[00:00:00] Phil: imagine this like persistent [00:00:01] collective memory of all your past campaigns, hypotheses, experiments, results in the form of an agent, [00:00:09] Aboli: to be able to access somebody at five years back, who was like historical learning. That's the value, right? [00:00:15] Phil: this idea of like a living documentation architect that basically converts every decision, every workflow change, every meeting, every slack thread, into structured, browsable, auto updated documentation layer. [00:00:30] Aboli: this is so cool. I wish it exists [00:00:32] What is exciting to me though is the Egen browser stuff that is up and coming they just asked the agent browser to set up a campaign and it actually did a pretty good job of like drafting it. [00:00:43] ​ [00:01:10] In This Episode --- [00:01:10] Phil: What's up everyone? Today we have the pleasure of sitting down with a bully gwar, senior Director of Lifecycle and Product Marketing at Credible. In this episode, we explore self-healing, data quality agents, campaign, QA agents, a hivemind memory curator, and AI browsers that could power living documentation, all that, and a bunch more stuff after a super quick word from two of our awesome partners. [00:01:34] ​ [00:03:37] Phil: Aboli. Thank you so much for your time today. Really excited to chat. [00:03:41] Aboli: Yeah, likewise Phil. I know we are trying this for the third time, so third time's a cho. Super excited. I hope the internet works, uh, well today. [00:03:53] Phil: Third time's a charm. Um, I appreciate your patience on, on the tech side of things here. I, yeah, [00:04:00] probably overthink it sometimes, but, uh, yeah, really appreciate you bearing with me here. We got a fun episode for folks though. Uh, the, the upside of. Doing this for a third time is that we've improved, uh, some of the research on, on the episode side, um, every, every time we've redone this. [00:04:16] So, um, yeah, we're, we're chatting about the behind the scenes applications for AI agents today. A lot of the folks talk about growth and propensity models and personalization is fun. Some of those applications are, we're talking about behind the scenes operational stuff for marketing ops folks, like data quality things. [00:04:36] And we're gonna be covering and exploring a bunch of different agent use cases here. Um, I dunno how many, we're gonna have a chance to cover Aboli. So for folks listening are are watching on YouTube. We'll, um, in the description or the show notes, you can look at the chapters, jump up with a timestamp to a specific agent you're really curious about. [00:04:53] Um, [00:04:54] 1. Agentic Infrastructure Components in Marketing Operations --- [00:04:54] Phil: but I wanted to start Aboli by maybe just like, grounding ourselves a little bit in some def. Technicians, when we say agentic infrastructure, uh, what are some of the components here? Like, uh, we'll put up a table here for the folks watching on YouTube, but we've got like the data layer, the agent orchestration, the execution layer, feedback and learning, but there's also human in the loop sometimes. [00:05:15] What are your thoughts on, on these components? [00:05:19] Aboli: Yeah, definitely. I think. For any organization or team or function, right, to build out any sort of agent infrastructure, uh, where, you know, you, you'd expect the agents to autonomously, uh, work and talk to each other and, uh, you know, connect data with each other. I think there is sort of, you know, you just said it like the detail layer, right? [00:05:43] Where, um. Everything needs access to data, right? So the agent's gonna need access to data too. And is your data in one place, you know, is whether it's your data warehouse, the snowflake, Databricks, right? Like wherever that is, like, uh, that layer needs to [00:06:00] exist. Um, you know, as you're thinking about creating agents or connecting them to each other, uh, then there is the agent orchestration. [00:06:07] So how do these, you know, agents talk to each other? And, you know, my background is a lot on the lifecycle and product marketing. So, um, if I were to take a lifecycle marketing use case, uh, if I'm sending out an email campaign, right, like I could have a email copy agent, an email HTML coding agent, right? Um, how do the, how do I copy say, like, how does my copy agent intake. [00:06:34] Copy. You know, like if there's a Figma agent that build, builds out the email or like visualizes the email or designs the email and then, you know, there's the coding agent, then that quotes out the email. Like right now, I think where, um, where a lot of teams are are, you know, they're probably building out these agents, you know, individually, uh, like testing around it. [00:06:54] But then you need an orchestration layer eventually for these agents to sort of pass the copy to [00:07:00] Figma from Figma to the HTML billing. Interface, right? So that it's all like autonomously done. And, um, so I think at some point, right, that orchestration layer needs to exist. Um, then there is the execution layer, right? [00:07:13] Like each of this agent needs to like do the tasks somewhere. Um, and so that could be right, like in the copy agent, well, it could be like a, you know, just a custom GPT, right? Like executing within that in OpenAI, you know, chat GPT interface and like writing the copy based on brand guidelines and whatnot. [00:07:32] Or sometimes it could have to be connected with your marketing automation platform, right? Like, so that layer needs to be like, thought through as well. Um, and then finally there is, I guess not finally there is the two other components there is the feedback and learning, you know, how does then data pass in and back and from these agents, uh, to one another, right? [00:07:51] Like, how does like learning improve over time? And then finally there is the human in the loop, right? Like, how does the copy and make sure the [00:08:00] human or the marketers checked the copies on brand, like it makes sense, you know, it's legal approved and whatnot. And then from there, like they're sort of, okay. [00:08:08] And so, you know, how, how does like all of it work? I don't, I, I think, uh, I think we are in that evolution phase where I think people are playing around with like one or two agents and like seeing, hey, like, can an agent individually solve like one thing for me? But I think. As we think about like building sort of that agent infrastructure, like the connectivity of all of these things is like gonna be really important. [00:08:31] Um, I think the last two things I'll say is, you know, there are platforms out there like NA 10, uh, Zapier. I know several people on the podcast have spoken about it in the past too, where some of that connectivity, right, these tools are enabling, you know, that connectivity, uh, of, you know, if you need a slack connected to. [00:08:50] A Google sheet for your data analysis and like reporting, right? Like some of that orchestration could happen in this, in those tools. Um, and I know the MCP, [00:09:00] uh, model context protocol, right, like that is, you know, another mechanism potentially for tools to like talk to each other and like orchestrate. Um, so I think there is a lot of technology and tools, you know, out there that is like evolving pretty fast and something I think marketers should keep an eye out and figure out, you know, what fits best for their use case. [00:09:22] Phil: Love it. I appreciate the breakdown there. Um, yeah, maybe we can start diving into some of these. Uh, like I, I want to maybe say to start like, Aboli and I ne neither of us are