SPEAKERS

Tejas, Steven, Intro, Spencer

 

Intro  00:00

I’m Steven and I’m Spencer. Welcome to Ragnarokast, your podcast for all things marketing and Martech. Hello, everyone. We’re the CO CEOs of Ragnarok.

 

Spencer  00:11

Welcome. Welcome. All right, here we are. Tejas, great to have you.

 

Tejas  00:16

Great to great to be on.

 

 

Spencer  00:18

Little disclaimer that this is not like other podcasts that you’ve been on. Yes, we will be talking about marketing stuff. We will be talking about AI will be, you know, generally, doing podcasting things. But you’ve met Steven in person, so you know, he’s, he’s a, he’s a silly guy,

 

Tejas  00:32

I met both of you also, yeah, I’m prepared for this.

 

Spencer  00:36
It’s gonna be a little silly, but we think, still impactful. Well, also, welcome Steven, I guess a

 

Steven  00:44

Well, thank you for not welcoming me back. You know this, 

 

Spencer  00:46

Welcome back.

 

Steven

I’ll take it.

 

Spencer

Let’s go Birds. We got, uh, we’re, we’re a few days out here from the Super Bowl. Steven’s representing his current City’s team, the Eagles, which are, you know, hopefully they’ll win the Chiefs. Come on. We gotta, you know, like, it’s not that I don’t like the Chiefs, it’s just they won last year. Let’s, let’s give the Eagles a shot. You know what? I mean? 

 

Steven

Yeah

 

Tejas  01:10

I won’t pretend to be following NFL this year.

 

Spencer  01:16

That’s fine. That’s fine. It’s just like, it’s more like Travis Kelsey, you know, he’s got Taylor Swift, and it’s just Eagles got to get something, you know? And Philly, if you’ve ever been to Philly, is a sports town, actually, no, it’s an Eagles town. It’s an like, they have the Phillies and everything else to sort of tied them over. But really, it’s a it’s a football town. It’s an Eagles town

 

Tejas  01:41

Philly, cheese steak is, is not relevant to the culture and stuff, right? It’s just football.

 

Spencer  01:45

Yeah. I mean, I don’t know, like, how often the average Philadelphian actually eats a cheese steak, Steven? Do you know

 

Steven  01:53

They all come in from Jersey, so it’s hard to say.

 

Spencer  01:57

It’s kind of like, I mean, I’m sure they know the good spots, but I think it’s still, like, kind of a touristy thing, because it’s a lot to eat too.

 

Tejas  02:05

I can relate to that. I’m from Nashville, and, you know, everyone asks me, now that I live in San Francisco, have you had Nashville hot chicken? She eat it a ton. It’s like, actually, I don’t even remember hearing about Nashville hot chicken or Hattie B’s when I when I grew up. I think all that stuff. So he went extremely thirsty.

 

Spencer  02:22

It was just chicken. It was a wasn’t hot chicken, and it was just chicken. 

 

Tejas 

Exactly, 

 

Spencer

All right, so Tejas, do you mind giving us a little bit of a background on yourself and high touch as the real intro to this conversation?

 

Tejas  02:36

Yeah, sounds good. So I’m Tejas founder and CO CEO at HighTouch. My background is in engineering and product and technical stuff. Grew up programming and got into like coding and stuff, pretty young, and came out to San Francisco about 10 years ago and joined a company called Segment, which was a super tiny company that not a lot of people knew about back then, but is now one of a big, big players in the marketing technology space, CDP, customer data platform and owned by Twilio. I started Hightouch with a couple of friends, Kashish and Josh, about five years ago, and we’ve been growing super strong work with some big brands like PetSmart, Warner Music, Grammarly, whoop so forth. And the high level idea is that we want to help brands make all their customer data as valuable as possible, to improve marketing personalization, make it more effective, and help them use AI in that process, in a in a real way, right? So, no. Bs, all real ways, all easy to implement, all easy to try out. And yeah, we’ll talk more about it, but it’s a little bit of intro on myself now. 

 

Steven  03:47

Now wait a minute, you weren’t the guy who was telling my friends to tell me that they shouldn’t I shouldn’t go buying CDPs, you weren’t that guy. Were you?

