[00:00:00] Phil: you actually made the pitch to your CEO at SEMrush to leave your 35 person brand team to build a marketing and AI ops team. [00:00:07] Olga: There's consumer AI and those chatbots, and you, you request something, does tasks for you, but there's no material shift, [00:00:15] this B2B ai, this is what replaces jobs and humans. It's better to build systems that anybody could use that is built on API, so it would be centralized. [00:00:26] it's all starts with, this huge writing. automation so people would fill in the form, then it like searches this big database, where we have knowledge base, blog articles, we have USBs, we have, transcripts of, calls, every wow moments about our product then it would produce the content. and the Slack bot would take that copy to that channel. And then someone will be editing the document and then they would click finish and only then the person would receive the document. [00:00:57] ​ [00:01:24] In This Episode --- [00:01:24] Phil: What's up everyone? Today we have the pleasure of sitting down with Olga Andrienko, former VP of Marketing Ops at SEMrush. Olga spent the last 12 years at SEMrush. She went from leading social media, global marketing, then became VP of brand marketing, and at the time of recording, she was building a new marketing and AI ops team. [00:01:41] But she recently announced that she was leaving SEMrush to go out on her own to build a marketing SaaS and advise others on how to build AI agents. In this episode, Olga shares agent examples. She built out SEMrush, including a huge content automation that includes a context database, She shares how to beat AI imposter syndrome, and how to prioritize internal AI [00:02:00] projects. We talk about AI's end game and speculate about a future of play to earn in mandatory human quotas. But most importantly, Olga shares how to optimize your body like a MarTech stack. All that and a bunch more stuff after a super quick word from one of our awesome partners. [00:02:14] ​ [00:03:21] Phil: Olga, thank you so much for your time today. Really excited to chat. [00:03:24] Olga: Thank you so much for having me here. I'm very excited. [00:03:28] Phil: Many folks that we chat with are in marketing ops and a, a lot of folks outside of marketing ops are predicting that marketing revenue operations folks are really well positioned in this. Like I. Transformation of roles with AI right now. Ops roles are already like heavily about finding ways to use AI and tools to automate and speed up internal tasks and increase the productivity of people inside the company and training them on it. [00:03:55] 1. How AI Agents Reshape Marketing Ops Roles --- [00:03:55] Phil: You're so excited about marketing Ops folks becoming this new like resident [00:04:00] AI tech implementation experts, that you actually made the pitch to your CMO and your CEO at SEMrush to leave your 35 person brand team to build a marketing and AI ops team. So maybe we can start by asking you like, what was that triggering factor? [00:04:16] Probably a few different things, but maybe chat about like that last touch moment that made you go like, all right, I'm leaving brand. I'm going all in on marketing ops. Chat to us about that. [00:04:24] Olga: Yeah, so, uh, this was this year. And, uh, the absolute trigger was, um, me attending the, uh, two day weekend course on AI agents. And, uh, before I was experimenting, I was, well, I was working in chatbots a lot and I was thinking I'm very quite advanced with prompting. And I used, well, I was very proficient with Claude and projects and, uh, well, the chat tea and Claude like perplexed, I was like on all chatbots. [00:04:58] Uh, and I was not [00:05:00] really a fan of Gemini at that point. Now I am, but um, and then I joined a course and I thought, okay, well this is some additional information. Um, but, and then it wasn't an eye-opener because not how I differentiate ai. There's consumer AI and those chatbots, and then they will make you work. Better because it will have, like, it'll work with you, uh, on, and you, you request something, it, it does tasks for you, but there's no material shift, [00:05:33] uh, with chatbots. And then this B2B ai, as I call it, this is a agent ai. Um, and this is, this is how, well, this is what replaces actually jobs and humans. Uh, not yet, but uh, this is what can radically transform and improve systems and at scale. So then [00:06:00] I had like a huge reflection moment, and then I instantly went into like this panic mode that we don't have it yet in marketing, and then just we need to, we need to roll it out and then we are already behind. So it was like, and um, when I think that like, we are a very tech savvy company. We have, like, our tech stack is incredible. [00:06:20] You can just come in and then you can play with all sorts of tools, but then it's like our, my assessment of us being behind is still quite well. We are like probably, well in, like in 5% of the companies that are very, very advanced in, uh, in the tech space, but. It's still like we are already like, well, every company now is behind, I think, [00:06:43] uh, because the, the, the speed of this, I think it's just every, like, every two months is 10 x of [00:06:48] what it was before. [00:06:50] So that was like actual moment when I thought, okay, like this is where something's need. Like something needs to change. We are a big company. Uh, that means [00:07:00] that, uh, I, there needs to be someone with authority, an internal authority to make the change, like, well, to make it fast. And, uh, so this is where I thought that I could be, like, I'm interested in that. [00:07:13] I spent all of my weekends playing with [00:07:15] ai, uh, and then I'm really good at unlearning things and learning like new stuff and I have authority and I know how to like navigate within the company. So. And actually this was a week where I was doing the quarterly business review with the CMO and CEO. So it was, and then I sent them like a, uh, day, day, day before I sent them an alternative presentation when I called it. [00:07:44] And this is where I, well, entech AI radical change. Uh, also another trigger was, uh, the, the, um, the cost of, um, subscriptions for chat chip tea and Claude that I [00:08:00] kept seeing [00:08:01] Phil: Mm-hmm. [00:08:01] Olga: company. And I'm like, okay, this adoption speed is crazy. And also I was thinking, okay, if we give, get everybody an account, that's, that becomes like an insane spend. And instead of this, it's better to build systems and just, well, automation that anybody could use that is built on API, uh, rather than just everyone having an access, which I would like, we would control in a bit, so it would be centralized. Uh, so I just, yeah, I calculated the annual spend and then I thought, okay, like, well, agent AI is, is something that we need to enroll in. [00:08:43] So there was like two and together and off I went with [00:08:48] Phil: Very cool that I want to like, um, un unpack that, that first one a little bit more. Like [00:08:53] 2. How To Beat AI Imposter Syndrome And Start Using Custom GPTs --- [00:08:53] Phil: you mentioned the pace of change and how you thought at some point that like you were experimenting enough and knowledgeable enough to make that leap and, and make that suggestion to, uh, the senior management team. I think like this