like, I. Experts in, in, in AI agents, like we're, we're not building any of these tools. We're not, uh, developers or, or engineers in this space. [00:09:39] So the point of this is really like exploring what exists today, what could be used and, and what's useful, um, for marketing ops pros that want something that's a bit more behind the scenes. And maybe you haven't heard of some of these applications. [00:09:52] 2. Self Healing Data Quality Agents --- [00:09:52] Phil: So the first one that I wanted to explore, Aboli is autonomous. [00:09:57] Data quality agents. Uh, so, you know, [00:10:00] everyone talks about data quality 'cause like garbage in, garbage out. If you give AI crap, like it's gonna give you a polished version of crap. Like everyone understands the importance of data quality. There's a few data hygiene agents that I've seen pop up that claim to be. [00:10:14] This like thing that auto fixes 200 plus common data quality problems based on patterns that they've mapped from existing customers they have. So some of the product marketing, uh, copy, uh, and you'll appreciate some of this stuff here on, on the websites that I've, uh, researched. So like you plug in the agent solution to your warehouse or your existing data layer so you don't have to like, rebuild anything for these tools. [00:10:36] They kind of like. Meet you wherever you're kinda living. And the agents basically watch and check clean data. As data is changing, new records are coming in, uh, and they're on the execution layer as I understand it, meaning they like the, the work instantly and proactively. You don't need to like, approve everything. [00:10:54] Uh, maybe they're giving you reports after like a problem happened, but for the most part. The system keeps [00:11:00] improving as it has more data and learns from corrections. What are your thoughts here? Like have you seen some of these in action? What does that actually look like in terms of implementation? [00:11:11] Aboli: I haven't come across, uh, these tools. So thanks for exposing, uh, them to me. What, what I do realize the importance of those two. Those though. For example, more recently I was playing around a little bit with, uh, snowflake's copilot, uh, you know, which allows, uh, potentially like a person is not analyst to like prompt, hey, like, can you pull me a list of so and so users who meet so and so criteria and, um, I think. It was, it was really good in terms of like understanding, you know what I'm trying to like find out, but I think it was like looking at the wrong tables, the wrong fields. And I think that's where I feel like if it had the right metadata, right, the property dictionary or if I also had access to some of that, you know, with an internal company documentation, I think I'd be able to like [00:12:00] navigate that and like ask it to correct the tables it's looking at. [00:12:03] Right. So I feel like. There I could see a huge application of it. I feel like marketers, right? Like, or our ops teams too, right? Like will often rely on the BI analytics partners to sort of pull in like those ad hos SQL queries. But with copilot you could, you know, as long as it does have like the date, right? [00:12:23] Data quality, the right metadata. And so I think going back to the auto autonomous data quality agents, I think if there are tools out there that sort of, uh, help. Sort of cleanse the data, like, you know, it's like unique rows, unique users, right? Like all of that in one place. I could see like how valuable that is. [00:12:43] Phil: Yeah, it's, it's super interesting. So maybe some of the folks listening are just like, we're doing a lot of, if this and that, like work inside of DBT in our, our data warehouse to do some of this automatically. But like through rule-based workflows, like a lot of [00:13:00] this stuff exists. Uh, for, for a long time. [00:13:02] Um, but some of the, the websites, and we will, we'll shout out some of these tools here, but like, um, these agents seem to be operating by using pre mapped patterns. So similar to like, if you were doing this like as a workflow, um, there, like you can give it best practices and learned experience, but the. [00:13:21] Benefit of these tools is that like you're not the only customer using them. A bunch of other customers, maybe similar industries and diverse customer sources to basically keep the model learning or the tool learning and they can fix data problems that maybe you're not even thinking about, you don't have rules for yet. [00:13:38] And, and all this with like very little human intervention. So the features that they're calling it is like data issue detection. Auto-generated rules so you don't have to like create those rules. Auto resolution workflows. So sometimes there's like a conflict, you know, we see a record that's matching the ID resolution rule kicks in, but we don't know which record wins. [00:13:57] So there's like some work there. There's stuff like [00:14:00] self-healing pipelines. Anyways, there, there's a good overlap of this with like existing tools in the data observability space, and we've chatted about this on the show. Uh, we had a couple founders that, that are building in that space. Some of these companies are actually moving away from data observability as a term, and they're calling it agentic data management as a category. [00:14:20] So one of the examples is an Indian startup called Excel data.io. Uh, Gartner actually has, uh, a category on their site, a quadrant four augmented data quality solutions. We've got a big enterprise players in this space, like Informatica, Qlik, uh, but one of the leaders in, in that quadrant is DQ Labs. [00:14:39] They're an agentic AI data management platform. Um, so we'll, we'll drop a bunch of these like links in there. Folks wanna like check them out. We're, we're exploring, you know. Because obviously there's like tools that exist that claim to be able to do this. Um, there's a cool startup called Delphi that raised a seed round this year. [00:14:57] Uh, they're calling it autonomous data [00:15:00] quality agents, so they're delphi.io. They're a French startup. Um, I also found this company called Elation that has a suite of agents, so we'll chat about them for, uh, some of the other agents we're chatting about. But, um, they're like agentic data intelligence platform. [00:15:14] One of their agents is a data quality agent, so. A lot of people claim to be doing this. Um, I've never played around with it, so, um, I'm excited to get my hands on some of these. I got two consulting clients that are in this, like, exploratory phase with AI tools and so, um, these might be some, some cool use cases there. [00:15:33] So, uh, it, it exists. Like, we're not talking about something that like, oh, this would be cool, like some of these other, uh, agents that we're, we're chatting about later. [00:15:42] Aboli: No, I think that's like a great exposure, uh, to all the tools out there. I think that. As a marketer on the business side and like figuring out, you know, the holy AI landscape, one of the things I've been actively doing is just demoing tools, like listening to vendors. Like I think there are like, [00:16:00] uh, just like listening and like what is possible out there can like, make you think like you don't have to necessarily go and buy those tools, but there might be some, uh, adjust in solutions you could build in-house. [00:16:10] So that's what, you know, I keep telling. Everyone in this space, um, like, you know, look out for new tools. There's a lot of like, cool and exciting stuff out there. Um, I think, you know, to your point, Phil, like, and