 

Tejas  03:56

I was the guy who was, you know, telling everyone to tell their friends to you to not buy a CDP, both in person and on T-shirts at conferences that somehow our marketing team let us print, which I think is, is pretty cool. Usually, the marketing team is supposed to tell people not to do crazy stuff, right? But our marketing team is like, pretty I have to tell them not to do crazy stuff sometimes and in person and at conferences and online and blogs. Yeah, so we published a pretty hot take. I don’t remember how long ago it was, maybe like, three years ago, called Friends don’t let friends buy a CDP. On our blog, it’s still up. If you Google friends don’t let friends buy a CDP, or probably just like, don’t buy a CDP, or a why not have CDP, or whatever, you’ll probably find it online. And really it was putting out this thesis that CDP as a concept is really valuable. Like, yes, marketing teams should have a database with all their customer data, and they should have tools to build audiences or journeys or segments on top of it and get it into different. Marketing and ad platforms like that is a very valuable concept. They shouldn’t be bottlenecked by data teams or IT teams, or not knowing SQL, not being an engineer for every task, but that the whole idea of collecting all your data or copying it from different places in your company and restructuring it to fit it into a traditional CDP, like Segment or Salesforce data cloud or Adobe, RTCDP, or whatever it is there’s, there’s so many, honestly, at this point, is a bit cumbersome, and for a lot of companies, they’re already investing in these data warehouses, like a Snowflake or Databricks, Google Cloud, or whatever it is for analytics. And there should be a new model of CDPs. We call it the composable CDP. That’s kind of our original product here at high touch, sits directly on top of the warehouse and just enables those same capabilities and doesn’t require a perfect warehouse. We can help you. Help you got there too, but there’s no need to go create another sort of source of truth. So yeah, that was our original product at Hightouch. Hundreds and hundreds and hundreds of companies using it today, and now we’re thinking what’s next, and jump in on the AI train here, but try to build some real value for customers.

 

Steven  06:09

I love that. Do you know how many people I talk to every week? Or half of them say we’re thinking about a compostable CDP, not a composable but a compostable. One gotta throw the lettuce on there. 

 

Spencer  06:22

You know, you gotta be environmentally friendly these days. You know.

 

Tejas  06:26

You have to be. I mean, it’s, it’s important, you know, if we were a public company, we definitely have to rename it to compostable CDP. I mean, think,

 

Spencer  06:34

Honestly, like, I think if you’re talking about doing crazy stuff with your marketing team, definitely steal that and do something with it. You know, on Earth day maybe.

 

Tejas  06:44

You know, yeah, maybe April Fool’s Day. We’ve never done a good April Fool’s joke. There’s been lots of good ideas that I’ve had to shut down. I think they’re a little bit risky, but we’ve never done a good one from a company, brand perspective. It would be a good one. You know, iOS frequently corrects composable CDP for me to compostable. You actually, I fixed it when I type, but when I say it aloud, it’s still like, What the heck is this like? Let’s put compostable CDP. And if I do, it’s speech to text, and I just embrace it. So, yeah, I don’t correct it anymore. I just go with

 

Steven

Love it

 

Spencer  07:20

You’ll be a good April Fool’s joke is, you know, be like, oh, like, throw out a random large tech company that no one’s ever heard of, saying, we’re, we’re being acquired by Twilio or whatever. Because, like, you see, there is, like, three CDPs in recently that were bought by random large companies that I was like, I think I’ve heard of that. I don’t know what it is.

 

Tejas  07:44

So very relatable. And I think it’s more we’re going to be counting. What are the CDPs that haven’t been acquired soon versus the ones that recently got acquired? Is, uh, is my take. 

 

Steven  07:54

It’s like, how? Like the Onion’s parent company is Global Tetrahedron, which I just think is awesome.

 

Tejas  07:59

I’ll act like I’m not smart enough to understand what you just said. 

 

Steven  08:03

The newspaper is Global Tetrahedron, which is just awesome. 

 

Tejas  08:12

I think I’ve seen that. I haven’t read an actual article from the onion for a while. I do see it. See them in my feed, here and there, and just read the headlines.

 

Steven  08:21

Snap the article right there. 

 

Spencer  08:23

Before we get into our main topics. Here something I did a few episodes ago Tejas, I feel like you’re a good guinea pig for this, which is. all right, so we got a challenge for you. No, we’ve got someone outside your door. They have a clown suit, and you’re gonna have to run around the block. We’re gonna time you. No… 

 

Tejas  08:44

Luckily, I’m traveling for work, so I think I’m untraceable right now.

 

Spencer  08:49

That’s what you thought we so I want to ask you, do you have any hot takes? More tech hot takes, something that you know, it doesn’t necessarily have to be like, antagonistic or contrarian just to be that, but if there’s something that you’re like, I don’t know if I think the market doesn’t agree with me on this, but I feel very strong in my conviction.

 

Tejas  09:16

Yeah, I mean, composable was definitely a hot take and that like marketers were going to embrace the data warehouse and actually just use tools built on top of it. So super hot take a few years ago, a lot of people ask me, like, Hey, why aren’t you guys talking about composable all the time anymore? It’s like, I think it’s not a hot take anymore. It’s just about what product does it well. New hot takes. I mean, I would say AI agents is a hot topic right now, right? Everyone’s everyone’s talking about it. Everyone you know Salesforce is helping out with that, right? They’re talking about agent force, left, right and center. Not a lot of people know what it means. I think the concept is pretty simple. It’s like, how do we use AI to not just give us answers on the screen, whether it’s a predictive model or like asking chat GPT a question, how do we use it to actually do work that people would otherwise have to do manually and in some tasks just at the same scale as people? So that’s like customer support and stuff. But I think in marketing, it’s like, how do we do that a much greater scale than people, and give give marketers a lot more leverage, which is super interesting. 

