View Auto-Generated Transcript
The below transcript has been auto-generated for your convenience. Please reference the source video/audio for direct quotes or to clarify any errors.
EVAN: There's a huge opportunity for AI to summarize and recommend at scale. To a human operator, a human strategist who then takes that as input.
CONNOR: What follows is a conversation with Evan Dunn. We talk about his early work in AI back in 2017 and why it didn't work then. AI versus human agents, the categories of AI software he's most excited about, actual GTM work he's done leveraging AI, when to systematize versus when to experiment, and the absolute best way to test your demand gen strategy, AI or not.
Let's listen in.
All right, Evan, welcome. What's up, man?
EVAN: Hey, thanks for having me, Connor. Great to podcast with you again.
CONNOR: Good to podcast with you again. Our producer at the beginning of the show asked Evan if he knew how to Riverside. And Evan was like, "yo, man, we like, Connor and I met on a Riverside. Like I had him on my show. So like, I kind of know what's up."
EVAN: I have this theory, Connor, that podcasts exist for networking that's kind of…,
CONNOR: It's not theory. You're just a marketer.
EVAN: Right. Right. Like, I meet great people through podcasting and those are the relationships I get more than the podcast producing, like, a big audience and tons of revenue, you know?
CONNOR: Totally. Totally. It's like, come on my stage and I'll talk to you. And like people are like, "Oh, I like stages." And then you're like, ah ha!
EVAN: Now we will be partners and friends forever. Yeah.
CONNOR: All from the podcast. So before, before we get into anything maybe a good place to start is like. What's your background? Why are we, why are we talking to Mr. Evan Dunn? I met you when you were at Syncari but you have been doing cool stuff in the tech universe as a marketer and a product guy for a while. And so we can start wherever you think it makes sense.
EVAN: Yeah, you're the one who asked me, so I should ask you why you're talking to me. No, I, I have been in marketing for over a decade now, which makes me feel really old. I, I didn't know, I never really knew what I was doing, but I had the good fortune of landing like Macy's and Verizon Wireless as clients through Twitter. By reaching out to them back when marketing was growth hacking, if anyone remembers that time and now sales is growth hacking. And that's what we're going to talk about when we talk about GTM with AI. But the, that consulting job eventually pivoted into an analytics firm that eventually pivoted into an AI product for media, working with HBO and Disney, Sinclair broadcast groups and really cool things we did. Which I know we're going to talk about in a bit.
CONNOR: What year was the AI stuff? Oh yeah, like, I dont want to date you.
EVAN: No, it's all right. It was like 2017 to 2020 basically.
CONNOR: Pre AI coolness.
EVAN: Pre-AI coolness. We, half the time we called it machine learning because was like kind of cooler in some ways and sounded more robust, but now I feel like most of the time AI is mentioned without machine learning in the loop and there's AI that's not machine learning now so like, It's just a different Wild West, still a Wild West. And then joined Convoy and it's a performance marketing. I actually was going for a product marketing role coming from a head of product role at Resonance AI and they asked if I had WordPress expertise and I said, "yes." So all of a sudden I was in digital marketing.
CONNOR: You're you're you're the guy that does the website. Yeah. As the digital marketing guy.
EVAN: That's basically what it became. It was a good time. We had a good run of it. In fact, I left before Convoy, Convoy's Zenith time period was over. So it was still mountaintop experience. And if you, you know, Convoy, if you're not familiar listeners, is gone completely. And if you still look for load board, which is a 14, 000 searches per month term, they're still number two. This is why SEO is really important because it lasts for a long time.
CONNOR: Trust Evan.
EVAN: Yeah, right. It's like spam. You know, you put it in the basement, you come back when the apocalypse hits, it's still there ranking number two on Google. That's a weird analogy. Never used that before.And then joined Airwallex as U. S. growth marketing lead, really struggled with product market fit in the U. S. And And so that was a brief stint we can talk about too, and then joined Syncari for a year and a half, really great stint, really great product, and then went on to a head of marketing at ServiceBell, which is where I'm at, at now. And I think the themes that make this a fun conversation for me to have, Connor, I've always tried new tools, new tactics, new tech, I'm kind of a junkie for it. I think a lot of digital marketers are. And I've just seen so many interesting things now with prospecting, the role of the SDR, sales and AI and AI for everything. There's it's kind of hard to wrap your head around it and parse the themes out. So that's what I love to talk about later in the show.