I've listened to like other podcasts guests, right? Like it all depends on the use case, right? [00:16:27] So yes, for people who perform like data quality is an issue. Um, I think this is like certainly worth looking into, but I, I also like [00:16:36] 3. Data Activation Agents --- [00:16:36] Aboli: when I talk to like other marketers fears in the space for marketers, right? Like. The more common use case that I've heard is sort of the data activation part, which might be a good segue into the data activation agent tech use case where, you know, quality may not be as big of an issue. [00:16:53] The data is, you know, out there pretty, pretty fine. Of course data's not perfect, but getting, having access to [00:17:00] that data and like actioning upon it, um, has been challenging. And I think that's where, um. You know, on the marketing automation platform side, we've sort of evolved from, um, you know, writing, building audiences, using SQL to like brag and drop filters to now, um, uh, uh, prong based potentially, right? [00:17:23] Like audience building with some of the newer tools out there. Um, and I think in terms of data access too, right there in the past have been here in the space for a decade. Like earlier, I used to remember like for every. Single field that I used to need to like personalize the emails right there is, I mean it still exists today, right? [00:17:41] Like there is a heavy reliance on the bi data engineering team to sort of make sure that calculated field, that attribute exists. Then like doing some work to like make it available in a format that fits the marketing automation tools, tools, format, and then, right, like there is some process of like getting that data in the marketing [00:18:00] automation tool. [00:18:00] And I think now we're seeing a lot of. Not just like, I think tools, I would say like general tools, without like the AI part, right? Like, which is making it easier to like activate data and like having marketers access that data, like easily and like maybe like one click import the field into the ESP. [00:18:17] Um, but then now there is a lot of, and that, that, that. I think what I have, from what I have seen that still like requires some bi engineering, you know, help in like figuring out the tables are mapped and like talking to each other, but now I'm like seeing a rise of now the AI agent tools, which like really make it like a promise base, Hey, like I need so and so field, can you like calculate it, create that feel, and then map it to my ESP. [00:18:44] So all of those like multi-steps are happening in one go, which is like super exciting as a marketer too. To be able to like, like access that data without that bi ticket and like getting it into our ecosystem. So, um, I'm super bullish on, on [00:19:00] that part 'cause uh, you know, we all talk about like personalization and I know we'll talk a little bit more about it. [00:19:06] Um, but again, for personalization too, we need that access to data like pretty fast. Um, so yeah. Curious, what have you, have you seen, like, do you feel bullish on that use case and what have you heard? [00:19:19] Phil: Yeah, data activation is really interesting. 'cause you know, it just, just a couple years ago, the whole category was just called reverse, ETL, like it's just recently just been kind of rebranded as data activation makes me, makes more sense. Like it's what we're kind of doing with reverse ETL tools in the first place. [00:19:35] But yeah, like you, like I've been a user on, um, the in-house side of this like latest wave of technology where. You know, we were using modern marketing friendly UI drag and draw builders, and it removed a lot of that SQL, uh, from the picture. Um, and so, you know, all these reverse detail tools that call themselves data activation, a lot of them are, are lumped into like bigger customer data platforms now. [00:19:59] [00:20:00] Um, but a lot of these tools, like you said, like still need analysts or engineers, like I, I was always. You know, I could always like stretch it, but I always got to a point where I was like, ah, like I, I still need that other thing out of the warehouse, or we need to calculate that field now. It's not perfect in the way that it is, or we need to like tweak the format of it before we can push it into our automation platform. [00:20:21] So you kinda tease this out a little bit, but these like new tools in the agen AI space are thinking about like using natural language to skip the SQL part and like. Altogether, data engineer, like, I'm still kind of skeptical about that part there, but some of the claims on, on the websites that I've kind of researched are like some of the features or the potential, um, use cases that are like, gather and understand data from different sources, analyze and process that data, figure out what needs to be done next, take action automatically based on the analysis that's been done from the agents. [00:20:55] So like activating marketing campaigns, generating reports for you, [00:21:00] updating customer segments for you automatically, because there's a new data field that came into the database and they do all this without constant human help or manual coding or, um, SQL like, uh, updates and, and your. They're following goals or instructions given to you in plain language. [00:21:16] So it sounds like theoretical in a way. I know there's some tools in the space doing it today. Some of the examples are, you know, wide variety of MarTech and data. Like, um, we just talked about data quality and I feel like that kind of fell into one category, data activation. Overlaps with a bunch of different, uh, areas, right? [00:21:36] Like at data warehouses and data lakes like Snowflake and Azure. And IBM Watson Snowflake has a data science agent. Um, you kind of talked about that. Like they're using NLP to democratize access to, to data. That's kind of the language around it. But yeah, like there's a lot of overlap with reverse CTL tools, composable CDP Growth Loop, for example. [00:21:56] We just had a. A couple episodes with folks in their team. [00:22:00] They're like, they have an audience agent and a journey agent that are automating some of this stuff. But I think that, you know, you, you named Zapier in your previous answer there. Like there's a lot of overlap with IPA vendors when it comes to activation. [00:22:13] Also, like trade AI or cado, Zapier, they fall in this category too. Like we had Trey on the show. They're completely pivoted away from, uh, visual builders to, or. They pivoted away from being just an IPAs to a visual builder to manage AI agents. So the space is changing a lot. And, um, I think that there's even like other categories, like content management platforms that are playing in this space, like, like a content stack. [00:22:40] Uh, they have like a real time data activation solution, but they're almost like a. Bigger enterprise, like DXP platform, but there's like product analytics tools like Amplitude that have data activation features, warehouse native capabilities. So yeah, it's, it's a weird one. Data activation. 