My hot take on it would probably be that in the domain of marketing, I think the most popular AI agents will actually use traditional machine learning and data science and reinforcement learning more than they’ll actually use this generative AI wave we’re seeing with large language models and tools like chatgpt. Obviously, that stuff’s super valuable, but I actually think the first generation, in first class, of AI agents that will really take off and do things like marketing, orchestration, journey, building stuff like that, will actually use reinforcement learning and more data driven Machine Learning Technologies versus, you know, operating at the language level.

 

Steven  10:58

This all sounds very familiar to Tajes. It sounds like you’re putting your money where your money where your mouth is on that hot take.

 

Tejas  11:05

Yeah, it is a product of ours that we recently put out called AI Decisioning. It’s kind of built on that hot take, I may say. But the idea is that, you know, in the future and in the present, I think for a lot of campaigns, it doesn’t really make sense for marketers to be having to not justify the strategy for the campaigns. I think they need to do that. That’s great. They need to work and define the content and creative. But a lot of time a lot of people on the marketing team is spending that can really change the efficacy of the campaigns is building out these like very specific audience segments of who to send this content to you. Or building out a calendar of you know how you should spread out your content, which is very, very common in retail companies over the next weeks, months, quarters. Or building out journeys and a journey builder to say exactly how many times you should follow up with content a versus content B. And what we’re finding is that a lot of those tasks can actually be automated with reinforcement learning, where you tell an AI agent to give it some goals, basically like, hey, I want to drive for my batch and blast marketing calendar. I just want to drive clicks and I want to drive conversions, and just keep optimizing to figure out what does that best for a use case like trying to help healthy customers cross-sell into a new product category. You can, you know, specify, go around cross-sells and weight it on which cross-sells are most important. And basically, give the AI some boundaries, like, don’t email customers more than x, factor in on subscribe rate, and basically have it, you know, do automatic experimentation and figure out what types of content, follow-ups, frequency timing, all those dimensions that marketers have to think about themselves right now and do guesswork or do an AB test to understand what works better. Do that all for you and get it smarter and smarter at optimizing your company’s marketing over time. 

 

So we’ve packaged up some of the technology that powers things like Tiktok speed, where, you know, flip through and you get a really personalized feed for you, and it’s just getting smarter and smarter to personalize it for you in a kind of scary way. We personalize some of that technology and package it up in an ICU i and platform for marketing teams that sits directly on their data, wherever it is, like a Snowflake, etc, and, or CDP and, and also plug this directly into their marketing channel tools, and just allows them to really build these one to one journeys with AI. 

 

Spencer  13:33

When you’re sitting in front of your your toughest customers, like the CFO of a large brand, they say, Wow, this all sounds great, but how does this like? You know, we have insert CDP name already, or we have this in house thing, or whatever, or maybe they don’t, but just, how does this make me money? How does this save me money? Well, you know, what are your when CFOs are throwing these, like, very revenue and efficiency-oriented questions at you, like, what are like the biggest value adds from a from $1 perspective,

 

Tejas  14:04

Yeah, actually, there’s kind of another hot take embedded in this one. But I think those CFOs, and even though there’s a lot of discussion about AI right now, like, if you want to get press on AI, you probably need to talk about how it’s helping you save money or cut jobs or replace human work. I think AI Decisioning is interesting, because you know that obviously we do save marketing teams from having to do some of this tactical, monotonous work, but it’s not a majority of the time marketing teams are spending because you just can’t do this stuff efficiently as people. So we end up doing it inefficiently by building these one off journeys and not rigorously experimenting them until they’re perfect, or building these calendars and just calling that the calendar for the month or calendar for the quarter. So really, the goal of AI decisioning isn’t to save work. It’s a part of it. It allows you to be more strategic, but it’s to make more money. It’s to to. So it’s under the notion that you actually have really good content as a brand most likely, but you’re not using it super effectively, like you’ve created all this content over the last 10 years, and like you just need to send it to the right customer at the right time, and actually factor in everything you know about a customer and how they responded to content in the past to give each of them a really effective, individualized journey. And what I like about it, too, is that we can actually prove the efficacy of the of the technology. So some of the one of the things that we’ve baked into the product is the ability to pick a part of your marketing, whether it’s a cross sell initiative that you have at your company, so trying to get customers into the store more, to use a new product on your Multi Product mobile app. Or you can pick another initiative, like converting your leads database into signing up or win backs, whatever it is. And you can actually do a hold out test or an AB test. So we can just split, split the audience, take, take 10% of it, for example, don’t send any marketing to them, or still send your traditional marketing program to them and and then compare, after some time, what’s actually the performance of AI decisioning versus the traditional marketing. Now this, this can be, this is a good way to get started and see like, is this driving real value? What’s the best ways to use this technology? But I think the future is probably not running these AB tests for every single marketing program, because it’s just tedious to set it up twice, one via AI and one manually.

