Nir Weingarten Podcast Transcript

Nir Weingarten Podcast Transcript

Nir Weingarten joins host Brian Thomas on The Digital Executive Podcast.

Brian Thomas: Welcome to The Digital Executive. Today’s guest is Nir Weingarten. Nir Weingarten is the Co-Founder and CEO of Eikona, a startup using gen AI and reinforcement learning to transform life cycle marketing. A published AI researcher with a master’s degree in machine learning from Reichmann University, Nir spent over a decade leading multidisciplinary technical and product teams across the fields of AI, data, and performance. 

Together with Co-Founder and CTO Omer Hacohen, Nir built Eikona to disrupt an area of marketing that has barely evolved in 20 years. While social media feeds have been algorithmically personalized for over a decade, the emails, SMS, and push notification brands used to retain customers still rely on slow, manual A/B testing. 

Well, good afternoon, Nir. Welcome to the show.  

Nir Weingarten: Hi, Brian. It’s a pleasure being here.  

Brian Thomas: Absolutely, my friend. I appreciate it, and I appreciate the fact that you’re in the Tel Aviv, Israel area making time to traverse time zones and calendars to get to Kansas City today. So again, really appreciate it. And Nir, we’re gonna jump into your first question. 

You’re a published AI researcher with a machine learning master’s degree who spent over a decade leading technical and product teams across AI and data, and then chose to apply that background not only to the most hyped AI categories, but to email, SMS, and push notifications. What did you see in life cycle marketing that made it the right place to build a company? 

Nir Weingarten: Right. So I, can I ask that back with a question? Can I, can I answer that with a question?  

Brian Thomas: Absolutely.  

Nir Weingarten: Okay. So Brian, in the last week did, did you see on social media a video, one or two videos you think that were generated by AI?  

Brian Thomas: Absolutely. Every day.  

Nir Weingarten: And did they involve furry mammals?  

Brian Thomas: I think I saw one. 

Yeah. Mm-hmm. Yep.  

Nir Weingarten: All right. And these are the ones I get a lot. Okay. And, and now for the second question. Do you know a single person that bought a plane ticket with an agent, with an AI agent?  

Brian Thomas: Yes.  

Nir Weingarten: Okay. Did they get to the place that they wanted to?  

Brian Thomas: Absolutely.  

Nir Weingarten: All right. That, that’s amazing. I don’t know anyone that bought a plane ticket but I see videos like that every day. 

And the point I’m trying to make is that from my experience, from my point of view, I think AI today at least is inherently good at, it’s inherently talented in in content and in engaging people hacking their dopamine systems and understanding what makes people pause and look at a video and pause and read something. 

And there’s, there’s other things that it struggles with. I think it’s much harder, at least from my point of view, to create a system that would do a series of actions that would actually have a, a high stake purchase at the end. Much more easier for the technology to create engaging content. 

And when you think about what’s the simplest form of content that this technology can actually automate end to end and that has a lot of market value too, that’s how we got to lifecycle marketing, and that’s how we got to stuff like emails and SMS messages. Think of an email. An email in many cases, it’s an image, some microcopy, a subject line. 

And now don’t get me wrong, creating an email from scratch for a mid-market or enterprise brand, then that, that email would be good enough to actually be shipped to tens of millions of of people is extremely difficult. But it’s possible to automate today end to end with today’s technology. If you look at a website, for example, to create a website or to change a website dynamically with AI, with not breaking stuff there, that’s a couple of orders of magnitude more complicated. 

Also videos, by the way. So lifecycle marketing is in my opinion, in our opinion, the lowest hanging fruit for AI automation in terms of what the tech can do and in terms where there’s a lot of commercial value. So I would say that’s the, the first, the first and most important aspect. And also it’s a very underserved market. 

I think that it hasn’t been really disrupted since the early 2000s with the first marketing automation platforms. More or less we’re doing the same thing when it comes to marketing automation since, right? We have user customer journeys. We built state machines. We send out blast emails exactly the same. 