'cause like it, there's so much overlap with all these tools. [00:22:59] Like [00:23:00] I wouldn't be surprised if in like a couple of years, like, what is this data activation category look like? It's just like, it's it's solution. It's not a category anymore and it's just lumped into a bunch of existing MarTech. [00:23:13] Aboli: Yeah, for sure. I mean, I think, um, the part that marketers do after they get access to the data is also now having, right, like, um, personalization agents or um, decisioning agents, right? And so I think that is sort of a question to be asked on. Yeah, like, do you spend time on like, you know, activating data one at a time or really focus on some of the personalization agents that, you know, you give one time access to like all your data and then like it figures out, right? [00:23:50] Like what's the, um, what's the like content mix and like subject lines, that combination, like what do you send? Um, [00:24:00] and so I do think, like right now, for the next one or two years, like, yeah, we should, like, it's. Saying like the use case specific, like where, where are you spending the most time on? Which is the biggest opportunity to like, bring a step function change either in operational efficiency or like revenue generation and like sort of uh, prove out, prove that out and go from there. [00:24:20] Um, but I think just to like close the loop on data activation, I will say one other trend that I've observed, um, there is another startup in the space called assortment. And there are like probably other players out there that I'm not exposed to, but you know, they, they do claim to sort of being able to like activate the data without, you know, potentially a data engineer. [00:24:45] But, uh, what's even more interesting to me is like many startups that I am talking to are offering to sort of coil some of these like, you know, journey agents or like, um, audience building agents. Right? Right. [00:25:00] Where, um, I think. From a business standpoint, there might be like a resource crunch internally, right on, Hey, am I gonna be able to get like engineer AI engineering resources to build out my specific use case? [00:25:13] And I think that that is where there is sort of like an like a conversation to have to be had about bill versus buy. But then a lot of this startups are like now approaching saying, Hey, like we, we don't have this agent, but we could co build this with you for your specific use case, which is like. Kind of exciting, but also scary. [00:25:30] 'cause like we are kind of the Guinea pig. Um, but I, I am like entertaining, uh, or like even at least discovering and exploring. 'cause you know, even if we might not end up building, like, at least, at least it gives us like exposure into like what I earlier said, like what's possible. So yeah, really, uh, really interesting out there. [00:25:50] Phil: Yeah, free, free. Shout out for, uh, Priya there, uh, at assortment, uh, we were chatting by email a little bit and he was kind of updating me on like, when they [00:26:00] started they were like almost competing with customer engagement platforms and uh, automation tools. Right. And you said that. They've really positioned themselves now to, like, we don't compete with ESPs and customer engagement tools anymore. [00:26:12] We're kind of like an add-on and we're, uh, the first AI data agent built for lifecycle marketers. So it's always fun like looking at the product marketing. Messaging for some of these folks. And, uh, I, I know you guys are friends, so may maybe you've given feedback on, on some of the messaging there, but, um, it's an interesting space for sure. [00:26:31] I think is is the right call to not necessarily go, go head to head against, like some of these bigger platforms. Um, the, the, the question mark I have there is like. What happens when some of the bigger players in the space just like roll out a data agent built for lifecycle marketers and it's already ready to go and the tool itself, and they're an add-on tool. [00:26:51] And so anyways, the, the space is, is really interesting activation in itself. [00:26:56] 4. Campaign QA Agents --- [00:26:56] Phil: Um, the other agent that we wanted to kind of explore [00:27:00] together is this idea of a. Quality assurance agent, and maybe we can lump in compliance agent here as well, because I think these kind of go hand in hand. And our, our friends on, uh, the legal side and the, the data governance side, we'll be happy with, um, the advancement in this space here. [00:27:17] Agent AI can revolutionize the campaign QA process specifically, so not just like qa. Uh, a bunch of different things, but if we chat about like the campaign quality assurance process, you know, they're automatically checking code copy links. If you're rendering, uh, right across different devices, like right now a lot of teams are doing manual reviews, a lot of cool tools like KN sponsor on the show that does a lot of work on like translation and gathering everyone together on one platform to approve templates. [00:27:47] And there's some really cool work on like QA agents that are validating and doing, uh, some of this work for you. Um, it, it acts as kind of like an intelligent copilot and, and gives marketers like realtime feedback and you can [00:28:00] reduce costs on like, some of the QA side of things. There's a lot of like non-AI solutions for this stuff, but like, do you think there's added value in an agent solution here that can help with, uh, campaign qa? [00:28:12] Aboli: Yeah. Yeah, absolutely. And I think this is something. That ev like anybody can build, uh, at least like a POC or MVP, right? Like, um, a custom GPTI know everyone knows about it, or like, I think Gemini, Gemini has gems or Claus projects, sort of the equivalent of it, but. I think like at the very basic, anybody should be able to sort of feed your QA checklist, you know, your campaign doc, your email proofs or an SMS push copy, whatever that is. [00:28:44] And like have it sort of do that initial review at least, like are your UTM parameters coded in properly or, uh, you know, is the. Because, you know, sometimes people will have like the brand guidelines and email creative guidelines and the padding [00:29:00] and all of that, right? Like, does it like all like work as expected? [00:29:03] Uh, so I think that like, that has been a low hanging fruit. I, or like in our workshops, some of the workshops like you've been doing on the lifecycle human side, we've been doing a lot of workshops around it and people have found it like a simple use case, a repetitive use case, low hanging fruit. Like people should definitely look at it. [00:29:23] I think where the next level of the QA part, like there is two components, I think, at least from a lifecycle marketing standpoint, right there, is the, we build all these like complex journeys. I don't think there is, uh, anything, at least like, at least the current marketing automation platforms that I've seen might have like a journey QA specific thing, right? [00:29:43] Like here's the branching logic and all of that working as, as expected. So like, would love to see more exploration there. Um. And then I think the, where I think there might be a lot of opportunities for like companies like Email and Acid and Litmus. Uh, again, this [00:30:00] is very email specific too, but again, today, like why do we check the rendering, right? [00:30:05] Of like emails and all these platforms. It's like, as a human, I visually check, Hey, is the padding off? But then technically these are just pixels and like, I just want two engineer, right? Like, I think all of those guidelines could be technically fed and like. Um, so I, I think like whatever the human eye looks for the AI could look for like it's math and like, uh, proportions and ratios. [00:30:28] Uh, so I would, I think there is like, 'cause today, right, like the most marketers will still send out tests to these platforms and check how it's like rendering across iPhone. And I think that's where it'll be cool to see like how some of those u QA use cases evolve there and save time. [00:30:43] Phil: Yeah, it's a classic anxiety