We all know that customers want personalized content, but your human team isn’t big enough to create all the content variants needed for your hundreds of thousands of customers, and simply sending a prompt to Gen AI doesn’t give you the control you need over your brand messaging.
But what if you could have 1 agent per customer that kept your messaging on-brand, learned exactly what that individual customer wanted, and was able to adapt to their preferences over time?
This episode is brought to you by MoEngage. MoEngage is an Agentic Customer Engagement Platform trusted by over 1,350 global consumer brands and combines deep customer analytics, agentic marketing workflows through its Merlin AI Agents, and omnichannel orchestration across messaging and surfaces, helping marketing and product teams deliver experiences that feel personal at any scale. Learn more at moengage.com
Welcome to One Amazing About Aampe by MoEngage. Today we’re talking with ZachDorner, Lead – Solutions Consulting and he will be sharing one amazing thing about their platform with us today.
Watch the Demo
About The Guest
Zach partners with our marketing and product teams to understand their CX challenges and design tailored technical solutions that drive results.
Zachary Dorner on LinkedIn: https://www.linkedin.com/in/zach-dorner-6ab081124
Resources
Aampe by MoEngage: https://www.moengage.com
The One Amazing Thing with Greg Kihlström website: https://agilebrandguide.com/the-agile-brand-podcasts/one-amazing-thing-about-with-greg-kihlstrom/
Subscribe to One Amazing Thing on YouTube: https://www.youtube.com/playlist?list=PL-BawE82qIM111hMVR5AJbE1uLdCMdYog
Get the latest news and updates on LinkedIn here: https://www.linkedin.com/company/the-agile-brand/
Check out The Agile Brand Guide website with articles, insights, and Martechipedia, the wiki for marketing technology: https://www.agilebrandguide.com
From the creators and host of the award-winning The Agile Brand™ with Greg Kihlström podcast, comes One Amazing Thing About… with Greg Kihlström, a show that puts the spotlight on the latest and greatest MarTech products and platforms by showcasing one feature that makes them… amazing.
Transcript
Greg Kihlström (00:00)
LLMs are great for a lot of things, but they aren’t great at everything. Plus, they can be expensive for some things and slow for others. What if you can integrate with Jev an AI decision model that can move more quickly than an LLM?
Welcome to One Amazing Thing About Tealium. Today we’re talking with Zach Wenthe, Director of Product Marketing and Customer Data Evangelist, and Sarah Smith, Staff Solutions Consultant Engineer at Tealium. And they’re going to be sharing one amazing thing about Telium’s integration with Jev with us today.
Zach and Sarah, welcome to the show.
Zack Wenthe (00:41)
Thanks, Greg. Thanks for having us.
Sarah Smith (00:42)
Thanks, Greg.
Greg Kihlström (00:44)
Yeah, yeah. looking forward to seeing this. I I gotta say, yeah, I’m very curious to see this in action and and we’ll do that in a second here. before we do that though, why don’t you each give a little background on you and your roles at Tealium?
Zack Wenthe (00:57)
Sure. you know, so as you said, I’m Zach Wenthe, director of product marketing. But my background, I’ve been in Martech for, you know, most of my career started in consulting, doing strategy work, doing, you know, implementation before I moved over to the tech world and I’ve been in a variety of, you know, CDP and customer data infrastructure ever since. And yeah, I was I was a marketing technologist before Scott Brinker made that cool.
Sarah Smith (01:23)
Hi everyone. So my name is Sarah Smith and I’m a staff solutions consultant engineer here at Tealium So my background and my role is essentially to be one of the engineers for all of our field organization here at Tealium. I’m the lead engineer who pioneered the invoke your own model framework at Tealium and my background is much more in the computer science realm and also for machine learning, everything from machine learning models, LLMs, Gen.AI.
So everything right here is really, really focused on bringing the power of models and AI to marketers and all of you today.
Greg Kihlström (01:59)
And we’re gonna take a look in a second here at at this in action. But before we do, why don’t you start by giving a little background on what are we gonna take a look at today? Who’s the audience? Why is it important to them?
