The Biggest Mistakes Marketers Make with AI (And How to Fix It) with Phil Lakin, Head of Enterprise Innovation at Zapier | Ep. 350

0:01Welcome to the Marketing Millennials, the no BS marketing podcast. I'm Daniel Murray and join me for unfiltered conversations with the brains behind marketing's coolest companies. The one request I tell our guests, stories or it didn't happen. Get ready to turn the f*** up. What's up, Phil? Welcome to the podcast. Good to be here. Thanks for having me. I want people to know your background so we could talk AI and prove that you actually are in the weeds doing this every single day. So could you give a little background of who you are and why they should listen? Sure. Well, hey, my name is Phil Blakin.

0:48I'm the head of enterprise innovation at Zapier. and I came into this role about a year ago through acquisition of my company, NoCodeOps, where we were a big part of the NoCode movement. We built the first observability tool in the NoCode and automation AI space, as well as built the first community dedicated to operations professionals, not just entrepreneurs and founders building stuff with NoCode and automation. So think marketing ops, sales ops, rev ops, people ops, et cetera. And so we've been working with that audience for quite some time now. And since I've been here at Zapier, I've been really spending most of my time doing content for our top of funnel and doing a lot of content led innovation stuff, all the way down to product packaging and pricing through helping one of our newer programs called the AI Transformation Package, where me and my team go into companies, embed into companies and help them go AI native within one year.

1:47And so with their teams getting hands-on and building themselves, we're not just building the stuff for them. So I'm spending all my time translating real-world problems that different departments have, especially on the GTM side with marketing, sales, revenue, into orchestrated AI solutions. So I'm both talking very top of funnel about this stuff and making sure the terms and the things make sense and that we're talking about the stuff in the right way and marketing this as well as the person on the very bottom, like helping folks build the programming and automations to help the whole company thrive.

2:23So that's me. Amazing. So yeah, everybody. So when we're talking, Phil really understands because you're actually implementing this today. So, I mean, I used to use, I've been using Zapier for 10 plus years and it used to be like the go-to tool for automation and now it's still great the go-to tool for automation but now it's moving to an AI first approach so the question I have for you is like what is the biggest mistake marketers make when adding AI into their workflow oh man there's a bunch so I guess let me rewind that for a second like break that out into two areas, right? Because there are several ways that you can use AI within a tool like Zapier.

3:10And let's talk about the mistakes after that, right? How they kind of map. So look, Zapier got famous for like automation, like if something happens over here, then do something over here. So very like simple triggers and actions. And those expanded over time into like paths and conditions and, you know, 8 ,000 different apps, 30 ,000 different actions and triggers without having to write a single line of code. So that grew and grew and grew. And when Gen AI came around a few years ago, we brought it in in a few different ways into the product. So one is, just like every other SaaS product under the sun, we have some AI features that are really helpful.

3:46We have a co-pilot that can help you build Zaps. We have a co-pilot that can help you build entire systems of multiple Zaps and other tools like tables and interfaces within Zapier. You know, we have AI to help you do error resolution, right? So you don't have to like read code or anything like that anymore to figure out errors. So we're using AI in a lot of ways to help you use the product better. But also the most popular version of AI is the second one, which is you can bring an AI step into Zapier. And so what that means is, let's say someone submitted a form and you got a bunch of fields from that form and say you delivered all that stuff to your SDR to reach out to the person if they were qualified, say, hey, write this person email.

4:32Maybe here's a template of the email. Feel free to customize and send. And it might just be like some basic template, right? Well, now what you can do with an AI step is instead of that marketer taking all of your stuff, throwing that into ChatChap and rewriting it for themselves, you can actually have the AI step in Zapier use all of the fields that were submitted, some basic prompt from you, right, with some past examples that have worked really well, and generate a personalized response to that person that then you're sending to the SDR to press approve or reject or whatever on. And so you're saving a whole, you're doing two things there.