written, uh, practice for email marketers. Like, how am I, how's my email gonna render? And, and Microsoft Outlook like. Damn. It works everywhere except for Outlook, freaking Outlook. [00:30:56] ​ [00:32:52] Phil: but yeah, [00:32:53] 5. Compliance Agents --- [00:32:53] Phil: I kinda related to this is the compliance agent space. So like QA agent, we mean like double checking stuff, like links and does everything working properly. [00:33:02] Uh, but on the compliance side, um, I, I know like you work in a highly regulated industry at credible, um, I had a stint at a startup that was in health tech. So, you know, it was my first taste of. PHI and PII and, and all that world and is quite different from just like B2B SaaS. Um, and, and a lot of those like highly regulated industries, there's a lot of companies that have like compliance reviews. [00:33:27] It's kind of like a major bottleneck, especially when you're like in growth marketing or lifecycle and you're trying to do something fun and exciting. Um, so this idea of like an agent compliance AI could help speed up some of this process. Like they can come in and pre-screen campaigns for risky language. [00:33:45] Uh, they help you enforce like legal brand guidelines. They flag improper data use, um, even maintain like audit trails for approval. So right now, when you're trying to get approval from. The legal team, like we wanna do this crazy campaign idea. Instead, [00:34:00] like there's like you trigger some of these agents to do some of that upfront research work for you, and when you're like presenting that for approval, maybe it's already done kind of behind the scenes by like embedding compliance rules directly into the campaign creation process. [00:34:14] And they get, like, you get instant feedback instead of like waiting for days or. For the legal person to get back to you. I dunno if you've seen like, tools in, in that space or, or what are your thoughts there? I only found one, like when I was researching, I'm sure there's a bunch of other ones, but there's this company, a startup called Clean Lab. [00:34:30] Uh, clean lab.ai. They have what they're calling like detection agents and remediation agents that are collaborating with your existing agents and they ensure safety and compliance. But I dunno, what are your thoughts on that space? [00:34:43] Aboli: Um, no, I think it's a huge opportunity for regulated industries like FinTech. Um, I here to write like. Uh, we, like you could always also try building out a custom GPT and like feed all like your [00:35:00] internal like advertising rules and, you know, all of that. Um, that's like one way, but there are like a lot of limitations on the custom GPT in terms of like the audit trails and like who has the accountability and, uh, so there are I think things that need to be figured out, but there is potentially like a MVP path to like at least getting. [00:35:21] You know, like with compliance stuff, it's either you can do it, you can't do it, or it's, eh, like a gray area. Like, uh, let's evaluate the risk courses. The, so I think the yes and no cases are like straightforward and don't, like, that's something like the GPT could potentially throw out, but the gray areas where then you need to go a little bit back and forth. [00:35:39] And so hypothetically, if there is like, if at least you get the yes and no use cases out the door, then like that could be a big win. I have, um, come across a company in the space that I spoke to recently, uh, called Hello warren.com. I dunno if you've heard of it. Um, the company's founded by a FinTech marketer [00:36:00] actually who comes from, uh, prior, you know, FinTech companies like Brex. [00:36:04] And she was in the product marketing space. Ha understands the issues and, uh, yeah, they've been trying to build out sort of compliance AI agent where, you know, uh, a marketer could go in, upload assets. Um, you know, they would give back feedback and then you could then kick it off to like your, uh, compliance team for like additional review or they have the audit trails, right? [00:36:27] Like, you know, all of that. I think the company's pretty new. I think again, and legal and compliance in itself, like yeah. Do you wanna have, have your audit trails out there, right? That is another question. So I think compliance agents aren't gonna be a little tricky, to be honest. Um, just given like the risk and like how much document, like how much exposure do you wanna have? [00:36:49] Um, you know, et cetera. Uh, but there are tools out there that like, we're like, I'm definitely exploring and yeah. What was one of them? [00:36:58] Phil: Yeah, we'll [00:37:00] link@tohellowarrant.com. Um, I'm just like reading this. Putting my like compliance shoes on. Warren's proprietary AI automates brand compliance, legal and partner reviews for FinTech real estate and insurance. And like, uh, I'm, and just like having a hard time picture these tools like convince. [00:37:18] The legal team to implement these. 'cause there's like a self preservation factor in all of these, right? Like, they're basically saying like, Hey, we will replace your legal team. Or at least like, we'll help speed up their lives. And, but as soon as you get into like that, like you need less people on the legal team angle to this, but the legal team needs to be your champion to implementing this. [00:37:39] Like, it's just this weird self preservation space, right? And you need someone that's really thinking innovatively who doesn't care about, you know. Well, creating systems and automations in place that like automates him, uh, himself or herself out of a job. Like those are the champions you need for some of these tools to be like, successfully implemented within companies, eh?[00:38:00] [00:38:00] Aboli: Yeah, but you know, Phil, so is true for, I feel like every AI use [00:38:04] Phil: Yeah. Yeah. That's [00:38:05] Aboli: like the whole, we're talking about marketing operations, AI agents, and if you build out a journey agent, you build out a Q agent, then you're probably like gonna lead fewer people, right? Or, but I, I think what, the way I view it is. [00:38:20] You know, businesses, you know, like good businesses always keep growing. There's always more to do. So, uh, the way I view it is like, instead of hiring like two more headcounts in the next two to three years, uh, I'm gonna, I'm not gonna need that headcount. I just, you know, I'm gonna be able to operate, you know, within the existing team and, uh, sort of become more efficient with within the existing team. [00:38:41] I think that's how, like I view it. Um. And I think if you don't do it, somebody else is gonna do it. Uh, right. Like, so it's Right. How do you stay ahead of the game? Like, how do you, um, sort of start thinking about this so that you could be, that, [00:39:00] you know, AI Orchestrator in the company, if you feel like your job is at risk. [00:39:03] Um, on the compliance age, uh, AI like side, I feel like the compliance team is always swamped with things like they just have so much to do versus, so like for the, um, at these companies that I've been at FinTech, yeah. I feel like they would love actually this kind of a solution, but it's just like, yeah, like the intricacies of audit trails and like, uh, you know, all of that. [00:39:27] So. [00:39:29] Phil: Yeah, we're, we're friends with compliance on, on every team. It's funny, it makes me think of, uh, the episode we had with the VP of Growth at Wealth. Simple like the, we did the, the teaser for his episode. But he, he said at one point, he is like, uh, he was talking about like crazy growth ideas and he is like, if