Zack Wenthe (02:12)
Yeah. Well, as you said, Jev is a is a new model that was released just recently by by Typesafe. And if you’ve been online in like the last two weeks, it’s probably been all over your feed, whether you’re on on on X or or LinkedIn. And fundamentally, it’s because it’s a very different type of model. Traditional frontier models like, you know, anthro anthropic, you know, Claude or you know.
Chat GPT are text-based. Yeah, it’s a chat experience. You’re, you’re giving it a prompt. It’s giving you words back. Yes, it can code and do other things, but it’s typically tied to first and foremost to letters. Whereas Jev is really about giving a an answer back, right? It it’s it’s trained a little bit differently. So it’s a kind of reinforcement on giving this you know, response. And that can either be a yes-no response or it can be a you give it a rubric.
So here’s what we kind of determine is, you know, maybe high, medium, and low. And it will give you back a structured you know, JSON object response. Why is that important? Because now you can start to do you know, decisioning right in you know, right in a stream in real time. It’s very, very, very fast millisecond response and it’s very cheap. in fact, just before we before we dive into the demo and kind of show you some of the stuff.
We were kind of doing some estimates. If we took a customer’s profile, like their entire historical profile, and we have a kind of benchmark of how many tokens that that generally takes, and sent a million of them to to Jev, we could score all a million of those profiles at their at their regular rate for about sixteen hundred dollars. and and typically you wouldn’t send the entire historical profile with all of the details,
Now, if you did that in like Anthropic or ChatGPT, it’s about $400,000. so you’re you’re talking orders of magnitude cheaper, faster, and you know, in a much more structured manner. So you can kind of get a you know what to expect when you when you you know ask for a response back. so it just opens up a whole new way that that marketers and data analysts can can build better customer experiences. in line, on the web, on mobile devices and and whatnot.
Greg Kihlström (04:36)
Well let’s let’s take a look. Why don’t why don’t we take a look at it in action?
Sarah Smith (04:41)
All right, so this is Tealium platform and essentially what this is, is Tealium functions. And this is really the foundation of where Invoke Your Own Model comes into place. So traditionally with Invoke Your Own Model, and we’ve pioneered this, it was essentially a way to be able to use Tealium functions, which lets you put in custom code, call out to custom APIs, and especially call out to models for machine learning models, send prompts to LLMs. and be able to get back a response like scores, reasoning, and then Tealium will act on that in the current session and also distribute the updated profile downstream. So from the natural progression, because yesterday at first was machine learning, then we have all of the different types of LLMs. So it’s really easy to essentially insert the Jev decisioning framework right here in Tealium functions. And I’ll show you how that’s done.
So there’s going to be a little bit of code here that we see on the screen. But for all of our business users and marketers, Tealium does have an easy way just to ask AI, and Tealium will write the code for you. But let’s take a look at what this is. when I have, essentially taking a sample, this is her profile. This is a retail customer who’s been engaging on mobile app. And she’s abandoned her cart. So she’s checked out had items to the cart and she left a checkout process. Typical classical marketing use case. And Tealium especially by the way of how we have a rich data layer with all of the different attributes in business language, this is ideal for essentially asserting directly into the state for Jev Because Jev takes the state, in this case, we’re asking it to evaluate the visitor profile as customer behavior that’s happening right now.
on your own and operated properties, formatting a few different items based on how it is like taking out key dates, information, lifetime value, as well as what just what they’re doing right now in their current session and making that available to Jev From here, special Tealium functions, when you need to really quickly be able to keep up to keep up to date with all the changes in
LLMs like yesterday, everything’s been so far with LLMs and trying to send the different prompts. Now we have Jev, which really shortens that time. And tomorrow, like we even see with OpenAI, they’re coming out with their own decisioning tool endpoint as well. So, essentially we’re just creating the questions and asking information about your customer as they are in their session and being able to really quickly iterate.