idea of imposter syndrome for AI is, is really real. [00:09:11] Like, I think there's a lot of folks in similar shoes to you where they see this future transition. They want to like take that leap, but they don't feel a hundred percent comfortable yet because they don't know if they're like knowledgeable in AI to like chat about AI ops or, or go into that stage. [00:09:28] Like a lot of folks are just like, ah, the engineering team is gonna figure that out. Like the developers are, are [00:09:34] gonna like, do that. I'm a marketer. I, I don't know enough about ai, but I. Like you were in the brand team and you were playing around with the tools and, and you said like learning and experimenting during the weekends. [00:09:46] Like what advice do you have for folks that want to take that similar leap or like want to make that similar bet then they're just like unsure if they're like knowledgeable enough about ai. [00:09:58] Olga: First of all, anybody who [00:10:00] says they have figured AI or they like, they're on top of this, like they are lying to you. I don't know anybody who doesn't have the fomo. [00:10:08] Phil: Yeah. E even Sam Altman, [00:10:09] Olga: yeah, it just, I, I think like he has a lot more gray hair than he had like two years ago. Like, he just crazy. You can see the interviews like, uh, you, you can see this guy doesn't sleep well. [00:10:21] So, um, and I like that. That's what I have every day. Uh, but I think the, the key is the focus. So this is like the main word that you need with ai. Um, because like you need to figure out, like you need to tackle it, use case by use case. And uh, just if you go like, it's just very easy to get distracted. [00:10:47] And, um, so take one. [00:10:51] Well just what I identify what I spend most time on. So if we look at this as a malow pyramid, like the routine [00:11:00] tasks and the repetitive tasks are the baseline. So you just see and analyze what, like what is the time suck and, uh, that's like, that's what you tackle head on and figure out how maybe you build custom GPTs, which are like a lot easier to build, or you build the project in CLO and then you still have this, um. You like, you start with the like basic stuff. If, if, and it actually also will set, will save a lot of time for the teams, uh, custom GPT if there's a repetitive task. If you are, well, if, uh, you have like a vague text and then you just need even to structure the text and then you do it every time. And that's where you could just build a very easy custom GPT if you are localizing text, well, it's not that great yet. [00:11:50] You still need the human editor, but, uh, you could still say how custom GPT with your instructions. Well, and then you translate [00:12:00] everything to, well, to speak from English to Spanish. And you don't need to do this every time. They, they can already be, uh, the instructions or even, uh, the prompt engineering. [00:12:11] You can grade custom GPT on how to create the perfect prompt. That would also save you a lot of time typing. So these are the simple things that like, that you could start with and that would, that would save you some time at least like one hour a day. And that's already a lot that, well, you could allocate this time somewhere else. [00:12:32] So if this is like, well ground zero a, uh, knowledge, then it would be just, uh, those, uh, custom GPTs, uh, or creating projects where the AI would have like a lot of context and it would keep that context and, uh, produce texts that are better or just, yeah, do the repetitive tasks that you have in an instruction. [00:12:55] So, and then you already start learning more.[00:13:00] [00:13:01] Phil: Yeah. I love your approach about taking it use case by use case. Um, a a lot of folks like think about too far ahead and they're trying to like run before they can crawl first and, and making the use cases a bit more basic. To start off with, I. Um, but I want to chat with about use cases because like you, when you were making this transition, you kinda like shared on LinkedIn part of your slide deck that you actually shared with your CEO and, and CMO to make that pitch for, for the new role change. [00:13:28] 3. How AI Content Agents Generate Drafts Using Internal Context --- [00:13:28] Phil: The first use case that you had in that slide deck was all about content. And this idea of building AI agents to help content creation by uploading a bunch of context, like you just said about own content products, unique selling propositions, competitors, uh, and then giving it instructions maybe through custom gpt or other ways of just like, um, I. [00:13:49] You know, tone of voice guidelines, templates, best practices. So this idea that you had kinda mapped out in, in that slide deck still had like a human [00:14:00] triggering [00:14:00] the action or, or the request, right? Um, a lot of the hype around agent is fully autonomous Agents who operate behind the scenes do the stuff for you without instructions. [00:14:10] So instead of like having a team member ask in that custom GPT, what content should I create next? And then ask it to like, help me write this case study. You would log into Slack and you'd have 25 new articles that morning, ready to go, ready to review based on events last night based on NPS scores that came in potential testimonials. [00:14:30] And you just have to like review which ones to like go live or not. Talk to us about that agent content opportunity and, and how that's going so far. [00:14:40] Olga: So ideally and in the future that could run on its own. Uh, I still think that it all depends on the quality of the output, uh, that you want. And, uh, really one great, great thing about ai, um, actually is that you instantly [00:15:00] understand, uh, what the, the definition of good enough for any person. So if you see the content and if person published that content, then you kind of assume that the person read this and then thought, okay, this is good enough. So the level of quality that everyone requires is very different. That's why you see like very, very bad content. That's because someone read this and thought, okay, this is good enough, and that's their level. And the higher a level is, then the more human in the loop you want. There's still like a lot of people that already are happy with the, um, AI output, they would just not check it. [00:15:41] And that's where I think one very important, um, skill, uh, that you need is a critical thinking, uh, because well, AI is hallucinating and then well, the more context you give it, and if you give it the usps, [00:16:00] if you give it the, um, information about you versus the competitors, if you give it the blog post, the knowledge base, like everything about your product, it's a lot of context. [00:16:09] The more context you give it, the more it has to hallucinates. So with some small information, well it could produce something, but if you give that like customer calls, system modules and constantly feed it more and more content, then the more checks you need to ensure that it's actually right. So. We are, um, I divided this in two steps. [00:16:31] Uh, so what's already built, and it's in a testing mode, is the quick fix, and that's where people write content themselves. And then they add the add-on that has the guidelines and it has the, uh, instructions. And we also analyze the good