legal likes your campaign or is instantly like giving you the thumbs up for it, like it's, it's not a good campaign. [00:39:50] Aboli: Yeah. [00:39:51] Phil: Like your goal should be to get legal to immediately say like, no, we can't do this. Like, if they say that, like, it's probably a good idea and you should push it. [00:39:59] 6. Hivemind Memory Curator --- [00:39:59] Phil: Um, [00:40:00] let's chat about this other one. So this is, uh, I, I don't know if this actually exists yet, so this is me kinda like thinking out loud here and curious to get your thoughts on this idea of like a hive mind memory, memory curator. [00:40:13] So when I worked in house, especially at WordPress, we had this like really cool, uh. Data team that was doing a lot of like in-house tooling for experimentation and like teaching and training. And so we had a bunch of like experiments going out all the time, but we lacked this like central repository for saving results of experiments and then like. [00:40:35] Doing stuff with those after, or just like feeding that into a machine somewhere that like keeps learning and like it shares that with other teams. So I've been like, um, meditating on this idea of just like, you know, lots of models do a really good job at taking input. Recommending an output and reinforcement learning obviously like helps with a lot of automation around, or optimization around this space. [00:40:57] But sometimes like operationalizing all [00:41:00] of those learnings from cross experiments and all of your systems and into like micro learnings from experiments is, is really hard. And I've like, I'm, maybe there's companies doing this, I'm curious to get your thoughts, but like imagine this like persistent. [00:41:13] Collective memory of all your past campaigns, hypotheses, experiments, decisions with rationale, with like empathy, emotional context, real outcomes, results in the form of an agent, and it helps you eliminate repeat mistakes. You're saving time by not. Doing experiments that you've already run. Um, you know, then you can make the argument of like, well, it's a different time. [00:41:37] Like maybe it's gonna get different results. But just like automating campaign debriefs, you're enhancing like cross team learning, drawing patterns from different teams within the org. I couldn't find a tailored made solution for this. Like most of my research led to like products like I Pass or Mine Studio is one that kinda came up. [00:41:55] But what are your thoughts there? Bowie? [00:41:58] Aboli: Um, yeah, I think that would [00:42:00] be a dream to have as a, a marketer, I think, to, to be able to access somebody at credible, a market at credible, like, I don't know, five years back, who was like historical learning. That's the value, right? Like why do you wanna retain employees is because. Obviously they have a lot of like historical knowledge, uh, about what works, what doesn't. [00:42:20] Um, well I think I've, uh, I'm excited about Atlassian's ro, uh, product that they're pushing on, which is supposed to be like, you know, like Confluence and Ev like is Jira is where like a lot of companies will operate, right? In terms of their tickets and bi tickets, test results. Um, and then like documenting like campaign documents. [00:42:44] Um, you know, et cetera. So I am actually, and then Google has like, you know, Gemini embedded within like Google Docs, like, so whatever, like, uh, or I don't, I haven't used Notion, but like, I'm sure like they might probably have a similar equivalent [00:43:00] of it, but I actually think some of this would be accelerated by the, you know, LLLM models behind these like Atlassians and like Google Suite of the world, where if we are. [00:43:12] As marketers storing and like documenting everything to the T in one place. You know, like one's doing it in Confluence, I was doing in Notion like, or one's doing it in Asana. Like, I think as long as it's in one place, like I am kind of bullish, um, I've looked at like some of the videos from Roho and seeing like what, like you could prompt it, you can prompt it to create more documentation. [00:43:35] Like I could see like where I could just ask like some of these AI like. I think ROA is supposed to be a chat agent too. Like I could ask ROA that, Hey, like how, like, I'm thinking of this idea, like, what can you find from existing documentation or like test results that could support or not support this idea. [00:43:54] Um, so yeah, I, but I haven't, apart from that, I, I [00:44:00] haven't seen like yeah. Any other tools? I think where a lot of tools are building the space is sort of those personalization agents, right? Like they're saying that, hey, like. We take this memory, store it, do our reinforcement learning, and then like personalize experiences for your users. [00:44:16] Um, obviously what that takes away is like a marketer's ability to access those learnings or like figuring out what work, what doesn't. Um, but I think I, in this world, in the next couple, like 1, 2, 3 years at least, I see a space for your like hive mind memory curator kind of a place, right? Because. There is, and especially larger companies, larger marketing teams, you know, a hundred, 200 people, marketing orgs. [00:44:43] Like I could see like, you know, somebody's testing something, an email and somebody's testing something in like Met ads and like, are they talking to each other? Like I could see like huge value in something like this. [00:44:54] Phil: Yeah, I feel like one of the counter arguments to this, I'm like talking myself out of it, uh, is this, like, [00:45:00] there's, there's a big trend of platforms that are basically saying instead of a human coming up with an experiment and then running the experiment test, like you should have a system of always on experiments, especially for email, like. [00:45:14] We have now access. Like there's some tools that folks I've chatted with on the show that are doing this already. Like there's no human coming up with like a segment of people and like a campaign idea and sending those out. They just have like a library of messages and content offers, and then a huge system in their data warehouse of people to message to. [00:45:34] And a model that like predicts the likelihood that someone will respond well to a certain message at a given time based on what they've done. And then the machine just like sends those messages out to those people and there's always experiments going out behind the scenes. So there's never a campaign getting launched to certain subset of people without there being a holdout. [00:45:53] So you're always comparing it to people not getting it. So you're always getting like incremental impact on everything that you're [00:46:00] doing. So you know, the counter argument there is like, what? What's the point of a hive mind with like all the results of experiments and it, it really just gives you data to like do another experiment. [00:46:09] Like what you should be doing instead is having always on experiments for all of your campaigns, at least like stuff that's digital over like some of the channels that you control. I don't know, like it feels super farfetched for some teams, especially in enterprise that are a bit slower. But I've chatted with some folks that are like using some of these products and they're just like, yeah, this, this is the future and the future is now for some of these companies. [00:46:33] Aboli: Yeah, I agree. I think there is definitely a big spike in sort of the personalization or decisioning agent companies, especially in the life sector. In marketing space. many companies are definitely