#282 Chris O’Neill: How GrowthLoop is Using Agentic AI for Real-Time, Personalized Marketing

0:00If you think about what's happening with Google News, Spotify, Netflix, these recommendation engines are getting so much better. and why is that well it's actually not just old-fashioned machine learning and reinforcement learning it's actually artificial intelligence in combination with machine learning the humans have a very important role they'll be the gatekeepers ultimately these tools are in service of the relationship between consumers and brands so getting the data in the right place investing in a culture of experimentation making it okay to fail and velocity is probably even more important it's always important in business to move quickly and in the right direction but now with the models changing so quickly, the underlying technology, you got to move fast.

0:39Build the future of multi-agent software with agency. That's A-G-N-T-C-Y. The agency is an open source collective building the internet of agents. It's a collaborative layer where agents can discover, connect, and work across frameworks. For developers, this means standardized agent discovery tools, seamless protocols for interagent communication, and modular components to compose and scale multi-agent workflows. Join Crew AI, Langchain, Llama Index, BrowserBase, Cisco, and dozens more. The agency is dropping code, specs, and services. No strings attached. Build with other engineers who care about high-quality multi-agent software.

1:38Visit agency.org and add your support. That's A-G-N-T-C-Y dot O-R-G. Hi, Craig. I'm Chris O 'Neill. I'm the CEO of Growth Loop. I also serve on the board of Gap and have been really fortunate to have a great career in technology. Big companies like Google, smaller companies like Glean. I've started a few things of my own. So I'm delighted to be here with you today. Can you tell us what allowed you or drove you to leave Glean and become CEO of GrowthLoop? And just give a quick thumbnail of what GrowthLoop does. Yeah. So Growth Loop applies agentic AI on top of a data cloud like Snowflake or Google BigQuery Databricks, whatever data cloud or lake or warehouse you choose.

2:37And the whole point of that is that you as a marketer or a data team can drive growth faster. And we can unpack more specifically what that means. But this is a company I've followed, invested in a while back. I was fortunate enough to partner and work with the two founders of the company, Chris and David, dating way back to my time at Google when they were looking at Google and seeing that there just had to be a better way to do marketing. This is a company that was bootstrapped from the very beginning. It's not raised a dollar primary capital. Google was our first and our largest customer to this day.

3:15So it's been a fun ride. I've been an investor, advisor, joined the board, and most recently was asked to lead the company. So it's been a fun ride. And Growth Loop is really at the right place at the right time, really trying to help out marketing and data teams solve some really complicated problems about how to get personalized relationships with their customers up and running much more efficiently and effectively than has been possible before AI. Yeah. Can you tell us, tell listeners how GrowthLoop works? I mean, it started from the notes I've seen as a composable customer data platform. First of all, what is that?

3:59And then how's it grown since then? Yeah. So all these terms that get applied, like that's not really what the company cared about from the beginning. The company really cared about just very specific business outcomes. So it started off trying to figure out how to reduce churn or make it much easier to acquire new customers. So it was very much from, hey, what business problem are you trying to solve? And then it was a necessity or luck or good fortune was built on top of data clouds. At the time, there was this belief that, of course, customers are important and therefore customer data is important.

4:34Therefore, we should build these separate pools of data on customers. We should copy data all around and really queue up a bunch of workflows off of that. As it turns out, again, out of necessity or good fortune, the company built on top of data clouds from the beginning, not trying to have a separate pool of data, but to say, hey, smart companies in the future are going to put all their data into data clouds. That's why we've seen the rise of Databricks and Snowflake and Google's cloud business. So, number one, it sits on top. That's the concept of composability. It's not some monolithic trying to do all things all people.

5:10It sits on top of the data. And then it's an intelligent layer on top of that that basically applies AI different steps of a growth loop. Or you say, hey, I want to try to maybe acquire a new customer. Or maybe I want to make an existing customer more loyal. I want to expand the number of products relationship in some meaningful way. So the AI basically sits on top of the data and understands the data first and foremost. And then it suggests ideas, experimentation to say, hey, maybe you should try to offer Craig this, whether this is a specific service or a product, and then it launches that, learns from that, and then iterates quickly.