We have very GUI today and some more features, but the concept is the same, so it’s underserved market. And, and lastly, I would say as a data scientist, it’s a place where you have a lot of data. So when you’re talking about your CRM, if you’re a B2C company, a B2C enterprise, you have a lot of data about your clients, and you know what they bought, you know when they bought it, and you know what device they’re using. 

Is it an Android or an iPhone or a desktop? Are they Yahoo.com, Hotmail.com, Gmail users? What’s their zip code? And because they, they did a purchase, and if you know their zip code, and that’s your data, you can also infer other information about them, and you can use that rich, rich data to give them content that would work so much better and be so much so much more intimate and engaging for them and give them such a better experience. 

And that data today is almost non- utilized at all, and it allows us to build a great product, a great technology and, and then also a moat around that.  

Brian Thomas: Thank you. Appreciate that. Really do. Unpacking that for our audience today. And you’re right, there’s a lot of … And I appreciate the, the initial questions there, but today, AI can do a lot of things. 

It’s very competent in, in most tasks, but as you mentioned, there still are some gaps. But you’ve nailed it here in the life cycle. Marketing automation is certainly a UK- a use case for an AI to really knock it out of the park, as I would say. And of course, it maximizes under certain markets. 

So, appreciate that. And Nir, Eikona applies reinforcement learning from human feedback, the same technique used to fine-tune large language models to continuously adapt marketing content based on how customers actually engage with it. How do you explain that technical approach to a chief marketing officer who doesn’t have a machine learning background, and what does it feel like to use from the marketer’s seat? 

Nir Weingarten: Thanks for the question, Brian. Think it’s spot on. Well, I think reinforcement learning is something that’s very intuitive to understand because it’s very similar to how we as people go about the, the world. And I’ll just give a very crude example. So imagine a baby is born, and that baby crawls around the world and starts to discover it, and they don’t know anything about the world yet. 

So they explore their environment, and they try out different stuff. And then the baby crawls around the floor and finds a piece of candy. So maybe it’s on a plate, ’cause he doesn’t pick the candy off the floor, but he finds a piece of candy, and the baby tastes it and, and it’s sweet and it’s tasty. 

And the baby has been rewarded by this experience, and the next time they’re, they’ll see a piece of candy, they’ll try to eat it. And, and the next day, say that baby roams about, and then they found a, a, a slice of lemon. And they see the lemon, they try it out, and the lemon is sour, and they don’t like it. 

It’s not tasty. So the next time the baby sees a lemon, they’re not gonna try it. That is reinforcement learning. And as they’re very crudely… What do I, what do I mean by that? I mean, you have some sort of, of autonomous agent. In this case, it’s the, it’s the toddler. You let them explore their environment, which is try out different foods. 

And, and you let them learn by experience what rewards them and what not, what’s tasty and what’s not. And to complete the analogy, we can say that we can train an AI model to try out different creative concepts or different content or different messaging, different creative intents, and learn by how actually people click on that or buy from that, what’s engaging for these audiences. 

So we are replacing the baby with a neural net that generates content, and we’re replacing the taste buds with counting clicks and revenue. And so that’s the concept. I think it’s pretty, pretty intuitive. And, and I think that a lot of marketers, you know, the first thing we ask, and we, we, we talk with really hundreds of marketers we ask them if they A/B test. Did, did you ever A/B test something Brian?  

Brian Thomas: Absolutely.  

Nir Weingarten: And how did that work for you?  

Brian-Thomas: Great. Gave some great feedback on which direction we should go based on the results we got back on the testing. 

Nir Weingarten: Great. And, and, and would you say you, you A/B tested enough or did you want to do more A/B testing?  

Brian Thomas: Love to do a lot more. Sometimes based on deadlines, we didn’t have the time, but, but yes doing more, we wanted to really dial that in so we could maximize our results.  

Nir Weingarten: That’s the answer we keep hearing across the board. 

I never, never heard otherwise. And when you ask people why, they tell you, “Well, we don’t, we just don’t have the time. It doesn’t scale. And it was very effective when we did that, and you remember the sale two years ago, and we try, but we never do enough like we want to.” And, and then we say, “Okay.” 