 

Steven  16:29

Yeah, you, you know, I think one of the big pieces of this with anything with with agentic, or anything like that, that’s using more, we’ll say, statistic, statistical level of or have has more statistical rigor than maybe. Machines are a little bit more bound to statistical rigor than humans are, whereas humans have a little bit more intuition, right and right? There’s obviously some limitations to this. Like, for example, if I wanted to launch something new, is agentic AI, like the path I would go down to launch something new, or is it more like, you know, as a marketer, I should be focusing on the new things and then, you know, spend a portion of my time letting the like, setting up the agent, so to speak, to work on the stuff I’ve already launched. Like, like, at what point where do you feel like the the agent will be the most helpful? Like, should it be, you know, like, I think back to like, you know, 10, five or 10 years ago, when everybody bought Mixpanel to look at, you know, a to look at the people who, like, weren’t going down the conversion funnel, and then, like, digging into literally everything that they did. Like, that was, like, the marketers exploration path. And it kind of caused this? Like, well, not cause. It was a good thing, but it like, had people think more intuitively about what is the real customer journey, and not like the one that I predict people are doing because they click off my email, and then they go do this thing that I told them to do, and then they do this, and they convert, right? It’s actually a little it’s much messier in between the different ends. And so do you think the, I don’t know why that’s such a tangent, but do you think the the the agent is going to be better at, like, the Mixpanel side of the house, which is, like, here’s an interesting thing I found that we could, we could go and build a program off of, or is it going to be a lot better at? You know, here’s something you’ve already established. Let’s, you know, optimize that and squeeze another 2-3-4, 5% incrementality out of it.

 

Tejas  18:20

Yeah, it’s a good question. I think at a high level, what we’re talking about here is, what are the right use cases to apply a technology like aI decisioning in a marketing program, and where will we see the most efficacy? So this is something I’ve actually been thinking about a lot. I think there’s basically three kind of paradigms for marketing, orchestration, lot of different terms, a lot of different tools, but I think it all buckets in these three things. One is just like batch, batch and blast communications, which, which, for what it’s worth, all right, bang. I’ll talk about that in a second. But batch and boss communications, which is basically, you know, I’m going to send out a calendar of communications to my customers, I’m going to be proactively paying them, even though there’s not a lot of intent or anything like that. But I just want to stay top of mind, and that’s important, especially for like a retail company that may not have the frequency of interaction with their customer without something like that. Second is like triggered communications. So these can be built in audiences, like an abandoned cart audience, right? Customers who added something to the cart didn’t check it out, can also be built through like journeys oftentimes. So like, let’s not just ping them once when they didn’t check it out. Let’s ping them multiple times on multiple different channels and kind of follow them until we really drive a conversion. But they all start with a trigger. So customers who added something to the cart, customers who finished a purchase, etc, these are actually not too bad oftentimes, because while you have to guess the steps of like, you know what, you are emailing the customer or texting them like it all stems from some level than 10, so it’s likely going to be much more effective. Doing on a on a percentage basis than something like the passion blast communications, and I think a lot of brands aren’t yet capitalizing on those. And then lastly is this new paradigm that we’ve introduced with AI decisioning, where you’re actually not building out a triggered sequence and not building out a marketing calendar, but deferring to AI to figure out what’s the best way to orchestrate this content for my business goals, with batch and blast communications. I actually think there’s a lot of opportunity for AI decisioning there, and a lot of value to be created, not for every campaign, right? You’re always going to have your big brand push, which you’re not trying to optimize conversions on, frankly speaking.

 

Steven  20:39

This should be like your new arrivals, that type of stuff.

 

Tejas  20:43

Yeah new season, you know, huge new store in New York that you want to tell everyone nearby about, like, there’s just communications where you’re not necessarily going to, yeah, you’re using as a brand channel, using as, like, almost advertising and media, even though it’s on email or text. We’ll have those done manually. We’ll put them on the calendar, keep them as is, but what do I send to like every other day? What? What I see right now in the status quo is that we create a lot of new campaigns to kind of stay fresh and stay top of mind, push those out to customers, but we’re not using like everything we have about, you know, what’s worked in the past for different customers we don’t even know, because we’re not running experiments. Right? As marketers all the time on these batch and blast campaigns. It’s super tedious to do, and we’re not using all that info of what’s worked, what might resonate with different customers to decide the calendar, right? We’re just kind of building a calendar a lot of times, while AI agents aren’t going to make batch and blast you know, marketing have a 2% conversion rate instead 0.5% conversion rate or conversion rate or something like that, 

 

Steven

0.3 I think 

 

Tejas 

0.3, yeah, probably they can increase the conversion rate significantly, right? And if the volume is high enough, it can make a real difference, not by making it like amazingly personalized in this one-to-one way, and predicting rights on the mind, right? What’s on the mind of your customers? It’s not possible for a lot of brands, but just making it incrementally better in a lot of small ways that humans wouldn’t do otherwise. CTA changes for different customers, tone changes for different customers, automatically, incorporating things like product recommendation type insights into the mix automatically. 