5:48I'm simplifying a great deal, but ultimately it's about how do you reduce the complexity for, in this case, a marketing team trying to partner with their data and technology team so they could get through iteration loops or experimentation loops faster so you actually learn better and faster. And now AI does a really nice job at many of those steps, not all of them yet, but in the future, I imagine that will very much be the case. Yeah. And you said at the beginning that it's agentic. So, uh, it does it, do you have a conversational, uh, interface, uh, where people can, can say what they want, uh, in natural language and then does the, uh, AI layer find the data, compose the email or the, the online campaign or whatever it might be and then execute that?

6:40Yes, it can. So a lot of questions in there. It is an interface that you'd recognize for now anyways. Like chat, I think is an imperfect, it's an imperfect interaction mechanism, but it's the best we've got. We can talk if you're interested in terms of future design mechanisms, but absolutely it starts with asking a very simple question. What are you trying to accomplish today? Again, trying to increase lifetime value of a customer. You're trying to reduce churn. What are you trying to do? Trying to grow category sales? Whatever it is, that's the first question. And then the first agent goes to work to understand the query, understands what you're trying to accomplish.

7:17And then it starts to wake up the different agents. It says, okay, the data agent is going to say, okay, great. Based upon the data schema, the underlying data in the data cloud, what are some ideas? To augment the human idea generation facility. They say, hey, look, in the past, when you're trying to do a similar thing, here are three campaigns or three specific segments that seem to work. Would you like to build on those? And so suggesting some ideas. From there, you can basically say, yeah, I want to work on that segment. Then you say, OK, great. What journey do you want to take them on? So in other words, what do you want to offer them?

7:51Is it an app experience? Is it an email? Is it a paid advertising? whatever it happens to be in terms of the actual journey itself. And then it does indeed coordinate with all the different destinations, hundreds, in fact, through APIs to basically say, great, across all these different channels, let's just go get it done. Removing complexity, which is currently handled typically by teams, by channel, by tool. So every one of these steps that we're describing are manual and they're time consuming. And they usually involve interaction across multiple different functions, usually a data team and a marketing team that are going back and forth.

8:31So this can often take weeks, months, gosh, sometimes longer. And really that's just, to me, is unacceptable. So the problem is like, how do you reduce complexity? How do you increase the speed and reduce the manual nature of any one of the tasks, including the handoffs from one step of the process to the other? So that's really what this is about. And agents increasingly are at every step. And as I said earlier, some of the agents are really working great. Others we're really experimenting with right now because the models are getting better, but they're not quite there. An example of that would be like generating creative briefs and the actual images and the copy itself.

9:14Like that's pretty good, but it's not all the way there yet, if we're really honest. The other agents choosing who to focus on, we're getting really good at that. And that's been our bread and butter from the beginning to say, hey, we're about who is it and how can we orchestrate this conversation? We're really good at that right now. And it's just getting better at a really pretty impressive rate and not necessarily just due to our good work, although I'm very proud of the team's work. It's actually the underlying models. And we're comparing and contrasting all the different LLMs to see which model is great for which task.

9:47And of course, there's cost effectiveness and security and all these things that factor into that decision, too. Yeah. And how do you reach out to customers? Do you have various channels that then the AI or whoever is managing the AI can choose between? This is something I've been interested in for a while. I mean, how do you reach the customer? and they're, I remember talking to a company that's based out in Bozeman. I can't think of its name right now. They're an AI ad company. But they would, you know, they certainly had direct email, but they could position ads that would reach the customer through social media or through some other, maybe Google.

10:46I don't really understand the tech behind that, but, you know. So how many different channels do you have? Hundreds, hundreds and hundreds, and even more if you think about the extension and reach through networks. So that's, I think that's table stakes, right? So the connectors or the destination, we call them destinations, or sometimes we call them surfaces. The surfaces are really up to the marketing team themselves. So they decide. Often it's just an email or an in-app mobile experience, but it really is any permutation of any of the destinations that are relevant to their customers, whether it's paid advertising, whether it's social, whether it's in-app email, push notifications, it's any or all the above.