pushing for that now. [00:46:47] Um, I, but I've also heard that. Some of these like tests can end up being so complex. Uh, you know, especially in like larger organizations where there might be another product [00:47:00] experiment like going on and then you're like running this experiment on top of it. And like the number of like variations that could be in just the room for error is like so high. [00:47:10] So I think, uh, I think there is still like, I think that it is de I'm definitely bullish on that, but I, I just. I haven't heard like, oh, super, like, awesome, like this is working kind of, it's like, I think it's still in that POC experimentation phase. Um, but love the idea of like always on, you know, experimentation. [00:47:29] Um, I think that's one of the, you know, like how we do experiments traditionally is right. You'll launch something, wait or stat sick, then like take a next bunch of creatives, put it through legal, then like re write, like launch. So I can see. Value, like something like always on, obviously like regulated industries. [00:47:49] My, when I go out and like explore these POCs, my concern and I'm sure the legal and compliance team would also be like, how do we know what variations and combinations of like [00:48:00] copy are we sending to each user? Are those combinations gonna be compliant? Like what if it says send something like that? Just doesn't make sense. [00:48:07] So I think, yeah, I think those are, you know, all the things that we'll need to figure out. [00:48:13] Phil: It's funny, we're talking about the experimentation space and like Stat Zig you just mentioned, like it's reminded me that Stat zig the company, the, like the AB testing tool was actually recently acquired by Open AI for like 1.1 billion and I was like trying to figure out why, like I was shocked by why LLM purchase. [00:48:34] Primarily like a product focus experimentation tool. And you know, the acquisition potentially was like strategic to bring their founder on as like open AI's technology officer of apps. Um, kinda reflecting open AI's ambition to like strengthen, you know, some of the areas around product engineering, enterprise ai. [00:48:56] So it may might be less of like the. We're gonna build something for [00:49:00] experimentation and more about like, we want some of the human capital behind that. But do you remember seeing that? Like, were you kind of surprised by that acquisition? [00:49:08] Aboli: I do remember seeing it on LinkedIn. I have, I was, I am not super close to like how stats sync works, but I think if it's a product experimentation tool, like I'm wondering if Open AI is. Um, I mean, I think, uh, as they grow as a company, right, they're probably figuring out like the specific use cases enterprise companies are going out and like buying and spending dollars on other companies. [00:49:34] And like, if they're ultimately using OpenAI LLM model in the backend, like can OpenAI offer that, uh, product? Uh, and it's Right, right. It's like. To some extent, I think there is gonna be a suite of products that OpenAI will offer. Like Google offers all these like edges and products. And so I, maybe OpenAI is trying to build on that, but now I'm like using custom GPTs for data analysis. [00:49:57] But maybe OpenAI builds out its [00:50:00] own. Um, and I think that's what we see in the MarTech space to write where the offer fed, raise an acquisition. It's uh, to your point, right, like, yeah, there might be these smaller companies building out these agents, but what if like a bigger. Uh, automation platform builds out an agent, but it is likely that these bigger companies could acquire some of these individual, like data agent or journey agent, right? [00:50:23] Like these kinds of, uh, companies. 'cause I think there's a lot of nuance and like building out each of these agents, right? When I was just exploring journey building agent potentially, uh, it, there's just so many nuances, right? Like if you're editing an existing journey. How do you like, uh, figure out the branching out logic? [00:50:42] Like it still need access to like, all the data to write, write the branching out logic, right? So there is, it's just like so specialized. So I could see like why like a bigger Salesforce might not be able to figure it out, but like, you know, smaller player might be. Um, [00:50:59] Phil: And then they [00:51:00] just get acquired bigger. Shark eats the, the little fish [00:51:03] Aboli: Yeah. [00:51:04] Phil: such as the, the cycle of, uh, of startups and, and MarTech we're, we're all familiar with that one. So yeah, there you go. Hivemind memory curator, free product idea, open AI stat. Zig, uh, let's, let's make it happen. Marketing op folks. So would love it. [00:51:19] Uh, we'll do one more bully. Um, I, I, [00:51:22] 7. AI Browsers Could Power Living Documentation --- [00:51:22] Phil: I think the bane of existence, aside from data quality and like. Getting sales and marketing to get aligned together. Marketing ops folks live and breathe documentation and, you know, nothing gets done without documentation first. Or, you know, someone leaves and you had like, uh, your whole MarTech stack built on top of one person and now you're just like, I wish we had invested more in, in documentation. [00:51:47] I'm curious to ask you about this idea of like a living documentation architect agent, or. Documentation AI agent, like whatever. I'm not a product marketer. Like have you seen agents that basically [00:52:00] convert every decision, every workflow change, every meeting, every slack thread, every like structured, browsable, auto updated documentation layer. [00:52:10] You basically have this like SOP, uh, the campaign playbooks, platform configs, like they're all staying accurate by default without. Anyone having to manually document or write anything or update anything, it's just kind of all living behind the scenes. Maybe more like ambient computing style. A little bit like, I don't know how this would work in practice, but this is a dream for marketing ops folks, like no more outdated confluence pages or tribal knowledge loss. [00:52:40] What are your thoughts there? [00:52:42] Aboli: Um, everything you explain see is like, oh, this is so cool. I wish it exists. Um, but no, I, I haven't seen, um, I think Doc, I agree, like documentation is a big use case and like no marketer like likes to spend time on it, but it's like, ah, [00:53:00] this is the last thing I wanna work on. Um, and so I think that I, I haven't seen anything, but I think like with where some of the Atlassian, like, you know, product build is going with and you all of that. [00:53:13] I think that could, I don't think it'll be like as visionary as you may, you've explained because of like how fragmented like our, uh, day-to-day like tech stack is, right? Like my Slack messages and the G uh, Google Suite is different. And then Atlassian and like, they're not connected, but you know, if everything was like, you know, some, and sometimes we use Zoom, but if everything was like within the Google ecosystem, like Google Docs, like I could see like how, you know, all of it's like talking to each other and Google can build out the product, but, but we don't use, we use Slack. [00:53:48] Right? Which is not a Google product. Right. So, um, that, that, that's why I'm like a little like, is that gonna be even possible? Um, but I think what's. [00:54:00] What is exciting to me though is the Egen browser stuff that is up and coming and I'm like, I feel like I question every single AI use case or like, you know, how it's done with, like, what I've seen with Egen browsers