CONNOR: Yeah. Well, let's maybe start with, with resonate something that you had said, I don't remember if it was in the, in the pre or the now, but you were sort of talking around like, Hey, we did this thing and it didn't maybe work or where it was. And. If that was 2017, 2020 what, what was it? And if you're, what do you think is either?I don't know, would it work now or, or is it just completely off base and was it ahead of its time? Or, I don't know, what's your perspective now?
EVAN: It was very expensive for us to do what we were doing. So let me talk about what it was. And, so early on, we did some work with Disney where we had social listening data. Mentions of shows, over many periods, I got, I became kind of an expert in social listening technology, which is also a category that's largely disappeared.
It was very made possible by Zerp you know, economics, where you're paying 40 to a hundred thousand dollars for a tool that just scans the internet for keywords, basically. And we combined that with like viewership data, right? Like how many people, households are watching a show across its lifetime.
And data about the content itself. So Disney piloted this, and then we landed a contract for the same kind of thing with more machine learning driven tagging mechanisms on the content for HBO. I can't go too much detail...
CONNOR: Sure. Sure.
EVAN: On engagement, but we were telling them why season 2 of a really popular show failed where season 1 was super successful and it was it was whenever all things held equal, and I'll talk about that in a sec.
It was when this one character, one of the protagonists, was on screen. People left and we're not retained. If you think about watching a show, it's not so much like, adding viewers, people don't hop into a show, there's no like, you grow a show through the content, it's you keep them engaged, right, there's drop off, there will be disintegration, if anyone's done video analytics, right, you know that it's how much drop off how many viewers were still present at 25 percent and versus 50 percent and 75 percent and completion is never 100%. It's 90 percent in, and all that stuff. And so we were tying that to 30 plus proprietary algorithms. Shout out to Will Henderson Drager, who helped make this vision become life. We called it a resonance score. So we would look at the color, the darkness, the transitions, characters on screen. Will and his team built the most powerful character recognition that I've still ever seen. Where if someone's back was turned, you know, it'd still track them across the screen. If
CONNOR: But when you, you're like, like literally the character in the show is what you're referring to.
EVAN: The character in video. Yeah, like the raw video. So we call true color footage, right? Like, the final output, right? And really complex stuff. And then you, you essentially render the content analytics, you know, many, many rows across the, the duration of the show and pull in two sets of second by second household viewership data and then check for possible influencing factors, massive algorithm, right? To essentially find where the drop off was. So we were outputting this and this is one of the first things I learned about AI. No one cares about the AI. They care that you can do a lot of things at scale very quickly, less expensive than hiring a bunch of people. We were innovating on top of like what
dozens of little you know, apps with people tagging stuff in them would do where it would like load manually tag stuff and have all these quality checks on time. These tools still exist for tagging training data sets for for algorithms, right? But we were trying to basically automate that entire process.
And the first problem is like, you can come up with these amazing, like, you know, our statistical analysis charts of, of influence and confidence, and no one's going to care about those be like, well, what should I do about it? Right? Like, should we kill the show? Should get rid of that character? Kill him off the next episode, this kind of stuff.
And this, this in TV world is called content strategy, right? Now it's pretty much all driven by gut. Even at Netflix, a lot of the content strategy. It is still just, do we think this is a show worth funding? They get lauded for House of Cards, which is the first time say that AI, like, picked up a show and promoted it.
But really, if you look at the Netflix engineering blog, AI is really mostly used for, like, recommendations of similar types of content. Surfacing clips, even though that produces some really lame trailer type videos in their app. Anyways, just a couple things.
CONNOR: Can you even still enjoy media now? Or are you just like, I worked, I worked at the sausage factory and I think it's
EVAN: Yeah, man. I mean, a great show is a great show. We loved we're watching Julia on HBO right now about Julia Child. But Mad Men, man, some, you know, some shows are just truly there's a lot of junk out there. And you do wonder, like, Netflix makes a lot of ripoffs. Like, are they using AI to analyze, like, oh, that's popular, let’s, What are the themes in it? Let's steal it. Anyways, so then the final product that, that we ideated, Will and I made was this resonance score negative 100 to positive 100 for influence, either driving people away or keeping people watching for every attribute from the algorithms, every character, every setting, every theme. And then we would take that to HBO to the customers and say you know, "this is what's happening in your content”. Ultimately, They didn't really take the recommendations.