11:33So that's pretty much table stakes for what we do. And of course, those channels grow and contract and collapse. But the real benefit isn't just the reach of all of them. I think, again, that's table stakes. It's really understanding what's the sequence, right? What's the timing? Ideally, you want to engage an existing customer on your own channels, your own and operated channels, right? So we're seeing a huge business with retail media as an example. So the Walmarts of the world, increasingly Costco and Gap and all these other companies, these big brands that have relationships with their customers, and they're actually facilitating advertising revenue from their partners that want to get in front of their customers.

12:19So it really depends on the context, but really table stakes is to have that reach. The real benefit is the intelligence to say, hey, what is precisely the right place and the right time to have a message that would resonate across the entire life cycle of that customer? So that again, it doesn't feel like an interruption, or doesn't feel like something that is antagonizing them, right? Often, these experiences, more often than not, are just kind of awful, really. And we're out to stop that, right? And if you're offering something at the right time in the right context, it can be a delightful experience, right?

12:58And we're trying to aim and help enable teams to do that more often. And that's really how growth is going to happen, right? You're going to have more loyal customers that care about you because you're actually feels like you're listening to them as opposed to just bombarding random, not relevant messages to them all day and every day. Yeah. And the classic example of that, you know, online marketing that drives me and I'm sure everybody crazy is, you know, I do a search for outdoor umbrella and, you know, I find one and I buy it. And for the next two weeks, I'm getting ads for outdoor and, you know, I only needed one and you don't need to keep serving.

13:44How do you know when someone has responded so you don't keep serving them the same marketing over and over? Yeah, it's a great example. And that example repeats itself in lots of different flavors all over the place. But this is when I talk about the importance and the paramount importance of data clouds. So when you have these siloed data pools, that's how you end up with an experience like that. We should know, a brand should know that you bought the Dart umbrella. In fact, all brands should know that, but that's a separate matter. How it works is when you have purchased that, the transaction data sits alongside the customer data in this case.

14:22So it will know that you purchased that product. And guess what? You will not receive that ad as quickly as the database can be updated up to and close to now real time right like we're literally real time we're talking about seconds so that's a very good example you'd want to suppress that for obvious reasons there's a whole bunch of other reasons you'd want to suppress things and you can do that in automated ways for compliance reasons right for all sorts of like privacy reasons like there's a whole bunch of things and it's really difficult and manual to do that now. And it doesn't need to be right.

14:57That's what we do. We basically allow you to basically have context. And that context is, hey, Craig's already bought the darn thing you're trying to sell him. So perhaps either leave him alone, or perhaps maybe talk about some other thing that you can do for Craig that would be of use to him. It really, I have all sorts of ideas in terms of what that kind of could look like, both form and function. But we'll get there in conversation if it's interesting to you. You've recently launched the Compound Marketing Engine. What is that? So it's interesting, right? So you mentioned the term composable CDP.

15:35This is an industry that's very crowded and there's lots of alphabet soup and there's lots of acronyms and so forth. And that's all well and good. In fact, GrowthLoop created what is considered the composable CDP category. And all that means is it sits on top of a data cloud. It's very flexible and composable you don't get locked in. What we're talking about with compound interest is kind of simple, right? If you think about the problem we're solving is the marketing cycles are simply too manual. Because they're too manual, they're way, way, way too slow. And they result in suboptimal experiences like what you just described as one example, but there are millions of examples where that is.

16:13So if you think about that, there's a better way, right? If you actually were to centralize your data and improve that data such that it could receive and participate with both marketers and agents, you apply agentic AI on top of that data, you could then have a much faster, more relevant context for the entire marketing loop. And if you think about the term compound interest, so I'm obsessed with the term compound interest going back to my days as an investor. We want to say, hey, what are the benefits of compound interest applied to marketing? So fundamentally, this is about how marketing happens.

