didn't downgrade my card until January 2025. So that means they would have had at least six months of significantly reduced activity for that to trigger something in their system to say, whoa, this person has significantly changed their spending, reduced their spending. And what if they had reached out in a non-creepy way? They just happened to send me a survey. They just happened to send me an NPS, something like that. Um, and just to learn more about what's going on for me and if I have feedback for them, and I probably would have told them, yeah, you're uh you haven't backed me on uh any of the charges I've fought in the last two years. I've lost almost $10,000 on charges you wouldn't protect me from, so I'm done. So looking at the slide, behavior triggers would have shown that my spending patterns started changing since they started changing in June. You probably would have noticed by July or August. Wait a minute, this person usually spends $6,000 a month and they're spending $50. Something changed here. We should probably get to the bottom of this. But once I decide to cancel, I've made up my mind. There and they tried to offer me ridiculous stuff to stay. It made no sense. They said, if you spend $3,000 in the next month with us, we'll give you. I said, I'm calling to cancel. I'm not calling to spend more with you. What is this? It's almost like you're not listening to me at all. You don't even understand what my problem is, and you're just trying to get me to spend more. I don't want to spend more. I don't trust you. You're not going to back me if I have to fight a charge. I've learned that. Uh Anna Lucia says this would be kind of like data analytics and UX initiative, right? Yeah, I think this could definitely be a cross-functional adventure. This could be data analytics, uh, whoever is working with your AI and LLMs and agents. It should be UX research, UX, maybe UX design. They have good ideas, uh, product management. It could even be people from sales and marketing, it could be people from customer support. Everybody has a reason to want people to cancel and downgrade less. We want them to contact customer support less. We want them to cancel less. We want them to be unhappy less. All of the people, the roles that I just named, are have a horse in the race. They want to see happier customers spending more. They don't want to see unhappy customers talking about it on YouTube, saying I don't use my American Express card anymore because they didn't protect me. Their promise that that I would be protected in case of crappy charges wasn't true. Um, going back to my slides. So, do we have any data that would hint at a problem? Oh, Anna Lucia says, I'm seeing this being implemented in a really dynamic way with data analytics, something to analyze behavior changes and help them do their work. Yeah, definitely. But again, I'd love to see it as a cross-functional initiative where, of course, we have data analytics, but we also have whoever at your company is behind AI and LLMs, could be data scientists or somebody else, engineers, product management, UX research, UX designers, and I would even imagine marketing and sales involved as well. So, can a behavior trigger correlate data? We don't know if a correlation indicates a cause, but we can notice a steep drop in my charges. We can notice that I have placed a number of disputes against charges, and that I've been denied every time, and that this is now a lot of money. I've now been, I've now fought about $10,000 of charges over three or four or five charges, and you've given me no money back. You didn't accept a single one of them. They were all denied. That could trigger a pattern as well. Why are we denying this person's charges? Could though could that be the could those be the wrong decision? Are are we we not helping that person? Are they thinking of leaving us? Um, so you know, every time I had a problem with my purchase and I was baited and switched, or I bought something I never got and I filed a dispute, American Express didn't give me my money back. And yet their promise is that they will. So that should be a problem. That should be something somebody cares about. I'm probably not alone. So it's uh a possible pattern, and it's uh we might not know the root causes here, but there is some sort of pattern going on where I'm not getting the uh experience that I need to get from American Express. So there you go. You've got a couple of examples there of behavior triggers. I hope everybody will start thinking about how they can use this on their team, at their company, for their startup. Um, can you create an AI, an LLM, machine learning uh algorithm? I don't care what you call it. Can you create something that is keeping an eye for certain triggers and behaviors or data correlations or behavioral changes that would show up in data? And find a non-cree way to reach out to people. That's uh that's what I think. So uh that's behavior triggers and um weird. Okay, I'm getting a weird email. Disregard that. Um, it's not from any of you. Uh, if anyone has any other questions or comments about behavior triggers, or you can think of an example of how you wish they were used because of your experiences or how your company might use them, uh, throw them in the chat. Let's hear what you're thinking. If you're watching this video later, drop a chat uh in YouTube. And um, YouTube has stopped notifying me of these things. I'll have to figure out why. But uh when I when I know you've left a comment, I will definitely respond to it. So let's give people a couple of minutes to see if they have uh ideas or examples or questions while I take a drink. So while we wait to see if anybody has any questions, a couple of show notes. Um, an hour from now, I will be streaming on my vocal coach Debbie channels on Twitch and YouTube. Um we're going to be talking tonight about the importance of performance when you are singing and you want to tell a story with your song. I'm gonna be doing a series of these uh analyzing how other singers approach their performance and storytelling. And tonight we'll be looking at a Peter Gabriel song called Intruder. Um, so that'll be an hour from now. Um, tomorrow, of course, we will do Ask Me Anything at 7 p.m. Italy time, 1 p.m. Boston time. Um, and we should have Dr. Nick with us, so that will be delightful. I just saw him in person. Uh next Tuesday, June 16th, 2026, I've got a big announcement to make, and that will be our Take Action Tuesday show. Um, and of course, uh the usual streams continue. So I hope people will consider following me on the Vocal Coach Debbie channel where I'll be talking about um is it a is it a good thing? Is what a good thing? I don't know what you're referring to. Is what a good thing. Um, yeah, anyway, uh don't forget to check dcx.to. I've got lots of uh new programs and things. Oh yes, it's a good announcement. Yes. Um it's a new program that I'm uh offering, and um I've done my best to make it affordable, but as always, if it's unaffordable, let me know, and of course I will uh do my best to include everybody. Um, so yes, this announcement will be uh I have some coming announcements of some new things that I'm offering, and I want to make sure people know about them because uh I think I think people are going to want to use them. So I want to do some formal announcements. Um uh okay, so uh I think that's mostly that. I'm not seeing any questions. I'm checking LinkedIn. Nope. Uh okay, so we will wrap up for today. I'll head over to my vocal booth and start setting up for the Vocal Coach Debbie stream. That will also be archived on YouTube on the Vocal Coach Debbie channel. So thanks again to everybody. Keep thinking about behavior triggers and how you can use them, and I will catch you all tomorrow for Ask Me Anything.