So, this is a very big problem. Now we’re going to create scalable A/B testing, automated A/B testing. The– So it, it’s called reinforcement learning, but it’s actually the next iteration of A/B testing. It’s more or less the same thing. It’s like A/B testing that just scales. And so I think for a lot of marketers, this is a very intuitive concept, both on how it works and the need for it, because people are really waiting a long time for that to happen. 

And, and on the marketer side basically we try to make the product integrate into the existing workflow that people have. So… And you want to make this as frictionless as possible. Nothing is a hundred percent frictionless, but you don’t change your tech stack. You remain with the same tech stack that you have. 

You go about your day the same way. The minute you hit send- To that email campaign or flow, you’re gonna be prompted by the system. It’s gonna create a Slack message. You’re gonna get that on Slack saying, “Hey, here’s 20 variations the system created, and select the ones that you like the best.” This, this is what the system recommends. 

This is– These are the pieces of candy the baby would like to try out. And you select the ones you like and then these go out to, to see how, how tasty they are. So, it creates another sta- another step in your day when you, you don’t just send out an email, you have to select the variations each time. 

Brian Thomas: Thank you. Really appreciate your insights and, and really, again, unpacking that for our audience. But I liked how you talked about that reinforcement learning. It’s intuitive to understand, and your example was that baby, new baby crawling around discovering the candy or the lemon and being rewarded for the candy and not so much for the lemon. 

But you are training AI to learn from consumers and their engagements or their incl- their clicks. And having that automated scalable A/B testing with your platform allows businesses to really dial in really what they want, the results that they want much faster and obviously much more accurately and to have a better results more return on investment. 

So I appreciate that. And Nir, Eikona is deployed primarily in retail, but is expanding into telecom, finance, insurance, healthcare, and travel. What do those sectors have in common that makes lifecycle marketing such a high-value problem there, and where are you seeing the most dramatic early results?  

Nir Weingarten: Right. Thanks for that, Brian. And basically we, we can serve any B2C business, and that’s large enough to have enough data for the system to learn. And mostly it comes about around a hundred thousand people on the, on the list that gets the, the content, the emails, the SMS. The bigger the, the list is, and the faster we can learn and the more iterations we can make. 

And we started off in retail because it’s, it’s just so… A, a market that’s very early, it’s, it’s, it’s, it’s– has a lot of early adopters, and it’s easier to build the product there. But as we went, we expanded to other verticals, which are in many cases bigger. And, I talked to… So, this is a s- a quote from a, a CRM manager for one of the biggest banks in North America, and he told me this. 

He told me, “You know what we do at this bank? We do two things. We need to very quickly compute the interest we give you on a loan.” And we need to keep you from going to the other bank across the road, ‘ cause they have exactly the same the, the, the same algorithms to give you– to compute your interest. 

So I, I think in many B2C services and B2C enterprise companies, retention is a, a very, very important aspect of business, and it can be, very, very profitable for these businesses to improve their messaging to their clients. So basically, we would like to work with brands that are– companies that are as large as possible, and also that retention is a big channel for them. 

In many cases, you’d see revenue from retention reaching as, as high as sixty percent. And, and we generate approximately between twenty and forty percent uplift on that. So that can be very, very impactful, especially for bigger companies.  

Brian Thomas: Thank you. And those are interesting sta-statistics, by the way, that revenue from retention can be as high as sixty percent. 

I thought that that was interesting, and then the, the uplift alone could be twenty to forty percent there. But I like how you can serve really any B2C business with the… You’ve got approximately a hundred thousand list of users that can be tested with with some great sampling of data here. 

So, again, really appreciate what you’ve shared there. And Nir, last question of the day, if as we look ahead in five years, you want Eikona to become the market leader in a new category you call adaptive marketing, where every message that reaches a customer is personalized to be more intimate, more warm, and more effective. 