5:09You're saving an entire step where the SDR doesn't have to go to check GPT to do that stuff. And you're allowing that not every SDR needs their own GPT, that like the best learnings are coming back up and hitting that earlier prompt that you control. So they're all getting the best, you know, the best version of a first draft. And that is super powerful. So, and then there's a third way, you know, using AI and Zapier, which is agents. And agents kind of break the mold of the traditional, like, you know, if this happens, then do this, this, this, you know, what I just described was AI in a workflow, unlike all of the LinkedIn gurus out there that think that that is an agent, that is not an agent, that is simply AI in a workflow.

5:51An agent is when you say you've got a trigger and then after that trigger happens, you've got, you know, here's some knowledge I'm going to give you access to. And that could be our wiki. It could be our HubSpot. It could be Salesforce. You know, maybe it could be the whole internet. Do some Google searching, right? I'm going to give you some instructions, like a job description, like I would give an intern. First, I want you to start by doing this and then do this and then do this and then maybe read this, maybe send this thing. And then I'm going to give you access to tools, right? So you're allowed to read this.

6:21You're allowed to send a Slack on my behalf. You're allowed to update this HubSpot field, et cetera, et cetera. Now, they're less deterministic than the other workflows. They're more inference-based, so they can make their own decisions. Now, the nice part about Zapier is that you can say, all I want you to do is this one job. So maybe do some internet research and do some qualification 1 through 10 about this person based on some of these knowledge sources I give the agent access to, right, about who's like our best ICP. And then you can bring all of that data back into a deterministic workflow.

6:55So you can get the advantage of both things in Zappi. You can carve out different things for different jobs, which brings me to the biggest mistakes that I see marketers make here, which is they think that now that AI is a thing, you can just like throw everything at an AI or at a chat or at an agent and just be like, cool, it's just going to fix things. It's not. You still have to think through the process. You still have to break down what parts of the process you want a workflow to own versus an AI step to own versus an agent to own and how you want those to work together. You've still got to battle test the outputs of the agents, just like you would never give an intern, like an overly enthusiastic intern with access to the internet, just like, hey, go figure it out.

7:36Then I'm never going to look at your work again. No, you're going to sit with them the first few times and battle test and see what's working, what's not working. You give them feedback. You're going to check in on them every now and then, right? So the same type of thing. You might have different levels of interns, right? You might have like, you know, interns that are level one and interns that are level three, right? And you're going to want to give them different types of work. And so you're going to break things out a little bit. You know, you're going to want to choose what you want to give to who, right?

8:03In intern level one, you may just like want to have run in a very simple deterministic workflow with a very little AI step. something you trust a little bit more and you've really checked the outputs of, maybe you want to give to an agent, right, where it has a little bit more free will and inference. So the biggest problem that I see is that people aren't breaking these things out and they're not thinking about what is the best tool for the job at every step of the way. And they're not spending time learning the difference between these tools and how they can actually work. And the funny thing is that 95 % of all of this stuff existed before AI.

8:35It's just AI made it super popular. So, yeah, you know, I always used to say to people when I was in marketing ops is you have to be able to do the process manually and know what looks good manually before you start automating the process. And I think it's the same as AI. Like, how are you going to know what good looks like or what great looks like if it hasn't been done before in your company or have someone who knows what you've got to speak the language? Because I could say, yeah, I'm like, I can say SDR, go do this. But if I don't know what good sales outreach looks like, I don't know what a good email looks like.

9:14I don't know what a good script looks like. I could just put it, the AI could be doing things. I'll be like, oh, that's great. But is it right? Probably not. So you always need to, I like what you just said, go through the process first, map out the process. it's the same 10 years ago with automations like map out each automation it's just now it's just doing a lot more in a little bit of time which is the good part of it saving us a lot more time is automation saved us the first 10 to 15 hours and now this is saving us another 20 30 40 hours to do non-manual work that we, more strategic stuff for the business.

10:00I think the other thing that the other biggest mistake I see marketers make here, I think there's two of them. Say one of them is they think that when AI does something, it means AI has to do everything. And that's not true. You can have AI be a part of a human in the loop process. So I'll give you some examples. Like if you have an inbound workflow, right? and I have an AI agent that I'm passing off some of that data to that came in through the initial form saying like, hey, look at all the knowledge sources that I just gave you access to in our company. Grade this person one through 10. And then maybe if they're like, you know, if they're a five or below, send them, just put them on a sequence and let them have fun, right?