CONNOR: Didn't care.
EVAN: Well, and like, what are we saying? Like a machine told you this is what you should do, right? Like it's really hard to base 5 million, 10 million, a hundred million dollar bets.
CONNOR: Do you think that that's like, is that a byproduct of the trust in the technology at the time, or is that a like a human condition aspect that you think is...
EVAN: I think it's both. And that's such a good question, Connor. I think it's both First, it's a by product that's still there. People are only making tiny, tiny, tiny risk averse bets on AI right now, make my email language and I'm still going to quality check it, right?
CONNOR: It's low risk stuff. Like, that's, that's what we see in any research we've done or any customers we work with is everyone's like, look, I really want to try this, but like, let's maybe like generate landing page copy. And you're like, what about, I don't know. A sales bot. And you're like, that really freaks me out.
EVAN: Yeah, absolutely. Like AI chatbots get deployment on, on tool teams that are more risk friendly because they have nothing to lose. They don't have a ton of traffic. They're not a big brand. They can't sacrifice brand equity. They have no brand equity to sacrifice. Yeah, I think that's a really good point.
I also think it is totally a human condition, like really well put. Like we still need to feel like we're in control. So what AI needs to do is surface options and say, here's what we think is going to happen to your pipeline. If you take this audience on and drop this audience, right? Macro level, things like that.
There's a huge opportunity for AI to summarize and recommend at scale. To a human operator, a human strategist who then takes that as input. But to, to prove that out, you're right. Like we've got to really nail the confidence scoring mechanism to surface the confidence scores to the end user. Right? Like there's too much smoke and mirrors.
ChatGPT has a lot of things. It's not transparent. So, yeah.
CONNOR: Okay. So, so previous AI experience didn't, didn't really go anywhere because of product and is it just like, Hey, this is super cool, but like we don't really care.
EVAN: Yeah, everyone fascinated. We were all, even as we were building it and our clients were all on the edge of our seats for like, what's this going to be?
CONNOR: How does, how does this bias you against what, like, whether it's real or not, like the very material hype train that has left the station on AI.
EVAN: such a naysayer. I mean, there, there some things, there are some things that will absolutely you use it for. Like one of our competitors at Service Bell is Drift. And I went to, and they have a lot of negative reviews on G2. So I summarize them with AI and I use that in messaging, right? It's great.
CONNOR: That's a that's a great strategy, dude. We should, we should expand on that. How did you do They that?
EVAN: There's an ask CSV tool that lets you import any CSV file and it'll summarize whatever column you choose, with AI. So, which I, yeah, it's, it was very simple, very lightweight. Yeah, but the, the, there's such a problem. And I observed this early on last year when ChatGPT was getting big and quickly I found like perplexity at AI, which provided links to the sources it was referencing.
When I would ask it a question or prompt, it would, it would return with a summary and where it got it from. So I could double check. And immediately was like, well, I'm going to use this and never go back to chat GPT because paper trails, man, like I need to know who's, who's, what the ingredients in the are.
CONNOR: We need like CYA dot AI. I'm going to go see if I can buy that domain right now.
EVAN: All right. I'm that's but that's really good. Yeah. CYAI cover your AI. Yeah, I think there's a lot, and there's probably stuff I'm not thinking about, but I'm sure product leaders and AI founders need to be the forefront of, of AI transparency and trust, not I, I'm not so concerned about privacy stuff.
Don't quote me on that, obviously, but, but like with AI, like what, yeah, with main concerns I have are like, are we going to start giving C suites these dashboards with AI recommendations and they go shift the company and then whoops, the training data was bad. CRM training, CRM data as training data.
CONNOR: Is that different? How is that at all different than, I think it was like, I think I'm stealing this, this idea from Mark Andreessen, who did a Lex Friedman interview, which I highly recommend, but and they were talking about like rockets and who decides if the rocket, like, you know, does it explode or not, basically, which is much more high risk thing.