What needs to be true technically, commercially, and culturally for that category to become the standard? And what’s standing between where lifecycle marketing is today and that future? I  

Nir Weingarten: love this question. I think to, to break it down like you did on the technical level, we’re, we’re there. We weren’t there a year ago, but we’re there today. 

Today we create end-to-end emails and SMS to some of the biggest brands in the world, and we do the complete email we do the creatives we, we keep the brand, we keep compliance, we keep everything there, and we do that with our proprietary tech stack. So techno-technology-wise, we’re there. And commercially, which maybe is the biggest thing I think the best analogy here is Bri- is Geoffrey Moore’s Crossing the Chasm. 

Did you ever get a chance to read that book?  

Brian Thomas: I have not.  

Nir Weingarten: I would really recommend it, one of my favorites. So, so Geoffrey tells us that in evolving markets, there’s always an early market and a late market. Early markets are triggered by early adopters of technology. These are people in or-organizations that are either visionary by their character or really sick of the, the opportunity to advance using technology. 

So, right now since the technology is pretty early on, despite that it’s working, the– it’s still an early market technology. It’s still for people that are that, that love, like to adopt new technology. It’s still the people that say, “I’m gonna get AI in the organization,” and really, really mean it. 

Okay? To cross over the other side to the mainstream, to the early majority where most of the market sits you just need more time to amass masses of, of the early adopters. So that’s, that’s at, at least the, the thesis in Crossing the Chasm, which I really connect to. Think of iPhone users, 2008, right? 

So you, the iPhone was very, you know, it was, it was the ver– it was a niche back then. N-not everyone had an iPhone. I had a Motorola. Some people had a Nokia. But then everyone had an iPhone, and it happened, like, real quickly. Or think of Salesforce. So imagine Salesforce coming out as a product and, and you have Marc Benioff telling people they need to put their CRM data in the cloud in someone else’s server No one wanted to do that. 

That was crazy. So I would put my precious CRM data. People had that on-prem. People had a, a s- a box in their office that had the data inside. And now when you think of it, how crazy does that sound? For someone to have a server in their office with the CRM data in there, and they have like a proprietary software to use that. 

So it’s just a matter of time until the market reaches there. And lastly c-culturally, like you said there’s a great term I, I, I read about. I really liked it. It’s called media panic. Media panic is basically the, the, the concept is, goes like this. So every time a new type of media is introduced the, the initial reaction is a suspicion. 

And because, w-we are what we consume we, we are, we think, our all co-cognitions around the, the information that we consume. And whenever that changes it’s, it’s a bit frightening. So a recent example would be social media. So if you remember when social media just came out, do, do you remember the backlash on that? 

Brian Thomas: Yep.  

Nir Weingarten: Yeah. You know, it, it was really frightening. It is frightening still, but you know, it’s still, but it’s a standard, everyone uses social media. So, it was the same when video games came out, and it was the same when, when television came out and when movies came out. And you know what? 

There’s even a quote that I saw by, that attributed to Plato where he blames writing for, for people being messy and forgetting stuff. And, and I think writing is pretty much mainstream today. And so I think that culturally, AI is, is very similar. It started off where people really resented seeing AI content, and today we see it all around us, and in a matter of a few years it’s going to be, in my opinion, as ordained as, as, as seeing a TV show. 

Brian Thomas: Thank you. Appreciate that. You know, we talked a few things. I always like to highlight some things here and technically and commercially; we talked about that your platform is now capable and moving towards this adaptive market category. And you shared some examples, iPhone in 2008, Salesforce CRM moving to the cloud, and you saw that these markets shifted very quickly, in fact, overnight, really. 

And then you talked about that cultural AI, right? You know, you– the example was that media panic. When something new or controversial is released, people kinda freak out initially, and then it just kinda slowly becomes part of our DNA and, and you shared some ex- some great examples there, and I really appreciate that. 

And Nir, it was such a pleasure having you on today, and I look forward to speaking with you real soon.  

Nir Weingarten: Thank you so much, Brian. It was a real pleasure being here.  

Brian Thomas: Bye for now. 

Nir Weingarten Podcast Transcript. Listen to the audio on the guest’s Podcast Page.