 

Steven  22:17

This is getting into a little bit of generative that when you talk about tone changes and like things like that. Or is that more 

 

Tejas  22:23

We’re not doing the general Yeah, asset management. If you put in a bunch of content of different tones, we can, you know, we actually do use Gen AI to scrape that out and generate more features on that. To say, content of this kind of tone works well for customers. If that makes

 

Spencer  22:37

It’s pre written, in this case, like, if you have seven different segments, you would have seven different versions of that tone or versions of that copy.

 

Tejas  22:47

Exactly except you don’t need to think about the segments. You can just say subject line here is several different ones, and it’s using go figure out which ones are not good at all, like they’re just not good ones, which ones are good for certain customers, etc, and we’ll just continuously optimize so it just allows marketers to throw a ton of content the system. And then AI decisioning is actually running continuous experiments on your batch and blast program, which is something not a lot of brands do, right? People have global holdouts. They understand the program’s working at large, but what specifically worked last quarter versus this quarter? I don’t know. So constantly running experimentation and then versus traditional experimentation, we’re not just picking one winner. We’re kind of seeing if there’s patterns so that there can be a winner for certain customers, a winner for other customers, honestly, on the batch and blast program, what we found is that there can just be a lot of long tail optimizations that add up, that don’t make your batch and blast program like crazy, amazing and right, what’s relevant for the customer not possible in a lot of contexts, but makes significant difference and can drive significant amounts of revenue. We’ve shown that with, you know, pretty large brands, like a fortune 500 kind of specialty retailer.

 

Spencer  23:55

So we’ve talked about, you know, the CFO, who is the, if if not, the decision maker on what tools to bring in is the decision maker on if the contract is signed or not, generally. And we’ve talked about how these features will it’s almost like adding a new layer, like I saw your article the other day too, but it’s like there’s AB testing versus AI decisioning, and it’s like a it’s almost like another, it’s another way of doing it, rather than it’s an alternative method, basically, for our marketers out there, who are the ones that are really, ultimately, really interested in Hightouch, and the ones that are going to be saying, hey, come present to my leadership. How is AI decisioning? Specifically, how is it going to change their day to day over the next few years?

 

Tejas  24:39

Yeah. So the way I think about it is that one, it makes your programs more effective. I think you need that so you can pitch your CFO, not just does it make them more effective? If a CFO asks a marketing team today, like, what out of the campaigns we sent last quarter really worked? Like, what changes have you made that are really incremental? Those are like, hard questions to answer today. So with using AI decisioning to orchestrate your campaigns, you just start answering all those questions by default. You have a ton of insights that can help you build better campaigns in the future and build better content, but also helps you present internally on like, what’s actually working, what’s effective, and just look like a way smarter marketer. Just be a way smarter marketer. Second, I would say, changes the workflow more materially, is today at a lot of enterprises, what we see is that there’s a lot of people that need to coordinate together to get every campaign out there, and a lot of CRM marketers that I speak to on day to day basis, almost feel like they’re just rushing to meet deadlines on this marketing calendar, especially for like, batch and blast programs, where just the expectation is really high to get new content out there all the time. And really what AI decisioning allows them to do is take a little bit of a step back, be more strategic. Reuse a lot of the content they’ve they’ve sent years ago. A lot of people didn’t open it anyways. You can send it back to them, add content. Add content to the system that is based on seeing insights like the CTA really resonated with this email, really resonated with customers in the past, and actually add content when it’s interesting, when it’s valuable. And instead of thinking a lot in terms of, how should I balance these communications out, etc. Just throw more content into the system, more creative, more campaigns into the system, and let let the AI system actually decide who to send it to. So it really just allows everyone to take a level like set a level higher up in the stack than they previously did, and instead of thinking about going into a Braze or Iterable or Salesforce and configuring every specific journey, or deciding, exactly, you know, should I send this email on a Tuesday? Because Tuesday overall has the best efficacy of my email, so I should send my best email on a Tuesday. I’ve seen this at Fortune 500 companies, right? Like, we’re gonna send our offer email on Tuesday, because it’s, don’t forget,

 

Steven  26:59

it’s gotta be at like, 2:30pm too.