and bad emails, good and bad landing pages. Uh, so it has the comparison and analysis of what good is and what bad is and bad is based on the sent [00:17:00] emails and the conversion rates, uh, [00:17:02] from those emails. [00:17:03] Plus it's also then like our email team's judgment on like good and bad, in especially in the copy. And the landing page is also based on the conversion rates, uh, from the, from actual performance. And, uh, so that system like works with now, like different API keys, well, like Claude has been like the best, um, that we've tested so far. So you have a document, you enable this add-on, and then it scans the, uh, text and it gives you the, uh, summary and the verdicts, like, needs a lot of work and then explains why. And, uh, in the first situation, it was rewriting the sentences and suggesting this, but then we figured out that it's only as good as the initial piece because [00:17:54] it, not delete the sentence, but like maybe sentence should not be there. [00:17:58] So it's like if you give it [00:18:00] shit, it will give, will give you, like, I hope I, I can say this in a point. [00:18:04] Phil: Yeah. [00:18:05] Olga: And so it will give you like a bit better version of the same, uh, but not anything that is like radically, um, high, well higher in quality. So that's why it's better to still educate the human, like, this is what's not great, this is what's great. [00:18:22] And then have them still think, because like ai, just like very quickly as if like, well, with the Google Maps, people don't think where they're going. It's like the same with ai. If AI writes for you, you kind of like stop thinking that, uh, eventually would be like a huge problem. I think. So that's like where we are at this point where there's an add-on for Google Docs, it's in testing mode and it'll probably released by the end of um, uh, next week. [00:18:54] Phil: Very cool. [00:18:55] Olga: and, uh, so second is this huge writing. [00:19:00] Uh. Um, automation and it's all starts with, uh, a Google form. Like not a Google form, but NA 10 form, where now the enterprise clients of NA 10, like as the, like most popular, not affiliated, like paying a lot for it, but like, it's one of the, uh, tools that, uh, enables like a lot of, like, it's just a connector. [00:19:27] And, uh, so, uh, it has a form. So people would fill it, fill in the form, then it goes, and then like searches this big database, uh, where we have knowledge base, um, blog articles, we have USBs, we have, uh, uh, transcripts of, uh, calls, like, well anonymized, but like anything, every wow moments about our product and how, like what, what people like. Um, and uh, also jobs to be done, like a lot of like product marketing stuff.[00:20:00] [00:20:00] So then it would produce the content. And that's where the tough part is that now I am finding and vetting the content writers who would be overqualified content writers. So I'm looking for heads of marketing, heads of growth. And uh, so I would have them on the Slack channel and the Slack bot would take that copy that was written with our guidelines, everything to that channel. And then someone from those people will take that and, uh, then they will have like, well, they will be editing the document and then they would click finish and only then the person would re re receive the document. [00:20:42] So, um, because I, I, I still think it's, it's good. Like we can, we can constantly tweak it better. And of course, AI, uh, system would be training on edits that, um, our, my editors would be doing, but. It's like, it's still, [00:21:00] um, this is the sea of sameness. Uh, and that's where I just, I maybe I wanna joke somewhere. [00:21:06] Maybe it just, it needs to, like, it needs to be less Well, well content dance. It just, [00:21:13] Phil: Yeah. Yeah. [00:21:14] Olga: just, people needs to, like, some person needs to judge this and edit this. [00:21:19] Phil: Very cool. So there's like two main, I guess, like systems that you have in place there. The first one is a human writing, the first draft, and. You have this add-on where you're getting feedback on where to improve that writing and, and that's ai. And then the second system is kind of flipped. You have, uh, AI writing the first draft based on a lot of context and information that you're giving it. [00:21:46] And then that's going to a subject matter expert and that person is tweaking AI's first draft. So as you're like talking about this, like we, we want to chat about like human in the loop, but like the next version of this, I'm al I'm also, [00:22:00] I'm almost seeing like the two systems at some point just removing the human from that loop. [00:22:06] Like if you have one system that is coming up with that first draft and it's getting to a point that it's really good and the other system is editing that first draft and it's getting really good, like. At what point do you think that, you know, we don't need the human in the loop anymore. Like, how do you balance that manual effort currently involved today to train the system and, and where that's kind of going? [00:22:28] Is that, is that where you're [00:22:29] kind of [00:22:30] Olga: Yeah. In, in, in a year, like in less than a year. [00:22:33] Phil: Hmm [00:22:34] Olga: Well, uh, but it also just, um, well, AI, sort of like everyone said that there's a, well, great equalizer. It's not an equalizer at all. [00:22:48] It's, uh, senior people are using that really, like, well, uh, junior people are already falling behind, uh, because no one really wants to train the [00:23:00] juniors like [00:23:00] everyone wants the seniors. Uh, this is also like a huge, um, uh, well, discussion point. Uh, because, well, I think that we will have a lot of new. Industries. And that's like, I'm, I'm talking like I'm taking a off track. [00:23:17] Phil: Don't do [00:23:18] Olga: Yeah. Uh, but yeah, there will be huge, like, there will be a lot of new industries, uh, formed by people who would not be able to find their place, like in the current work [00:23:30] Phil: Mm-hmm. [00:23:31] Olga: universe. [00:23:32] Um, because like no one wants to train them. AI does it better. So what they do, like if we have millions of people are just like hanging out well together and then think, well they have their needs and this will, this will be a good moment where like a lot of new economies will be formed, uh, with like millions of people there. And that's where I would feel left behind again [00:23:57] because with my like advanced AI knowledge, I would not [00:24:00] know anything that was happening in like in another universe. Uh. [00:24:06] Phil: So me, me, the folks listening right now are just like, I would rather not be one of those humans that needs to like, adapt to a whole new workplace and, and still try to be the person involved in the AI ops, which is kinda why you are talking about that triggering factor and, and, and made that transition. [00:24:24] Um, maybe we can chat about like some of the other use cases that, that you had on your list there. [00:24:29] 4. How to Use a Risk and Reward Grid to Prioritize AI Projects --- [00:24:29] Phil: You actually posted in, in your slide deck you had a really cool matrix of how to classify AI use cases, uh, based on like risk and reward. Um, you had a bunch of different ones in there on top of content creation, like ad