10:37Like, you know, if they're a, you know, eight, nine or 10, assign them an AE and write that initial outreach message from the AE using an AI step and send that to them. But if they're like a, yeah, I guess six or a seven, right? This is kind of like in that maybe zone. What I want you to do is send me a Slack message. And this is a step in Zapier called request approval. Send me a Slack message with two buttons in it. And show me in that Slack message, all the information you had about that person, why you graded them a six or a seven, and let me make the decision, right, of which path they should go down.

11:14The other thing too is you can have, like, say the email is coming directly from you. I can always, instead of just sending an email for me that was AI generated, I could say, just send me a draft, put it in this label and send me a Slack link to that draft. And so basically I just get a Slack link that I just have to go click, review, maybe edit it, judge a bit, and then send. So just because AI is a part of the process doesn't mean it has to take over every little step. I think marketers are a bit scared of that they shouldn't be. It's just a matter of who has what job. I love that. I think we, I mean, the scoring thing is that example alone saves a lot because we used to have like an ACP at my last company, like an acceptable customer profile.

11:57and we didn't really do we had to like have a whole process on this but if we can just have a tiger team of ai that we we put to put do the initial outreach that's a great step or people who are dq'd send let the ai send them like tutorials blogs they might be yeah put them a fast-tracked, have more of a design process, acceptable process, like accepting process where a human isn't involved because you actually need, humans still need to be part of the process, especially in sales because these are big accounts. You can't just blame an AI that you ruined that account. And that's where you can mix determinism and non-determinism, right?

12:46You can say like, look, if it's 500 plus people, I don't care who they are. I want to talk to them. You know what I mean? You can have overrides too. You can write those overrides into the agent. They're really good at responding to that type of stuff. But let's get super meta and crazy for a moment. A lot of these times, these lead routing things have been built into very deterministic rule sets, right? Like all routing is, or like all qualifications, just like a set of rules and scores, right? Not that crazy, right? Maybe we're replacing some of that stuff with an agent, but here's the cooler part.

13:16What I can then do is have a separate agent outside of this whole system that we just built in the inbound side here, right? Looking at every week or every month, right? So the trigger would be time-based, right? So every week or every month, your job is to look at all of our closed one or all of the things that are looking likely to close. Look at the messages that were sent to those folks, or the ICP or the demographics or the technographics of those companies that we know about, and write a report and show that versus where our current qualification agent is currently in that prompt, and make a suggestion to update our prompt to better score people based on what's actually working.

14:05So you can use live data. Yeah, that's amazing. I remember we had a six years ago, manually, every month we had a rescore based on new data that came in. If AEs are closing, closed one, which one's closed one or are the other qualifications. Now they can also, you can have the AI agent look for other qualities that better determine a closed one op based on the fields on the record too. So it's like you could go even more meta and find things that find qualities that you weren't even looking for because you can't look at the data all day, which I think is things that people aren't doing where you aren't looking for extra qualities that people – but I want to also go back a step.

14:58Okay, we talked about mistakes a lot of marketers made. Oh, I did have one more, which is really simple. Okay, you go ahead. There's the third, which is pretty easy. I find that marketers get really jazzed about like point solutions that are essentially just like some type of database, some type of workflows and some type of AI model that they're hitting. Right. But it's highly opinionated towards like a specific area. They just give you an example, like, I don't know, like a revenue orchestration platform or something like that, or like a lead router or I don't know, whatever. you know what I mean?

15:35Like where it's just like, uh, they're basically just like a few rules, you know, workflows, AI, and a database with like some fancy features and some opinionated thinking. And some of those in the extreme end of opinion anything I think are still super valuable, but I think there's a lot of them that like, if you just spend the time to sit down with Zapier or like a tool, like an AI orchestration tool, I guess, you could be replacing hundreds of thousands of dollars of point solutions in your stack. Should you learn the basics? And the biggest mistake is just like seeing something and being like, we want to buy that, we have to integrate it versus like, what if I just gave you the building blocks to own that on your own for like 5 % of the cost?