 

Tejas  27:00

Yeah, exactly. Instead of thinking through all that, like put the content the system, we’ll figure out that it’s, it’s that the promotion offer you’re sending out is extremely effective, and what we’ll send it to, we’ll send it to customers on the day that makes sense for them. And yeah, a lot of them will go out on Tuesday. Maybe most of them will go out on Tuesday, but a good chunk of them will go out on other days, maybe, maybe 40% of them, and your campaign will be automatically more effective. And you don’t need to think about it really. It just takes every job from the people creating content and and creative that didn’t have a lot of insight into what was really working before, gives them insight, allows them to be more strategic, from CRM specialists who are in tools, configuring campaigns, instead of configuring journey sequences and multi step sequences, and where we all know we’re kind of guessing, to an extent, it allows me to instead be like looking at insights, changing goals, to tell the AI how to optimize differently, like maybe, let’s double down on cross sells. Let’s Let’s go ask my data team if I can get another feature into Hightouch so that it could look at that like weather data, right? We’ve seen customers incorporate things like weather data and just be more strategic, and be more like thinking about data instead of just thinking about knobs, and then at the highest level, like, you know, the life cycle. Marketers and the directors of life cycle don’t have to be asking if we can experiment on, you know, A versus B, the system does that they can think more about campaign strategy, what type of content we should create, what type of programs we should run as a marketing team at large.

 

Steven  28:31

The way you’re sort of talking about the goaling and the optimization are, is this particularly looking at it on like a campaign by campaign basis, one of The things that we’ve seen in a lot of our experience with our clients is, you know, one of the challenges a lot of life cycle marketers have, or even analysts have, when they’re looking at the program, is they’re looking at it with too much of a lens on was this campaign better than this campaign as opposed to, is the program as a whole, From a business perspective, doing better, right? Am I shifting not just shifting my, you know, attribution over to my email or my SMS or my push? I’m not also just pushing a purchase forward that probably would have happened anyway, but I’m truly incremental. On the business side, how does the agent understand that what it’s optimizing too. Is actually incremental, and is not just the way that we’ve been doing it forever, which is, is this campaign better than this campaign? 

 

Tejas  29:27

Yeah, really good question. So the only way to really understand incrementality, you guys debate me if I’m wrong here, you’re the experts. But it’s experimentation, right? So it’s to actually run an experiment where you do something on on Group A, you don’t do something on Group B, or you do something different on Group B. You compare the results. We really do that two levels. One, if you’re just trying out AI decisioning versus, you know, historical campaigns or journeys you’ve created. We can do a split on your your audience, of customers. And put some of them through the traditional path, put others of them through a decisioning and and compare the lift, see if AI decisioning is actually being incremental or just doing what customers would have done in the journeys and audiences you’ve built in your your email tool before. And the second thing is that AI decisioning, when it uses reinforcement learning, kind of has experimentation baked into the system that we can constantly be experimenting and figuring out if certain variant or certain content or certain campaign tactic is is better for different customers, and serving those insights to you, we can always tell you, you know, what would be the outcome if you did something different, that’s way too difficult and not possible. But we can tell you, like, what’s actually better? Like, is it, in what cases do customers prefer to receive emails on Tuesday, Wednesday, Thursday, for example. Is it, is it correlated to gender? Is it correlated to you, to age? Is it correlated to you, FICO score, right? What’s it correlated to that’s in your in your data warehouse.

 

Steven  30:56

And this is all they would be having. They would be at the same volume, ideally, right where whatever’s in your control piece here, your your AI decision treatment group, like, both would be sending five emails in total, or four emails in total. Or would you see that AI decision, you might actually send less or send more, depending on how it’s optimizing.

 

Tejas  31:18

So sometimes that’s a question we get right. Like, oh, isn’t, you know, isn’t AI decision just sending more emails, and that’s why it’s more effective. And I would say that’s one of the value props too, is that we can figure out how many emails we can actually send a customer about a certain topic without them unsubscribing. So no, it wouldn’t necessarily be the same volume. One of the value props that AI decision can figure out, like, yeah, how many emails should you email a customer about X or Y or Z without them unsubscribing, so you can actually have unsubscribes and opt outs and any other negative signal in your business is a negative reward in the ML model as well. And track that. You know, yes, sometimes more emails more effective. But if you really measure these over a long term basis, you can see if, if it’s actually more effective, it’s not an easy challenge, like if our emails was all we needed to do and was easy and we weren’t worried about unsubscribes. We all just send twice as many emails today, but obviously we are worried about about unsubscribe so that is actually one of the value props of AI decision thing, as well as to figure out that kind of optimal balance, but within some guardrails so that it doesn’t go crazy, experimenting

 

Spencer  32:21

With our remaining 10 minutes or so? Here, I wanted to bring up one more buyer. So you’re talking about CFO marketer taking a step back. You know, there’s also times where, you know the CTO or the data team, or even the IT team, are the ones that are making the decisions or have the budget or whatever. And so we’ve talked about the, I guess, the the campaign layer, but taking a step back, like, how does the in AI tools and AI decisioning? How does the, you know, data warehouse, or whatever cool term I know, we’re not supposed to call it data warehouse anymore, whatever it’s called the data

 

Steven  32:56

Data warehouse

 

Spencer  33:00

in the you know, how does a data warehouse come into play in AI decisioning?