campaigns, recruitment pipeline calls with customers, reporting a, a bunch of stuff. [00:24:48] So like, talk to us about what went into that classification system and, and has that evolved a little bit since you posted it? [00:24:57] Olga: It. Yeah. So it has [00:25:00] not evolved yet because I'm not, well, I, I. I like to start, uh, with few ca use cases, but if you start small and then you build something quick and then you gain trust a lot faster, then if you start with something major and then it's still like not done within a few months. So, uh, so I, I hired, uh, AI Workflow Architect, uh, and then it was like his first month. [00:25:32] First month we already built like three automations within that month. We built one for influencers team, where we just, uh, we tracked server voice from a certain number of influences. Uh, and then with LinkedIn is like very, very difficult. And LinkedIn just hates all sort of scraping. [00:25:52] But like with, uh, well, Twitter is way better with this. Um, so we, we, we track, we just [00:26:00] now automate, uh, like we track mentions from that list of influences. We track mentions of competitors, and then AI just like puts everything in the file. Uh, and then it just, what we already saved around like two hours, um, of, uh, of one team member's time. Like she was doing this like every, every month. [00:26:21] And that's manual work. And then there's like 40 influences and then seven competitors. Like that's just like, that's a part of a job. Like you, she obviously hates because this is just like coffee pasting, calculating the number mentioned, like just not, not, not enjoyable. So yeah, it just, I, uh, that that's already done. Uh, one other, um, so Mark, well, marketing analytics, which is like also part of, um, ops in like in a lot of companies. Uh, this is the like, toughest thing to, um, automate. Um, and especially for us, like we are a public company. Like, uh, my, [00:27:00] uh, the, the teams that are not a big fan of me right now, the like, procurement, legal, um, and security team, because I was just like, I want all the tools. [00:27:12] Uh, and, uh, yeah, just, uh, some use cases just did not, not, um, possible for us. Uh, while for scale ups and startups, like startups are just like, they're the, well, I would say scale ups are the winners, uh, because they're already big and they have budgets and then, but they're not, like, they have, like, their risk tolerance is good. [00:27:35] Like [00:27:35] we just, uh, uh, we can't put any data like, well that is the sensitive, like an open AI or philanthropic, [00:27:43] Phil: Mm-hmm. [00:27:45] Olga: But well scale up would have like, uh, definitely, well, just any private company also would have a bit of a, a different solution to that. Uh, startups, um, they can be very, very risk, uh, like they risk appetites, they're really [00:28:00] good, but, and then they don't have a brand they can damage. [00:28:03] Like, it just don't, don't have anything. So they can outreach like anybody, they can just like be very noisy. And then some of it, like if you throw a lot of shit on, like against the wall, some of it sticks. Like [00:28:14] that's what, like startups and AI situation right now. Um, so, but like my case is that I am, well, I work in like for big public company and uh, we, we still want to save our world time of analytics team. [00:28:32] So we found the solution in the Google Suite. There's a Vertex ai. And, uh, that's, well, they're like very complex, not really UI friendly, um, tool where you could, it doesn't gather context, so there's no project, but you can, uh, feed the data in and then talk to the data. So it really saves time. Like the, the estimated time saved was already right [00:29:00] around five hours a week for our head of marketing analytics, [00:29:04] which is great. Um, so, uh, that was like the second one. So there was like, well, so, and then the add-on is, uh, is another one. So within like a month, we already showed that we could do some really fun stuff. Uh, so, uh, that's, that's where I'm like going more, like for, for summer, for example, I plan to work more with digital marketing, and that's like. Really, really difficult department because there's a lot at stake. Like I cannot build them the, like automated bidding systems [00:29:42] for like a million dollar campaign. This would be this. So, uh, they're like, if you have big budgets, then it becomes a lot more like, just a lot riskier. Uh, but reward is higher. [00:29:56] Right. And then, but I, I, I chose [00:30:00] to start working with, uh, routine tasks internally, like reporting. That's also something that like semi, semi-automated but already collectively saved around 20, 30 hours, um, of teams time, um, per month also on reporting. So it's, um, uh, now I do this, this is like five hours of my time. [00:30:23] Next month it will be three hours of my time. But then if you compare this like 30 versus my three, then it's uh, it's way better. Um, so. That's, that's how I think about this. So this grid might be totally different for a scale up will be totally different for a startup. [00:30:41] So it's like specifically for SEMrush. And uh, then I also color coded some, that's what I didn't share. Like that's something in red, something in green that I could start with. And then, uh, hyper-personalized content is my dream. Uh, it's very difficult to do, [00:31:00] but I would imagine that, uh, uh, we would be able to send something very, like hyper-personalized the moment someone like will register the account with rs. [00:31:15] So we'll send them, like we will pull our API and then we will, uh, give them something very, very unique, tailored to their account, delivered to their email. So that would be like my dream workflow to execute. [00:31:28] ​ [00:33:19] 5. How To Use Google Workspace To Skip AI Vendor Approvals --- [00:33:19] Phil: I love your point about. The team, uh, that you're not the biggest fan of right now is security and compliance. And you also said that like scale ups have a really cool opportunity right now because they have money, they have a bit of a brand building right now, but they don't have the robust legal thing around them. [00:33:37] Um, I would even say like scale ups in non-regulated industries. Like I was at a scale, uh, at my last gig, but we were in health tech and there's a ton of sensitive HIPAA compliant information, PII stuff. And we were a pretty small team and growing, but legal was like involved in [00:33:55] everything that we did. [00:33:56] Like we couldn't launch a campaign without having legal in [00:34:00] legal in the loop. Um, and you know, this is the thing that I keep thinking about with AI moving really fast. It's that like a lot of teams have procurement in place and like, you know, the reality for most marketing ops team is that there's stack only moves at the speed of procurement, [00:34:18] not innovation per se. [00:34:21] A lot of folks see all the cool, shiny things and then they chat with it and security and legal, and it's just like, yeah, I've got 17 things on my plate right now. Like, I'm not gonna be able to look at this until, uh, a year and a half from [00:34:33] now. You know, like how do you balance the pace of AI and all the cool stuff that's happening right now? [00:34:39] And the pushback, not, not necessarily pushback, but