But, but I think the concept here is the same. I would argue to your point that that's not at all different than, like, the analysts who set up this data and who put it together didn't normalize. It didn't do these data sets, didn't have it. And now there are people making decisions off aggregated data and that data wrong. And...
EVAN: Now we're talking about the problem of data teams generally.
CONNOR: Sure. Yeah. But like, is the AI better at being a data team?
EVAN: If you train it to check for inaccuracies and discrepancies and give it some contextual clues, this is where the human resource needs to go. It needs go towards quality checking at the end, but also training set.
Optimization, right? So how do I give it ample context clues? A little bit of, of tagging on a training set goes a long way. For instance, we, when we did our character tracking, we would do human in the loop on about a hundred times on each major character, meaning yes, no, this is that person, this is this person.
I don't know who this person is. And it would feed into we do it on one episode for a 10 episode series, and it would then be perfect across the series. There are, there are analogies for that today. We're like. CRM data, like just go clean up that, that, deal notes field or, or give your own like prompt for how to interpret, here's the jargon we use, here's what these terms mean. Give it some context. So it doesn't summarize based on, because right now when you deploy an LLM or like a chat GPT type thing for your specific scenario, it's trained on something else. You see this with images, right? We're like, how does it make a new image of a person?
It has billions of images of other people. That's what it's using. So those images are not what you want, you know, and take that to text, take that to data and recommendations analytics. If that training data isn't, doesn't have enough of the signal that you want, it's not going to give you an AI version
that's better. It's actually going to be worse because it's going to be steps removed from the already bad data. So you're just rabbit holing down the wrong path.
CONNOR: Yeah. No, I, I totally agree that something I say often is, is I think I, when I was at a marketing ops conference thing last year and we were talking about this and, and it was always this question of like, oh, well, this means that. Do the ops teams and the people that are managing and building all these stuff and like, they get replaced.
I'm like, no, like, if they screw up, it's really bad now. Like, it's, it's not like, oh, yeah, sometimes that feels kind of weird. And like, we changed it then. And but like, just if you do this one thing, it's fine. Don't use that report, use this report. And instead now that's like, AI has no clue. It doesn't context.
It doesn't know that like, you know, Greg built a crazy formula and he hasn't worked here in three years. And every created under Greg's tenure is like off by whatever. And we normalized it. It just knows that the data set is the data set and here's what it thinks you should do
EVAN: That's right. And it has no way of knowing. And that's, you know, I think you're actually articulating the biggest gap with AI and opportunity. And, and this applies to sales AI stuff. We'll talk about in a bit, right? But like crawlers plus AI is huge. Crawlers plus an LLM summarization capability is huge.
And I say that because you have a CRM. It has thousands of fields on multiple properties, right? An inability to scan through those. Obviously, you're beholden to HubSpot Salesforce limitations and what they let you do. But if you can pull all those out, scan them, look at the names and what's in those columns, those those fields for similarities and say, "Oh, it looks like there's a couple of redundant fields on your deal object.
It looks like there's a couple of redundant fields in your company and your contact object, do you want these to be combined human?" And like basically human steps for providing documentation, then it spits out like a documented CRM that you then deploy right towards training. So like, can someone do this duly...
CONNOR: I think, I think that's going to happen. I think the CRM vendors have to do it. Like, I think that that's I don't know. I get questions all the time by basically CRM market analysts who are we'll pay you and ask you what you think. And I, I tell them and I'm like, I've, you just, I'd have a beer with you.
And I'm like, I love telling you, come on my podcast, but sure. But I think to your point like, I, I I think there's this sort of thing of like, Oh, well, what can HubSpot like can HubSpot add this AI
functionality and what could they charge for it? And could Salesforce add this AI functionality and what could they charge for it?
And I think that that question misses the actual core element of what's happening, which is, I think that this just moves into like, this is a standard expectation of what the software is and what the software does. And if you don't do that, and you are at your core, a software that is driving decision making or making recommendations, like you're useless, you're dead and you just lose.
And I don't think this is, it's kind of like, I don't know, maybe there was a time that people were like adding, no, this was happening. Remember when people would charge for API access, that was like a thing that you would have like a SKU on and you built like no one does that now ever because it's an expectation of like, and I think that this is going to be the same thing where that just moves to being baseline expectation.