 

Tejas  33:04

Yeah, so you’re totally right that, especially with anything relevant to AI data, but especially AI, the technical teams aren’t also going to get involved in this, and it’s for a good reason. I think there’s a lot of ML and AI features and a lot of different, you know, ESP solutions or marketing solutions that advertise themselves as they’ll just work in a click of a button and sometimes they do work, but sometimes they don’t work. And there’s nuance to that. For example, what’s the underlying data that the model is built on? And those aren’t always advertised front and center when you’re when you’re buying these solutions, so it’s good that the technical teams get involved, in my opinion. And one of the things that technical teams really like about AI decisioning is that what are what are data teams spending a lot of the time on at or at an organization? And what do they want to spend all the time on? Is they don’t want to be handing people reports all day. That’s why we have self service BI tools. That’s why they like to reverse ETL or activate data from the warehouse into marketing tools. Like, they want the business to self-serve. What they actually want to spend their time on is building a lot of interesting data products and a repository of data and interesting attributes about customers and a predictive LTV model. Like, that’s what data engineers, data scientists, data analysts, want to spend their time on creating really interesting insights about customers that the business teams can use. And AI decisioning sits directly on top of a company’s data warehouse, like Snowflake, Databricks, Google Cloud. We don’t care which one, where the where the data teams are creating these insights. And that means two things. It means one the data teams and IT teams are always like, happy about it, because they don’t have to move the data into another system that’s going to force the data into a different format and limit the potential of these ML models. They’re going to get control of the data that’s feeding into them right there in the warehouse. Two, it actually means the you. How the models perform better than a lot of out of the box models that are created in tools directly, like a Salesforce or a different email platform. And not because the ML technologies those email platforms use aren’t good they’re good technologies, but what features are fed into them, how data is feeding into them, that stuff makes a big difference when it comes to any sort of ML and AI technology in AI decisioning as a product, both are usage of reinforcement learning. So the type of machine learning that actually experiments and tries things with different customers and learns from that that’s a very, very, very critical component that allows us to automate customer journeys and automate marketing calendars. It wouldn’t be possible with just predictive models or predictive audiences, like you see in a lot of marketing tools. That’s really important. But the second thing that’s important is that we sit directly on the data warehouse and get direct access to the best data, and can even incorporate data science models, product recommendations, propensity scores, if your data team is advanced enough to be creating those today.

 

Spencer  35:58

The other hot topic right now, other than AI, at least in martech, is, you know, CDP. CDP is getting a are getting a bad rep for various reasons, at least traditional ones. So how does the CDP play a role in this AI decisioning and data warehouse, I guess, relationship?

 

Tejas  36:17

Really good question. So CDP, it means a lot of things to a lot of customers and people out there. But one of the things that it means, especially in traditional CDP context, like if you take a Segment, for example, one of the big value props is helping you collect a lot of digital data about what your customers are doing, interesting attributes about them, different pages they’re visiting on the website. We can do this as well. In our Hightouch events product, anyone can do it. A lot of companies can do it. Honestly. That’s a great foundation of interesting data to feed into any ML model, including the models that we create in AI decisioning. If you just have a table of contacts and some basic attributes about them, that’s all the AI model can be based on that and how they respond to your emails. If you have interesting data from your website and mobile app and stuff as well, and clicks and carts and all that kind of stuff, we can make that into the decisioning of what to send customers. The more data you give the model, the more, the better it performs. That’s kind of the whole idea behind chat GPT and a lot of the AIML stuff these days and CDPs have a really good source of data to to give customers, in this case, and it’s nice when there’s a little work that’s been that’s gone into organizing it, it’s not a prerequisite. We’re working with customers who don’t have a CDP. Use Hightouch as their composable CDP. Obviously, taking this to a lot of our existing customers, you know, use something like a Salesforce, data, cloud, or whatever it is. But if, if there’s any solutions upstream of us in the stack that provide good, structured data that makes everything easier when it comes to AI,

 

Steven  37:49

A lot of the applications you’ve talked about seem to be skewed towards B to C. But you know, mentioning Salesforce like is there also a B to B application you think for AI decisioning, or is the volume not large enough to really be successful.