just like the pace of procurement. How do you balance that with, uh, the glacial speed of, of enterprise tooling? [00:34:50] Olga: With Google [00:34:51] Phil: Hmm. [00:34:52] Olga: like, well, so we are, we are, we're using like we are G Suite, uh, company. [00:35:00] Uh, so now I prioritize every Google update, like, and every Google tool. So. We have, we can put our financial data in Vertex ai. Great. [00:35:12] Like, well, I, well the Gemini, um, deep research is like way better now than open eyes research. [00:35:20] Like [00:35:20] that's what I showed to our, like, director of social media. And we immediately like won one case we, like, we, we would've spent like $5,000 for research like this that we just like did with the Gemini percent. Like, so, um, that's like where my attention is, like to anything that Google releases, because I know that legal team would approve it likely. [00:35:46] Phil: Hmm. [00:35:47] Olga: Um, and, uh, there's no procurement involved. We already have it. Uh, it's included. Um, and, uh, yeah, security is also fine, like, because it's within the system. So, [00:36:00] um, that's, uh, that's what we like pay more attention on. Like, Gemini is not the best yet, but I think Google will win the AI war by the end of the year, um, uh, because now their latest releases are really good. And, uh, also their adoption rate would be crazy. Like everyone has the Gmail account and then a lot of companies use G Suite. So, but they, they enabled a lot of these features for free for their business users. Um, so they're not doing a good job with education right now. Um, and then also the UI is really, really bad. [00:36:46] I. Uh, but that's something that they can fix. But like they, they're like, they have the largest audience they could release stuff to, and that's what also like everyone really, really, truly cares about. They [00:37:00] also, uh, for like bigger folks, that's like, not in my wheelhouse, but, uh, some, well if we, well, by buying the Wiz, uh, the security company, I think this was like a really smart move in terms of, um, well, well now, uh, with Wi Vibe coding, like a, a lot of stuff that, well the security online becomes like very, very key. Uh, so I, yeah, I believe, um, yeah, Google is like very smart in that direction. And then that's where I, I, I kind of figured out very quickly the guardrails and then, okay, like Google is approved, like. Almost by default, but it, it's approved a lot faster. Everything else requires time. So I, I have like this workaround, [00:37:53] Phil: The folks that are listening that are on, on the Microsoft Suite are, are hoping that Microsoft [00:37:58] picks up a little bit [00:37:59] Olga: like I, [00:38:00] well, uh, yeah, but Microsoft that is not releasing the same, uh, type of stuff. But I also, I think that a lot of companies, they, they just, um, were in Microsoft. They, they also like with their legal team, like, would look at Google stuff like in a, in a way better, um, well site, uh, than most, [00:38:20] uh, but yeah, like everything else, uh, requires an approval and, um, every small tool like, well, but also one other thing that, uh, I have, I, I thought of. For, uh, so I started also with the social media and content workflows because this is third party data or customer facing, well, data in terms of like content that we would create. So, uh, and this is not sensitive, like legal doesn't care about this, uh, that as much because while you, our, our social media, uh, results are [00:39:00] available like in, well across a lot of third party tools because this is all just open information. Um, so for those, uh, and then if you, if you wanna scrape something, if you want to, well, to build the automation within social media and something like this, then uh, paying the agency or freelancer to build the full workflow for you and for them to buy those tools would be way easier, uh, than owning them yourself. But just, uh, writing this like in a, in a good way in the contract that like you own the [00:39:37] IP and um, everything to that. And then if you were to stop working with that freelancer, then you would like buy out, uh, the account as well. Uh, [00:39:48] but yeah, that for like, for the sake of speed for anything that is nonsensitive, just, um, have this in a contract with the, with the, their, like with your vendor. [00:39:59] Phil: Very [00:40:00] cool. [00:40:00] 6. How To Decide Which AI Agent to Use --- [00:40:00] Phil: I wanna ask you a question about the future of AI agents and, and orchestration. This is the question I asked, uh, a couple of recent guests on the show, and there's, there's different perspectives here, but I feel like there's this like race for AI orchestration ha happening in MarTech. Like a lot of MarTech vendors, it's already happening today. [00:40:21] Most of the tools in your revenue stack are [00:40:24] going to have, or if not already, AI agent, like features to them. Marketing ops folks are going to face this challenge of being, um, like someone on the show called it, like AI referees in a sense, like making calls on which agents to turn on and which tools and which ones to turn off, and more importantly, how to orchestrate all of this. [00:40:44] Like if you're turning it on in your customer engagement platform and in the CDP and you're using IPAs, you mentioned NN to like connect all that stuff. Like I, is it gonna be use case specific? Like how do we orchestrate all this? Do you think that like one of the vendors [00:41:00] is going to win? Like you mentioned IPAs as being the way that you're connecting some of these proprietary tools that you're building. [00:41:06] What are your thoughts there on just like, like what advice do you have for marketing ops folks that are just like in a year from now, everything has an AI agent, which ones do I turn on? Like how do I orchestrate this mess? [00:41:20] Olga: It depends on also the size of the company and what tech stack you're using because, um, well the big, the, the big, uh, SaaS will win here. Like, well just, uh, we're adding a lot of, uh, agent. Um, capabilities like inside of SEMrush and well, Salesforce is adding AI agents like, well, if you think of like any niche, like, well, sales, marketing, customer success will help support our building, like sales, AI agents, customer success, AI agents. [00:41:54] So, um, and uh, I would, if you're already [00:42:00] like using HubSpot then, and then they have the agenda capabilities, then I would not think of like going and testing out like 10 different smaller companies. Uh, just like, again, because of procurement, because of all the hoops that you need to, uh, jump through. And, uh, and then ultimately this is about, um, even tech technological abilities of the companies to store data and then just to process data and then secure your data. And, uh, and if, if you're a larger company, then you tend to stay stick to like more secure, larger vendors. And then they will still, like eventually just well catch on and then see what's like working for smaller companies and incorporate that. So I would not go and analyze like 10 different smaller companies. [00:42:56] I would just wait, well, or I just built [00:43:00] something custom. Or I would look at the existing bigger vendors and then just ask them like, do you have this in the backlog when it's coming? And, uh, but