If you don't have this, I won't work with it.
EVAN: Yeah.
There are analogs here too with Google Analytics, Google AdWords. been doing AI recommendations for a long time. Never thought they were good, and I never took them, but I'm actually not the majority of their users, right? It's always good to remember the majority of your users are not actually the domain experts,
right? Like they're, they're just people who got handed a tool they're not familiar with and are Googling their way into. So if you can, yeah, if you can circumvent their need to go externally to find guidance and give it inside your tool, absolutely. Yeah. HubSpot could absolutely. And you know, they would be the one to do it knowing their leadership, but,
if there was, and what we're talking about here really is AI co pilots, right? Like essentially increasing adoption by, if you have a platform that has many things you do with it, many possible things you can do with it, increasing adoption through a chat and recommendation interface for figuring stuff out.
It's no different than what AI things have been doing on the help center, but it's kind of like also looking at that person's installation and, you know, where is it at? What have they done? Have you noticed mismatches or missing fields that seem important, required fields that no one tried, wants to fill in, you know, these are all things you could help people navigate that would be huge to your business and theirs.
CONNOR: Where, what, so what are you working on now? Or where, where are you deploying AI? So I think the thing that's really interesting, right. As you talked to already of like, Hey, I'm a tinker. Here are the things that I've, and maybe you don't need to bash anything, but like, here's, here's what I've done. And what I thought was sort of like, here's where I feel like it's BS versus what are you actually using?
What are you excited about? What, what have you used that has been disappointing sort of being out in the field and, and experimenting.
EVAN: Yeah, there's there's a lot that I'm really excited about and I have to give a shout out to clay.com because of two things. They have claygent which is like your AI co pilot for Audience research, messaging development. You tell it a command to go scrape, you know, websites that are like XYZ and it comes back with results.
This thing is a little trippy and how powerful it is. And we're very early days still, but even before Claygent, you could incorporate ChatGPT for some really interesting commands. And so one of the reasons, that we're talking is, is I actually built a phone validation waterfall myself in clay.com so that it checks two data sources for phone numbers, sees if the phone numbers match.
And so I asked ChatGPT in the loop inside of clay, inside of the table, do these phone numbers match? If so, don't say anything else, just output the phone number that matches. You kind of have to coach it to shut up, right? Or else like, " w we found matching phone numbers like this". And like, I don't want to have to remove all this.
CONNOR: No, no more information. Just, just, are they the same?
EVAN: Yes. Honestly, this is one proof that AI is...
CONNOR: I love the idea of like the AIs are just gotten really mouthy and you're like, "stop talking".
EVAN: It's, it's very much the case.
How hard it is to get them to stop talking. Like I like five different versions of the command in there and it still says like output colon phone number. I'm like, "no."
CONNOR: You're like, no, no output, just phone number.
EVAN: You're making a whole nother workflow for me. Export, Google sheet, find, replace.
CONNOR: That's really funny.
EVAN: Anyways, so then I, and then I bring in additional phone data vendors, right?
And it keeps checking in a waterfall and it's, it's difficult to build. I actually found fullenrich.com. Shout out to these guys, they built it. That you basically go and buy, subscribe to a waterfall credit paste and for email or phone seven different sources just on a simple credit system, very easy to use.
Already testing a list with one of my SDRs, as we speak.
CONNOR: And you're, you're using these for lead gen for sales team with accurate data segments, like all this kind of stuff.
EVAN: Exactly, yeah, exactly. Phone validation, if you're not familiar in the wild listener, is basically don't call every phone number because that can lead to a lot of bad things and a lot of wasted time. Find the phone numbers that are most likely that person's phone number, or even for sure that person's phone number.
And typically what I've seen doing this validation is it's really good to look at multiple data vendors. And if they all report the same phone number, then they've all got high confidence. Some of them are referencing the same data sets since you want like four or five, six data vendors in the loop.
Yeah, so that use case is interesting. I'm really big fan of account research tools.
CONNOR: What, what's that? Wait, I, I, okay. So I don’t. I don't do any outbound at all.