where, um, somebody had. [00:54:18] Demo. Uh, like they just opened up, like whatever, you know, B2B marketing platform. They used to like set up campaigns and they just asked the agent browser to set up a campaign and it actually did a pretty good job of like drafting it. And I was like, okay, like what happens to the journey agent? Like, or, you know, if the agent browser is able to pick up all of these things with like very little context, like, and that's just like, this is not even year one of agent browsers. [00:54:44] Like where does it all go? So the, the living point being like the living documentation architect, right? Like if it, everything is still happening within the browser ecosystem where, you know, have the slack and like, even if you're like in fragment, like I could see the browser playing that [00:55:00] role. Um, but no, I haven't, I haven't seen anything quite to that level. [00:55:06] Uh, is this in the market yet? [00:55:09] Phil: Yeah, the browser stuff is, is really exciting. It reminds me like, 'cause you had a great point about like the, the biggest obstacle for a documentation agent is just like how many tools, how many fragmented workflows and systems are part of our day-to-day. And like auto logging changes from all of those day to day. [00:55:29] Like it's just. Not feasible unless it lives like on a different layer than on the tool layer. Um, our first episode this year was with, uh, Austin Hay. And, um, he's building a CRM today, but he used to, like, he created the, the MarTech course at, at Reforge, like he's the visionary on the space. And he talked a lot about this idea of ambient computing as kind of the. [00:55:52] Future of agents, like right now, a lot of agents are workflow based, goal based, and they're specific to 1, 2, 3, [00:56:00] or a couple tools. And with MCP, they kind of tie into some other stuff. But he talked about this idea of like ambient computing where something that just like intuitively knows everything that you do because it has access to like the operating system level. [00:56:15] So we don't have to like worry about connecting to Slack and connecting to Google Calendar and Asana. 'cause it's connected to your os and so like it's, it's already, I don't know if tracking you is the right word there, but I don't know. It kinda works by like embedding AI directly in your OS and giving it realtime access to like all of your files, all of your apps, all of your workflows. [00:56:36] And it gets like, to understand the context, it can automate tasks across different tools. We're getting a bit more like speculative future here, but I don't know, like some folks immediately are just like. Giving an agent access to my os, like forget about it. Like I'm using that work computer to log into my personal finances. [00:56:55] Like I don't want an agent to have access to that so that it gets kind of [00:57:00] dicey really quickly. I. [00:57:01] Aboli: Oh yeah. I think I would, I think the amount of people who are like skeptical about, it's a, it's a big number. It's not like one out of 10. I think maybe three to four people out of 10 are like worried about, and, uh, I would be too, in general, like I think I am generally okay with, I don't know, uploading my lab results at a chat giti. [00:57:21] But I think the moment that it like starts. Like, yeah, like my personal finance, right? Like, you know, all of that. I start, I am also like, what it, like, I wouldn't even wanna try the tool. Like sometimes I'm like, oh, should I even try it? And like, what is it? Get access, right? I think all of this is so new. Um, but yeah, super excited on the AI browser. [00:57:42] I'm keeping an eye out on it and, uh, seeing, you know, how it evolves. [00:57:47] Phil: Awesome. A is super fun. Uh, I feel like it's a long time in the making. I'm glad we, uh, we got through it today. Uh, hopefully folks found a new list of tools or potential use cases, or some of these ones I think are a bit [00:58:00] more exploratory, uh, especially the last couple ones there. But [00:58:03] 8. How to Stay Balanced as a Marketing Leader --- [00:58:03] Phil: I got one last question for you. [00:58:04] Bully. Um, we're, you're obviously a lifecycle marketer, product marketing leader. You're also a newsletter author, community builder, keynote speaker, and, uh, at home you're in, uh, a mom of, sorry. Um, at home you're also. I dunno why I had written it like that. It's, it says you're also an avid mom of an 18 month old. [00:58:28] You're also a mom of an 18 month old and you're a jogger or a yoga, uh, you do yoga, you're hiking. Can you tell, I've recorded two episodes back to back and we're at the end here, but one question we ask everyone on the show is, how do you remain happy and successful in your career? And how do you find balance between all the stuff you're working on while staying happy? [00:58:48] Aboli: um. I think for me it, um, like relationships and like connection, like friends, family [00:59:00] connections, right? All of that like holds a very important, like everything I do, like I go back to like why I am doing it before my family is spending more time with my, with them. And I think that's what like really keeps me going is yeah, like why am I earning money? [00:59:14] It's so that, you know, I have a, a quality time, you know, like. Uh, good quality of life with, with my family being able, like I am, my parents are in India, so like, yeah, can I make more trips? Go back home, right? Like, um, that's really, I think that that's what keeps me going. I do think like, like the yoga, the, all of that like actually helps balances it a little bit, right? [00:59:37] When I'm like too stressed, I'll just go out for a slow jog. Like, I think that, I'm sure everyone knows how important like exercising and all of that is, and why it's important for your brain. Um, but yeah. [00:59:51] Phil: I love it. Yeah. Great. Uh, great reminder to get outside, touch some grass, get some sunlight. I think it's easier to said, said than done. Like, [01:00:00] especially, you know, like when you work from home. Home. Like it, it's, it's easy to forget the, the importance of going outside, moving around. Like, I, I spent like the last like five, six years working remotely and I just like got so unhealthy by like, working on the couch, like barely moving around except for like dropping my daughter off at daycare in the morning and like, you know, health just like caught up to a point where I was just like, all right, like I need to do something about this. [01:00:28] Like, I'm in my mid. Thirties and my body feels like I'm in like my mid fifties. And so now I've like really taken that seriously. And now like I walk, like I lift weights now I eat so much healthier and like, I'm just like so much more productive at work and actually look forward to like going on my walks and listening to podcasts and getting my dog to also move around. [01:00:49] So yeah, it's a, it's a good reminder. [01:00:51] Aboli: actually, I was gonna say, dogs, like it's, they push you out of the house. They're like, yeah, you gotta walk them. I mean kids too, right? Like I have to [01:01:00] take, uh, my son to the park. Like I meet my, now I have like parents friends at the park. Like the social connections part is so useful. Like to just get out of your, like, corporate nine to five and talk to other people. [01:01:13] Um, so yeah, it's, it's all good. Hope people find time to do that [01:01:18] too. [01:01:19] Phil: Yeah, talk to other humans, humans and MarTech. It's what I do full time now talking to other humans. Appreciate your timer bully. This says you were fun.