if you're unhappy with the current vendor, [00:43:14] then it's like, it's good time to evaluate and then take this into consideration like what they have as the upcoming releases of the AI agents. [00:43:24] Phil: Yeah, that's, it's good advice. So I'll give you like a practical example maybe, and you can walk me through your thoughts. So let's say I work at a company that has HubSpot and we have, uh, Zapier as our IPAs tool, and we've got Webflow as our CMS. And, um, maybe we're, we're using census and five Tran as like our, our customer data of composable CDP and all four of those tools are telling us this is the tool to pick. [00:43:52] We have agentic features that allow you to create one-to-one personalized content. They all do it a little bit [00:44:00] different based on where they sit in the stack. Uh, Zapier says that like, we can do it by combining a bunch of different things together. Webflow says we can do it on page for people personalizing content. [00:44:11] Um, your composable CDP can do it based on first party data that's in the warehouse. Like all of these tools are saying they're able to do it for a similar use case. How do you decide which ones to turn it on? Like do you just experiment with the tools that have it already and you pick the one that works best? [00:44:29] What are your thoughts there? [00:44:32] Olga: I would not trust, like I would trust neither. Yeah. I, I just, um, I would, well, I would build if, well how you describe this, I would actually just, um, build the custom flow [00:44:50] Phil: Hmm. [00:44:51] Olga: and I, because none of these tools, they're actually specialized, like in, well, they just, it would be like an add-on for, for, for [00:45:00] them. They, they have all have a lot of integrations. Uh, so I would just, yeah, I would build something as well in a connector like in IT then where they, well, we just, and then the important thing is that you need to self-host it, um, for security and, but um, yeah, overall I would, I would, I would not trust any of these. [00:45:24] Phil: Very cool. Yeah, it's, uh, it's interesting to hear you say that, and I guess like your first answer was also like, it depends on the stage of company that you're at, like if you're a startup, and just this question of [00:45:35] like [00:45:36] Olga: You would not have all of the tech stack that you just described, [00:45:39] so just [00:45:41] Phil: definitely not. And if you do, you're in like the free tier and you have access [00:45:45] to [00:45:45] like [00:45:46] Olga: Yeah, yeah, You would have, well, you would just have a CRM. So like you would use everything that HubSpot gives you, um, with, with this. And then you would just be grateful, like, well, just for, for their limits for the AI agent. [00:46:00] Uh, and then you would like, we would think it's okay. It's good enough. [00:46:03] Uh, but, uh, overall, yeah, like if, um, for the scale up, then again, like, it just, it's all depends on the budget and then where it's cheaper. Um, so it's, it's a lot of different factors. And then who actually oversees this as well? Uh, and if you have technical capabilities, because like for example, I have now, like I, we've hired a person who would just like fully manage this. Uh, if you don't have that person, of course you'll just buy something off the shelf and then that's where you just experiment with three. Uh, you just go and like read their views on G two and then just make your decision on this. [00:46:44] 7. How To Build an AI-First Reflex in Marketing Ops --- [00:46:44] Phil: Yeah, I, I want to ask you about experimentation a little bit here, because likewe, we chat with different people at different spots of the adoption curve, like still today, surprisingly, there's people that are kind of polar opposites to you and, you know, aren't. [00:46:56] Fully awake yet to their reality that in a few years [00:47:00] maybe they might not have a job that, that they currently do today. And they're just now starting to use chat, GPT to summarize meetings, you know, like [00:47:08] very basic one-on-one stuff. Um, and a lot of folks like the advice they give if you're starting or you're very new, just like experiment, just like get accounts and a bunch of different things. [00:47:18] Play around with stuff, follow a couple of people read this blog post. Um, but like this idea of making time for experimentation is easier said than done. You called, uh, AI and marketing ops the biggest sandbox to play. To plan right now. Um, what makes this the highest leverage place for marketing leadership today? [00:47:39] Like how do you find time to add new things in that sandbox and keep up with everything that's, that's getting built? [00:47:46] Olga: If I have a task, the first question I'm asking myself is, can I do it with ai? So it's like, there's no, there's no backlog, there's like nothing. This is just, I have this task. How can I make it [00:48:00] quicker? And, um, if I don't know how to make it quicker, I, I would ask, uh, well, I would, yeah, I would research, uh, with, uh, with, with ai and, uh, I'm not using AI fully for work. [00:48:16] I'm using it for everything. I'm using this for relationship. Like I was walking the dog. And I was chatting to Claude, and there's like a huge project. I have all August Health as my Claude project with all of my DNA analysis, like, well, all of my blood tests and all that [00:48:34] stuff, like all of my uploaded, uh, grocery, uh, orders. [00:48:39] So it just like, it, uh, were, but yeah, I was, I was, um, recording. I was like, I, I told ai, just ask me all of the questions about relationships, about my emotions and what I care about. And then I was just in audio mode. I was telling this. Um, and then I, at the end of the chat, I asked, [00:49:00] uh, the AI to summarize this, and I copy pasted like a huge thing and I sent to my partner [00:49:05] Phil: Hmm. [00:49:06] Olga: and, um, so I, we didn't have a, like a ceiling lamp in the kitchen. [00:49:11] And I, well, I took a photo, asked Claude just to like, just to suggest this, like, well, five versions. It suggests five versions. Ask them to like, to translate this to Spanish. I then asked Open like Chad g PT to search for this in Spanish websites. And then like, we bought the ceiling lab like much faster because it took us like, no time to find this. [00:49:35] So, um, that's how I, I approach all of the work, uh, stuff, all of the personal things and uh, that just like I'm training myself, uh, to think AI first, like how can I do it faster? How can AI help me? And that's why, that's why it's the biggest sandbox. Like it just, uh, for example, um, if [00:50:00] you, I don't know if, if you, if you have wood and they just need to crush something with wood, like it only works if you have skills for this. But like with ai, if you are an artist, then it's applicable. Like if, if you are like a marketing officer, you're applicable. If, uh, if you're a mom, it's applicable. Like, it's like infinite use cases and it all is tailored to the, to you, like, to anybody. And, uh, that's why it's so great. Like, it's just, it's universal and the only limit of this is like what you imagine, um, this to, to do for you. Um, so yeah, that's like, I I have not seen anything, uh, like this. Um, the, so like you can, you can chat about AI like with anybody, and then they will have the use case that you never thought of, of, and then you'll have the use case, uh, they never thought of. And that, that's like you're talking about the same thing. [00:50:58] It just like, it's just a tool.