16:54It can happen both better and faster to ultimately drive business outcomes. So in essence, that's what compound marketing is. It's about how do you drive growth faster by basically getting people through iteration cycles and experimentation better and smarter so that you learn faster as a business leader and a marketer. So that's really in essence what it is. And we're really pleased with the reception. People tend to get it right away, perhaps because of the term compound, which is a very powerful word in and of itself. Yeah. Can you walk us through an example that illustrates it for the listener?

17:33Yeah, we're really fortunate to have a number of customers who have been on the journey and continue to lean in. The most recent one I'll talk about is a company called Allegro. For those not familiar with Lego, they're basically Amazon in Europe. They're one of the largest marketplaces. They're a very successful company, very analytical in detail. They have been using our compound marketing engine to, and they actually announced something really interesting today. It's the first end-to-end agentic campaign that they've run. More specifically, they're applying Growth Loop, our compound marketing engine, to automate and apply AI to the selection of audiences, right?

18:16So remember I was talking about who you actually want to address at any one point in time. Historically, that was a very manual process for them, right? So they basically had humans going back and forth with ideas and SQL to say, well, we think we want to target this person. And that would take a lot of time. Right now they do it instantly, right? It sits on top of the data cloud, in this case, BigQuery, and basically serves up specific ideas for audiences to help. And what has that done for them? Well, they're using it in this case to power their retail media network. So that means there's a brand that sells goods in the Allegro marketplace.

18:53They want to get in front of agencies. They want to get in front of audiences, specific customers, and then find them wherever it would be most relevant in their journey, whether that's on their mobile phone, in an app, or whether that's in a paid advertising environment like TikTok or Facebook or whatever. They were up and running in a matter of weeks. And then over the course of two months have improved their return on ad spend by 2x. We're really increasing the amount of volume through their marketplace and doing so more efficiently. So that's one example that we're really proud of. And they've been very public about sharing these results and really think that they're one to watch because of the investments they've made in the data teams and agents at every step of their marketing cycle.

19:44So that's one example. I'm happy to share others if it'd be interesting to you. Yeah. Well, on that example, so Allegro, if they want to push, I don't know what the range of products they have, but if they want to push - Think of them as Amazon. I mean, they have everything, right? It's a huge marketplace. Yeah, a shoe, if they want to push a certain shoe, a sports shoe. They first, your growth loop first would find customers in their data, is that right, that are most likely to be open to a sports shoe purchase? and then the growth loop player would then design a campaign and decide on a vector or a channel to reach that customer, as you said, whether it be mobile or social or whatever, and then execute that.

20:50Am I on the right track? Yeah, you're certainly on the right track. Let's build on that a little bit. Okay, so not just any shoe, right? There's always different trends happening. Like I serve on the board of Gap with lots of different brands. So it's amazing to me how quickly different fads and trends come and go. So it would be informed to say it's not just any old shoe. It'd be this type of shoe. It could be a platform. Don't ask, don't quiz me on my shoe fashion knowledge. But suffice it to say, there's usually a trend or two that really is an important input. The other thing would be you'd have a data team that would do propensity modeling.

21:29They would say, hey, Craig's propensity to purchase shoes is either high or low. So maybe we should talk to say it's high as far as a particular type of shoe. Their propensity models, and this is ML, this is machine learning, not AI, it'd say they'd have a lot of models based upon all the previous transactions. to say, hey, we think that people that look like Craig have a high propensity to purchase this shoe. And then you automate all the other things you described. Okay, what might the offer be? Where should we talk to Craig, et cetera? You get the gist. And that's very much what we're working on with them.

22:07I'm really proud of what they've done. They've basically leaned into actually the actual creative itself. So they have a team that's using the underlying models. So Google launched VO3 last week, which is a video one, but there's no shortage of these different image generation bots, which are really darn good, like mind-blowingly good. So they're actually able to not only just take all that upstream intelligence on propensity, they're basically able to have an ad that basically they created themselves, not some third-party agency. And then there you go. So it'll be really fun to watch the type of results that they're going to continue to get.