 

Tejas  38:03

I’ll answer in two ways. Right now, we’re super focused on B to C and like anything where you have a high scale of consumers, I’m talking, you know, at least 500k customers, but probably millions of customers, to be honest. And you want to optimize how you’re interacting with them digitally on some sort of own channel or channel where you can address them one-to-one, so email, text, push, web interactions, etc. We’re expanding our Outlook to paid media as well, and CRM advertising, that’s something that we’re going to work on this year, but we are staying away from like sales and also some of the B to B use cases right now, unless you’re like a super product led growth company with a huge customer base of individual users, like Notion or Grammarly or companies like that, are great, but they’re not the typical B to B Company, if we’re honest. And I think they’re pro there are very interesting applications of AI in B to B, and in sales and in everything I mentioned, but there’s, yeah, because of the volume of interactions, because a lot of the interactions are not digital, a lot of times, because you often want to optimize on the account level and how the accounts progressing versus the user. There’s a lot of nuances that I think the technical solution might be quite different, and we’re kind of straying away from it right now.

 

Spencer  39:22

I could see there being, like, maybe a use for the like, the companies that have, like, self serve models, like Klaviyo or MailChimp, Shopify, where that is quite like, even though it’s it’s B to B, it’s kind of B to C, SMB, yeah, like

 

Tejas  39:40

SMB B to B, high volume B to B. We can do that for sure. Like, you know, on the CDB side, for example, we work with some banks that serve SMBs. And sure, there’s just millions of them, and they basically operate like consumers. At some point there’s, like, a business owner that’s managing the relationship with the bank, or maybe plus one person. Yeah.

 

Spencer  40:00

QuickBooks, like QuickBooks Online and stuff like that.

 

Tejas  40:03

QuickBooks TurboTax, I guess, is B to C as well. But yeah, for sure, those types of businesses are fair game and can operate the same way, and oftentimes use the B to C marketing tools, right? A lot of them use Salesforce marketing cloud and Iterable and Braze and those tools sometimes in parallel to like their Marketo and their ABM motion for the upper tranche of accounts. So companies using the B to C martech, technologies, where we’re focused right now. You know, there might be a company that comes out and just does some really cool stuff in AI for the B to B side. I think there’s a different set of interesting problems there. Maybe we’ll tackle it in a couple of years, but this stuff’s hard, so we’re staying focused

 

Spencer  40:45

Quite a quite a green field here for you to tackle. You got a lot of ground to cover. I think focusing on B to C is okay for the foreseeable future.

 

Tejas  40:56

I agree that that’s probably, you know, that’s what we call focus. So we have another conversation. 

 

Spencer  41:03

All right, guys, well, we are at the one minute to the end here. Just wanted to thank you, Tejas for coming on.

 

Steven  41:11

Yeah. Tejas.

 

Tejas

Go birds!

 

Spencer  41:15

Do you guys have anything that you wanted to impart before we leave? Or, you know, closing words, final words, last words?

 

Tejas  41:25

I feel like I talked a lot, Steven, you have to say something crazy.

 

Steven  41:29

Okay, I think, from my perspective, and I, you know, and I got to talk to Tejas about this late last year was, you know, this is something that companies have been building or trying to build internally for many, many decades before, not decades, but years before, like this, 

 

Spencer

1920s 

 

Steven

Over the last decade. And I think one of the interesting things is, like, nobody’s really ever, like, thought, like, man, we should just build a product that does this, right? Like, I think some CDPs kind of touched into it, some ESPs kind of did, nobody really, like, built, you know, a system that just be the AI layer, right? And, you know, because they either they wanted to own more of the vertical, or they were like, I don’t know. I really don’t understand, like, why it wasn’t built before. Was always something you needed a large data science team to build. So yeah, Tejas, I think just like in from a from a business perspective, it’s vastly interesting that you guys, one identified like this as an opportunity, and two, like, solves a pretty big gap in what you would be, from a personnel perspective, to do it yourself, right? And so I think you’ve got a nice carve out here where, like, much once you’re in, like, I can’t imagine unseating yourself, right? I mean, maybe there’ll be competitors at some point. But, like, the only other way to do this is, like, hire 20 to 40 data scientists, right? Like, it’s not easy to do, and I think it’s just really interesting that you are bringing this and making it more attainable for marketers to actually, like, make their find ways to actually make their program improve without needing to either hire more bodies or hire a completely separate function.

 

Tejas  43:19

Yeah, and actually usable by the marketing teams too, right? So, 

 

Steven

There you go. Yeah.

 

Spencer  43:23

All right, guys. Well, that’s all for today. Tejas again. Thank you so much, Steven. Thank you. 

 

Tejas

Thank you.

 

Steven  43:31

And welcome back. Spencer.

 

Speaker 2  43:33

Where did I? To the podcast? 

 

Steven  43:37

To the podcast. 

 

Spencer  43:38

Okay, thank you. No one welcomed me. What the hell? Thank you. Thank you. Thank you. All right, guys. 

 

Steven

Bye