[00:51:00] [00:51:00] Phil: I love your, your story chatting with AI over voice while you're walking your dog. I actually started doing that recently. Um, like I'm still a solopreneur. I got a couple advisors and, and partners that, that help me with the podcast, but for the most part, my business partner is Claude and I take him on walks with me while I'm walking the dog. [00:51:21] And recently we just got into like, just hypothetical conversations and thinking like bigger pictures instead of like the local maximum that I'm focused on when I'm just like in the day-to-day, like busy work. And it's just like been eyeopening, like almost like a business therapist in a sense. And yeah, to your point, like everyone you talk to that, that does like the voice stuff, um, like you, you, you discover a new use case for it that you didn't think of before. [00:51:48] I, I love the shopping one, like using both Claude and and and Chad GPT 'cause GPT you can, you can still search stuff. I was still waiting for that to, to, to happen in, in Claude there. [00:51:59] 8. AI’s Endgame: Play-to-Earn and Mandatory Human Quotas --- [00:51:59] Phil: I wanna ask you about [00:52:00] the dark side of too good ai, though. Like we, we chat about a lot of the positives we kind of teased out at the start of the conversation, like the potential dark sides and job replacement there. [00:52:10] But so you've said in, uh, like one of the big questions is what happens when AI gets too good and the agents are better, faster, cheaper? Where do people go after that? Like, are we looking at a future where we need to invent new kinds of jobs, um, simply to keep up, uh, simply to keep people employed? Like what are your thoughts there? [00:52:35] I know you, you have some things there. [00:52:37] Olga: Yeah. Uh, so yeah, first of all, there will be new industries. People will create them. Uh, but I think, uh, and there are some people who like talk about this in a, well, in a very sophisticated way that, uh, well if AI is. So great that like a lot of people would just not need to work. Um, and, [00:53:00] uh, we'll just have like a, um, well just people just, well getting the money, but like not working. Uh, which I think it just like sounds too, too too good to be true [00:53:12] or like too scary for me. Um, but I feel that, well, if people just are not like, they don't have like a regular job. Um, everyone loves gaming. Um, and, uh, so I really believe this in like in a play to earn industry, um, where, well, it's like very niche now and more Web3, but if a lot of people just, well go and well play more, uh, then the gaming companies would be very happy to, to adopt this and then have more people play than it means like a lot more advertising money. [00:53:50] Um, so. And yeah, people just enjoy, uh, what they would've enjoyed anyway, and then they just earn while playing. Uh, so I [00:54:00] feel that this would be the fun and then that's the future that I, I, I kind of can envision. Um, and there, yeah, there just a lot of like to earn stuff. Like, [00:54:13] Phil: Yeah. [00:54:13] Olga: just, uh, and it would be, it would be just people looking for those opportunities where with their free time, where they could do simple things online. And, uh, they also will get probably some money for their, from the government. If, like, if AI is so great that it could do stuff for them, then I wouldn't worry about, um, the payments. So, [00:54:36] Phil: Yeah. We're, we're almost describing, have you seen the movie, uh, Wally? [00:54:41] Olga: yeah. [00:54:41] Phil: Yeah, [00:54:42] it's, it's almost like that future of Wally where like all humans are just like, really bad health, don't move around anymore. They're just like floating around in their chairs on their computers, earning money by playing or just like providing information, filling out surveys and, and stuff like that. [00:54:59] And [00:55:00] yeah, it's a, it's a fascinating thought, thought, uh, exercise. I've actually had this exercise with, with Claude, like, uh, a, a few times about this and like thinking about other business models that allow you to shift to that like earn model that, that could be attractive for, for advertisers. Um, but like one thing I wanted to ask you is, like, LinkedIn right now is like littered with founders and leaders that are saying things like. [00:55:30] Ai. Now I, I only have to hire one person instead of 20. Or like, my goal next year is to build a company that hits $10 billion with only 20 people on staff. You've actually said that in this context, governments might impose quotas on human employees at companies. Do you see this as like a likely regulation [00:55:53] or [00:55:53] Olga: Yeah. I, I see, I see this as like. This is a guaranteed future, uh, [00:56:00] because it would be, well, I thought, well, regulated industries, uh, will have this anyway, but, uh, it could be if you have an audience of 10,000 people, then you're obliged to have like five customer success, [00:56:13] Phil: Hmm. [00:56:14] Olga: team members and this is how it could, uh, yeah, they could regulate and it would have like places for, for, for humans there. Um, so yeah, I, I see this as a very likely scenario, um, because well just like, otherwise you can't really, um, well, yeah, regular, like stop the pace of just, uh, like a lot of AI companies just going fully zero human. [00:56:41] Phil: Yeah. [00:56:42] Olga: But there was a, there was a very funny. I don't know if you've seen, there was a company who had this huge, uh, times Square, um, banner, like stop hiring humans, and it just like, all like, just all about the [00:56:56] sales AI agents and, uh, the founder, [00:57:00] uh, was he he posted the LinkedIn post saying that, yeah, we like huge, had huge success this month and also just our viral campaign, uh, stop hiring humans. What I'm also proud of is that we hired a team of amazing, uh, amazing, uh, professionals. And then he's like, they're like two contradictory things together, like that. They have like a huge disruptive campaign of like not hiring anybody at the same time. Like the, one of the highlights, uh, is like him hiring a great human. [00:57:35] So, uh, I, yeah, I will find this. Uh, so [00:57:39] Phil: That's too funny. Yeah, we'll post that in the video version of this, but yeah, it makes me think of like this, this potential like counterculture that we might encounter where. You know, the companies that might come out of the woodwork are the ones that say like, Hey, in the next five years, like everyone is not gonna be hiring humans. [00:57:58] We are not gonna be [00:58:00] using ai. This is built by humans. We have no chat bots here. Like we're anti chat bot. We're all humans. We hire humans. We support humans. Like this whole counterculture of like people being mad at a [00:58:13] company because they're using AI to do [00:58:15] this, and [00:58:16] Olga: it already is already, uh, happening. And, uh, and uh, so there were like two memos, uh, of, uh, [00:58:23] CEOs. [00:58:25] Phil: Duolingo, Shopify. [00:58:25] Olga: And like Shopify was really well received while Duolingo really wasn't like, they had like a pushback. Um, and