22:43the early results were astounding really like 2x improvement for a company that is already really good is pretty pretty impressive so we're really really lucky to have them as customers do you augment uh customer data for your uh clients uh so let's say it's allegro and they have a a profile for each of their customers that includes everything they've ever bought. And I don't know what, but I had a conversation about a year ago with a company that uses AI to help political fundraisers target individuals. And it was amazing. They had built out, you know, I don't know how many columns in their data field, but everything from what magazines people subscribe to, to their zip code.

23:53So, you know, which gives some information about the political leanings and, you know, just all of this information. and then they could, the AI would sort through and identify people that was, you know, the ROI was high enough for them to spend time reaching out to. Do you help augment that customer data or is that a different part of the pipeline? Yes. Yes. Short answer is yes. Longer answer is we do it in a variety of different ways, typically with partners. If you think about these data clouds, there's a lot of enrichment. So just getting the data in both for the first time. And then what I'm talking about, and we talk about loops, it's feedback loops so that you actually provide feedback in so that you learn not just the first time, but as you continue to do things.

24:47But I'll give you some examples. So we partnered with companies like TransUnion to basically enrich a profile to understand more about a potential customer or an existing customer. Companies that do a lot of performance marketing use companies like LiveRamp or they'll use third party informations, again, to augment and get a holistic picture. The term, the buzzword is customer 360. You want to have a 360 degree view of your customer. So there's no shortage of people that offer the ability to do that. We've chosen not to lean specifically in the act of doing that for reasons I won't bore you with today.

25:26But suffice it to say, we help through partnerships to get people to ingest that information and get it in a centralized spot. We then kick in at the orchestration. So the identification and activation and orchestration of that data. But it's a very interesting field. And if you think about it, it's not just things like magazine. If you think about what's happening with Google News, Spotify, Netflix, these recommendation engines are getting so much better. And why is that? Well, it's actually not just old fashioned machine learning and reinforcement learning. It's actually artificial intelligence in combination with machine learning so that it's not just a magazine.

26:07It might be a tone. It might be a regional. It might be a political leaning. It might be a thousand different things that are really almost imperceptible to humans that constitute surface area to then experiment. And gosh, it explodes your brain to think about how companies like Walmart and Gap and others think about a product. A product isn't just a pair of jeans, right? And it used to be, it's as equivalent and maybe even more profound than the difference between Yahoo and Google. If you recall, Yahoo was this directory. It was like, it was a dropdown. It had to categorize something into a directory.

26:46Along comes Google, right? And doesn't have a directory at all. It basically understood signals by using backlinks in this case, but other things too to say, this is really what you want. That same thing is happening in all elements of business today. Think about a product again, a pair of jeans. It might be the cut of the jeans. It might be a vibe. It might be who they're connected to as an influencer. I mean, a million different things potentially. So that's really the fun here is to say this isn't just about random blue jeans and trying to say, hey, Craig needs jeans. No, no, no. It's much, much more nuanced than that.

27:23So it's about understanding that context. And I believe that each prospect and customer will have their own agent, too, to basically represent what you're trying to accomplish, both in commerce or entertainment or any walk of your life. So it's kind of a nutty thing to think about it when you stop and think about it for a minute. But it is exciting to me because it's going to mean more meaningful engagements, more personalized, because the context has really exploded, I think, in exciting ways. Yeah. And you mentioned early on about not being intrusive. And hopefully, marketing systems are becoming increasingly sophisticated through things like growth loop, where you're not just being bombarded with irrelevant marketing.

28:17because there's so much coming at you now. How do you manage that so you're not exhausting the customer or adding to the noise so that you're really delivering a message at the right time that's going to resonate? Build the future of multi-agent software with agency. That's A-G-N-T-C-Y. The agency is an open source collective building the internet of agents. It's a collaborative layer where agents can discover, connect, and work across frameworks. For developers, this means standardized agent discovery tools, seamless protocols for interagent communication, and modular components to compose and scale multi-agent workflows.