16:13i love that because i also now where marketed marketings have gone as they like their data their own own database is the source of truth where a year 10 years ago salesforce or some other tool was the source of truth data now you can connect that to your database or something like that and have rules out the database which changed a lot more of like letting a platform control all the rules that you have, and then connecting, like you said, a point solution to that platform because that platform didn't know how to do that one thing. Usually, it is just a bunch of point solutions attached to one big platform.

16:57100%. And now, imagine the power of building most of those point solutions in one system with the same fundamental building blocks. It puts you in control. It allows you to customize. Look, it does come with a cost of maintenance. I don't want to hide that, right? There is a cost of maintenance on some of the stuff, but oftentimes that cost of maintenance is way less than the cost of a point solution, right? I would say there's a cost of maintenance of all these, like someone has to run these tools. You still got to go maintain the point solution also, right? Yeah, exactly. Someone has to run that tool.

17:29Someone has to run the connection between that tool to the other. And someone probably owns the point solution and IT may be owning the database. So you have to work with five different people to get one thing done so that's but i wanted to take a step back and also say for marketers just starting like implementing ai they they probably have used chat gp to chat gbt a little bit um they they may be inputting data into chat gbt a little bit what are some i wouldn't say like too detailed example what are some quick wins that marketers could do right now that or what are some you've seen with some of the customers your customers examples that they can implement and today that could help their marketing or or change their marketing or i mean obviously whoever's listening there is different for every company but there are some there probably are some quick wins that you're seeing that people aren't even thinking about that are pretty easy to implement.

18:36Yeah. Look, I'd say on the thinking side, one thing that you can always do, right? And I see people do this a lot, like make this mistake a lot where it's just like, come up with a good marketing campaign idea for this, right? And it's just like, no, that is like telling an intern who has no idea about anything, very little about your company to go do that. I always say overload with context, right? Always put in a lot of context. But beyond that, find the public work of people in that area you're asking AI to build the thing. Go search for those people and ask AI, like ask your chat GPT, if you're interested in chat GPT, right?

19:17Being like, I want you to give me feedback on this initiative, or I want you to create this initiative from the standpoint of these two or three people working together, right? So, it will research those folks and how they think about the world. And now it's going to give you a lot more specific of an output, right? As opposed to just that general thing. So basically, you're telling the intern, like, look, you don't know much about our company yet. Here's a whole bunch of context about the company. Here's a whole bunch of context about, you know, XYZ. Oh, and by the way, here's three people to go research that I really love.

19:47Take all of that and then give me back a first draft. And you're going to get a way better output. So the more specific you can get with AI, the better you're going to get an output for. Yeah, I love that. I want to just stop on that point. I think too many people are not spending enough time on the prompt. They're just saying, hey, AI, do this or that for me instead of details. Just like an intern, the more information you give them, the better job they would do. The AI still needs that. And so many people are falling short of just saying one little line, hey, could you figure out how to do this instead of telling them who they need to act like, what they need to research.

20:34Here's context of our company, because it could be different from every company, and going in more detail, prompting them instead of. So I think that's a great point. That's going to be huge. And if you're confused about how to write a great prompt, don't even feel like you need to research it. Just putting your initial, you know, okay level prompt and just ask AI, what do you need for me to improve this prompt? And then it's going to ask you, and then you're going to respond with that and then ask it to now write the prompt with that. Right. So just like with an intern, I could say, go come up with this campaign.

21:06I could make, but also like intern with the way that you like to communicate and the way that you work in your working style, what do you need to know from me? And then it'll just tell you, and then you can communicate with it. It's like working with a coworker, you know? Exactly. I'll let you go to any other wins that you think could be good too. Yeah. Look, I think I kind of mentioned one earlier where it's just like trying to replace, certain point solutions I think is very, very powerful. I would say find one area, just one area where you're like repetitively going to chat GPT or whatever kind of chat model you're using for a certain thing where you're like, okay, every week I'm writing this report, I'm having ChackGPT help me do a thing.