I think it just all depends on the tone, uh, because. Um, well, I analyzed the, both, like both and then it's just, Shopify's was more visionary and then about a growth mindset. [00:58:46] And this is how we help you succeed. Well, during it was more like business oriented, saying like, here, here's our trajectory. And uh, they are just like, they outlined the specific cases. They said that they will be [00:59:00] hiring, well, they, well, they will be using ai, not hiring, uh, contractors anymore. And um, the fact that they said like, we still care about employees and they will be supported. It came like later in the email and then it's still said like, change can be scary. So it had like all like a, a bit like different tone of voice and it's not about, uh, just, well, AI is here, we help you. Transform. It's like, well, just, Jo was going AI first. So it's like two very, very different. It's like same, the same thing. But uh, first like Shopify was saying, yes, we, we are here for you. We wanna lift you up. [00:59:40] And then, uh, Jo was saying, yeah, this is, we can't wait until like, um, it would be a hundred percent great. And then we just celebrated transitioning. So same thing, different messaging and like marketing does itself work. [00:59:55] So, uh, yeah. And like whenever people [01:00:00] feel now that, okay, this is ai first they care more about AI than humans. Whenever, like there is a sense of that, then it's this immediate pushback. [01:00:11] Phil: And it makes sense. It's like self preservation [01:00:14] kicking in there. And then the flip of the coin, there are people that are just like, oh, well, instead of complaining about people using AI and potentially losing jobs, I. I wanna work on ai. Like I want to be the person thinking about that stuff. And, and that was makes me think about like our ops people, like potential listeners preparing for the wrong future. [01:00:34] Like given what we've chatted about. And you know, potentially in a couple years we have like quotas on certain number of people being hired. Um, most ops leaders today are focused still on like learning tools and building integrations and cleaning data. But if like the future that you're describing plays out, should ops be preparing more for things like governance and ethics and workforce [01:01:00] design, and then just like the technical implementation of tooling and being like the, the people like implementing AI tools, like I think there's a whole other layer of that, of like governance and ethics and workforce design that like supersedes the tooling implementation. [01:01:16] What do you think there? [01:01:18] Olga: Ideally, but no one's like, well, ops are ops because they're more, uh, they have strategy, but they're tactical, like they make things happen. So they enable the strategic people and the visionaries, uh, to scale this operations and then make the like world, like well actually make things work. Uh, so, uh, that's, that's where I think now. We're at the stage where very little, well, very small amount of companies, um, are, can, can say that [01:02:00] they are like in a, well, they're, they're using AI on a really good level. Like there are a lot of like, streamlined processes. And this is where like ops need to like, help all of the companies step up [01:02:13] and then be very comfortable with ai. Once it's happening, then like, it's another step of thinking of, uh, of ethics and then of just, uh, how this gonna play out in the, in the long term, like learn term. But now it just, well, everyone's in a still, like most companies are in an infancy [01:02:33] Phil: Yeah. Yeah. [01:02:34] Olga: I don't even think that the, like the AI companies, like they're drinking their own Kool-Aid. [01:02:40] Like they just, I, I'm not sure. Um, and, um. So it's, uh, yeah, like if, if there's a startup that's building AI tools there, like there might not be AI savvy internally as a team. They are focused a lot more on customers, [01:02:57] but they're not like, so I would say [01:03:00] that ops like the now what we need to do is that we need to educate everybody and we need to enable everyone, like to really be well proficient in AI or help ensuring that we have like automated, um, workflows. [01:03:16] Phil: Yeah, that's such a good point. Uh, we had someone from OpenAI in, in episode one 70, and, uh, we were trying to like, get answers about how their dog feeding, uh, OpenAI at OpenAI and he works in uh, GTM systems and he was like, man, we've added like 10, like a bunch of people, we 10 x like just our team in the last like three months. [01:03:38] Like I am focused on like, trying to use it as much as possible, but like, dude, we're like, we're so busy with this and with that and focusing on customers. So I think that that relates to what you just said there. Um, I'm looking at time Olga, and this has flown by. Honestly, like next level of conversation. [01:03:55] Really appreciate your, your insights and, and taking the time to chat with us. [01:03:58] 9. What Happens When You Optimize Your Body Like a Martech Stack --- [01:03:58] Phil: I got one last question for [01:04:00] you. Uh, you're obviously a VP of Marketing Ops and AI ops at SEMrush. You're a team leader. You've also been marketer of the year frequent speaker. You're also a dog mom and an avid foodie. 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? [01:04:23] Olga: I don't have balance, uh, I have priorities and I believe fully that if you wanna have balance, you will be me mediocre at everything. [01:04:35] Uh, so I choose, well, I choose relationship. I choose work. And choose, like personal growth and cooking. Um, so those are my prayer and like, and, uh, how I stay, uh, happy is that I stay healthy. [01:04:53] Um, and then I have like routines and I, I'm true believer, like ever since pandemic, like I will, I wake up [01:05:00] every, well, every single day I wake up almost at the same time. I have three sleep trackers. Uh, I don't drink caffeine. I drink a lot of water. Like I, well, I don't wear sunglasses. I wear a cap. When I walk the dog. I have sunlight, uh, in my eyes as soon as possible. Um, and this is just like a lot of small health, uh, habits. I have that compound over time. I take cold shower, so it's like all of the things. And then I cook at home, I eat healthy. And that's like where like if my body functions. Uh, that makes everything else a lot easier. [01:05:40] Uh, so I invest a lot of time there. Plus, yeah, only a few priorities. Um, and like everything else is already secondary and I know that I can't win with everything, but I, well, I can win with things that I care about. [01:05:56] Phil: Very cool. Love it. Awesome advice. Big fan of routines. [01:06:00] Uh, love the sunlight in your face as soon as possible. Uh, I love being. Someone who's mindful about the first thing you do when you wake up isn't like, look at your phone and, and check emails and check texts, like, pet your dog or like, do something with your partner instead of, uh, defaulting to a digital sense of community right away. [01:06:20] Olga, it's been super fun. Really appreciate your time. Uh, won't get to obviously SEMrush and a lot of the stuff you guys are doing and building. [01:06:27] Thanks so much for your time again, Elga. Really appreciate it. [01:06:30] Olga: Thank you very much. Thank you for having me.