And then all we have to do is once again use Tealium functions and we have secure authentication to send the customer profile the questions and run it against Jev. And the results are really instantaneous. So right here you can see end to end tests. What happened? We took the visitor profile as a trigger from either an audience or you can trigger from an event for anyone who’s really focused on the event data collection and activation in real time sent it to Jev and got back response and triggered a Tealium event that will then be able to send to different downstream connectors, especially for customer journey orchestration, send to ad visitors, ad campaigns and social media platforms, and also do onsite or in-app personalization. And Jev, just like Zach mentioned, it’s really quick at being able to evaluate the customer profile pull out information. So here we have like propensity purchase if they don’t have any intervention, pulling out information about why they abandoned next best action. How should we go about retargeting this visitor as well as push notifications? Like what channels should we go for? this really, so Jev and Tealium’s invoke your own models framework makes it easy for companies to get started experimenting with Jev right now.
with tools they already have and be able to then quickly have Jev make these decisions for you instead of having to try and figure out what the best action is or how should you interpret the visitor profile.
Greg Kihlström (09:11)
Yeah, love it, love it. And I mean, for those watching, we didn’t like edit that or speed that up or anything. That was in real time here. So that is pretty, pretty amazing, I would say. So yeah, I mean, I think you know, from from my standpoint, and I think there’s been a lot of talk this year. You know, we started out with token maxing and like how how many tokens can we use, and then all of a sudden people got the bills and they were like, Hey, hey, let’s talk tokenomics. So there there started to be this conversation of
Well, do we really need an LLM? And LLMs aren’t even good for everything anyway. So but it was it was always this like binary, like do we use you know, ML, do we use LLM? This feels like some is it is it safe to say it’s kind of somewhere in the middle or like what how how should how should marketers think of this?
Zack Wenthe (10:01)
Yeah, I I I think that is the idea, right? It’s it sits in the middle from a you know, from a business perspective. Obviously it’s significantly cheaper. you know, as Sarah showed you, it’s you know, it’s fast, it’s efficient, it allows you to test and iterate very, very quickly. and because it’s a structured response, data engineers, data scientists, you know, they’re gonna like it, obviously, because they can kind of, you know, build a much more structured, you know, as opposed to having to parse out maybe a response back or force.
a traditional LLM to give you a JSON response. but for the marketers, like in that scenario, you don’t have to write that code. You know, it’s if you have somebody on your team who can kind of hack away at it, it all flows back into the UI. So it’s available as an attribute, you can, you can segment off of it, you can build off of it. And it’s yeah, it’s just the natural like next evolution of like pick the right tool for the job as opposed to forcing everything into a AI problem it’s it’s now, you know, it’s the first, it won’t be the last.
Sarah Smith (11:05)
And to add onto that too, a lot of our customers are using Tealium’s AI connectors to send prompts to different LLMs and get back response already. So another application of Jev in Tealium functions is to be able to act as a harness, decisioning harness to determine what type, when should the, when should, like, is this appropriate time to actually be invoking a LLM model? Or should we, or should you wait? And because of how cheap and fast it is as well. You can now start to have more AI responses and more AI decisioning throughout the customer journey instead of trying to very carefully balance cost versus real-time personalization.
Greg Kihlström (11:46)
interesting. Okay. So yeah. So you could actually get it to help you decide when to use it and even factor in the cost, you know, potential or or you know, yeah, the cost of it, right? Yeah. Interesting.
Sarah Smith (12:00)
Yes, absolutely.
Greg Kihlström (12:02)
Love it. Well, yeah, this is this is great to see. I know, you know, there’s probably plenty more, plenty more we could see, but if people want to learn more about this, where should they go?
Zack Wenthe (12:11)
Yeah, Tealium dot com. you know, we we already have a blog post up on it. Sarah’s writing a couple more technical specs on it. you know, this is there there’s gonna be a lot going on. or, you know, join us on LinkedIn as well.
Greg Kihlström (12:24)
Well, again, I’d like to thank Zach Wenthe and Sarah Smith from Tealium for joining the show. You can learn more about Zach, Sarah, and Tealium by following the links in the show notes.