21:52Or every day we get these types of requests in and I'm using ChackGPT to rewrite these things and send it back out, whatever it is. Something repetitive where you're just constantly going to ChackGPT. What I would do is try to use an orchestration platform like Zapier to say, the trigger is this, wherever that data is coming into you via, right um the next step is an ai step where it's going to mix some of that data from the inputs of that with a prompt that i'm typically using and then it's going to output into a slack message that comes to me for approval to send to wherever right that way you don't have to be going back and forth all the time the data will find you the draft will find you and like you can always update the prompt and the inputs and all that stuff.

22:33But like that saves me so much time every day, every week. So if there's reports you're prepping for certain leadership folks, if there's like, um, if there's reports you're getting from leadership that you need to translate to your different departments, like bring AI prompts into that workflow to help you come up with the first drafts where you can just dump in the original thing and get everything you need and then edit from there, as opposed to having to go back and forth and write like 20 different things and have 20 different prompts. Yeah. Another question I have is, I mean, you started a pretty big community for ops folks.

23:13You said a little bit earlier, one of the things you should do is say, go research these three people. But what are some ways that people, what are some communities you think about or resources do you think about that people can join to start improving their knowledge base of AI or finding these people? Because that's sometimes half the battle is finding where to go. To get inspired, right? Yeah, exactly. I think there's a few, man. I think, of course, the community that we started, no code ops is still owned and operated by zapier and claudia is our community manager who's incredible still runs it it's a free discord so it's just nocodeops.com um i think operations nation is another really great one i think that generalist world is quite phenomenal um you know more for generalists who are just kind of going across different areas i think marketers kind of have to learn how to be a little bit more because i think we're going to leave some specialization stuff.

24:19Like, look, just so you know, the world of sales is teaching us what's going to happen. A lot of roles are going to fold into like what the GTM engineer equivalent is, right? Because of like how fast people can now move. And so I think marketers are going to have to learn how to become generalists a little bit where they know a little bit of strategy. They know a little bit of, you know, content creation. They know a little bit of the automation side. They know a little bit of the system side. They can combine all of those things to be like a secret, you know, like a secret weapon for all of those in one.

24:48And so I love kind of the ideas of generalist world there. And then honestly, like I would just like I would be on TikTok and LinkedIn looking for this type of stuff. Like I get inspired every day from stuff I see on on TikTok and LinkedIn, just seeing what other people are doing. You know, make sure you're interacting with content and doing searches around AI and marketing and like your feed will, you know, kind of help you out on the algorithm side of that. But like, dude, I'm learning so much stuff from just curating my own feeds. So, you know, it's actually kind of funny just how you mentioned like the go to market engineer becoming the new role.

25:27But it's funny, ops used to be builders, and then they changed it into a role where they had to be builders and insights. But now it's gone back to you have to be a heavy builder again. Because before, it was a lot less building and more just connecting. And now you have to go build systems again, which is what the ops role was. At least marketing ops for a while. I think so, too. But I think we're seeing like a combination of all of those things, right? Where it's like, I'm seeing strategists have to become ops people. I'm seeing ops people have to become strategists, right? Like the reality is the first one to do both is going to win, right?

26:10And so like, you know, I think people have to get a little bit uncomfortable here, right? Because if in the ops world, creating a system that, you know, maintaining a system used to take, you know, 30 hours every week, now it only takes 10, right? You know, maybe save some of that time to go hang out with your family and kids, right? But like in the other part of that, you should be learning the strategy side. Like you shouldn't just be settling and stopping. You should be learning the insights side. You should be learning the data side. You should be learning the strategy side because I'm telling you, companies are going to get smaller with all of these things.

26:43It is what's – I'm very optimistic about all of this stuff and I think that new roles are going to evolve and change. But like the folks that are going to be the most well-suited and on the most stable ground are going to be the folks that are going to get a little bit uncomfortable during this time and be like, If I only really feel good in ops or I really feel any good in strategy, at least learn some of the language of the ops stuff. At least sit in an AI bootcamp where you learn what the heck is possible, right? Like, you know what I mean? Like get a little bit uncomfortable here. I think that, I mean, that was always, I feel like going out of college, the marketing role was one of the hardest roles to get.