EVAN: Don't need to
CONNOR: I have. I have. I'm not, I'm not, I don't know if I would be like, I'm like a heart. I think, I think for, for Aptitude 8 specifically, right? Like we're, we're B2B services. We're predominantly higher ticket. Like I compared a lot to heart surgery.
Like it's really hard to prospect for heart surgery and kind of weird. And so instead we focus on like, who are the physicians and we sort of think about partners and ancillary stuff and everything else. But for hapily stuff I don't know that I would say that we're like anti doing that ever, but it's been a long time since I've done that.
What, when you're thinking about sort of adding AI into that prospecting data creation loop, what is the primary value output that you're getting? Is it the time it takes you to do the workflow? Is it the quality of the data? Like what is, what is the thing that you're like, here's what I'm unlocking by
using AI in this, in this flow.
EVAN: It's no different. You know, one of the things we should talk about, it's no different than automation improvements ultimately, right? It's, it's cause it's the whole problem like, should I spend an hour building this special AI workflow, automation workflow, because it's always a mix of AI and automation, right?
When I have a step tool, like Clay saying, do this, then this, then this, right. IFTTT basically piloted this stuff for all of us back in 2014. And versus like 10 minutes of checking a hundred numbers across two data vendors to see if they're the same. Sure, I can do that one time, but then if I have to build lists multiple times, I really want that system.
I do think there's a risk of over systematizing things here, right? Where, and so that's why briefly, like, let's talk about how AI is not a category, right? You have to just, and what's frustrating about this is people treat it like one now. I get questions like, what's ServiceBell's AI roadmap? Like, well, what do you mean?
AI in chat, AI dynamic videos, AI
CONNOR: What kinds of software are you guys going to build?
EVAN: That's exactly the question, right? Totally, founders who are in the AI space really fall victim to this, thinking they are producing an AI product, saying that to VCs, saying that to customers. No one cares. People care as far as they look at your website and talk about it on social media.
That's it. That's why hype cycles happen, right? But buying, like literally, if you pictured like AI.com like what does it do? I don't know. It could do anything. But, but it can't do everything and it takes training and wisdom. And we were talking about the data input problem. So if the, what systems should I build as systems versus keep them in manual spreadsheet format, right?
That's the question I think for marketers like myself, ops people like you. To really grapple with today. One of the ones that I think is really exciting, the phone validation one, I think is big, but I think not, not everyone is a cold caller and all that stuff, but one of the things I think is really exciting for ops pros everywhere,
rev ops, GTM, ABM is account research. Everyone I see who's succeeding in the like, sequencing, automation and AI in the loop for for prospecting is automating, is building system for account research per segment often. So let's say you want a company that was founded. Before 2010, because that means there's a certain cultural milieu you want that they've raised around recently, or that they've done layoffs recently, like AI can get this for you with a little bit of help on where to go look.
Or you combine like virtual assistants, looking at the LinkedIn insights tab on the LinkedIn company page with AI summarizers and all that stuff, and you're a thousand dollars into, a you know, a gold mine of, of insight that you then hand for your 2024 target account list and boom, you've got something really strong.
So think about like, what are the trigger events, the really specific things that are happening in your potential customers or in your customers and working with, Jordan Crawford is everyone should be following his material on AI and content. And we're talking about a new product he's working on where you summarize, job description datasets for existing customers.
Did they open up a job for a title that should be a user, or should be an owner of our platform? What's in that, what initiatives are they going to be responsible for? Did they recently close a job? For some, like, think what you can do. You could basically trigger like automated workflows for upsell, right?
And companies across the spectrum are leaning into this. I'm talking with the chief of staff at a 1200 person open source software type type company who is trying to bring some of this innovation in and she's launching experiment pods of SDRs, like two SDRs and Dimension Marketer who are just working these new ideas so that she doesn't have to turn the whole ship right away.
And I think that's brilliant. So, so
what are the systems?
CONNOR: I want to I want to expand on what you just said, because I think that that is actually, I think 1 of the number 1 things that we've seen in some of these conversations and research we've done around this is the biggest obstacle as people being like, how do I start? Like, where can I go? And like, moving the whole ship is really, really hard.
I think what you just highlighted is if you're in a larger organization, the ability to pilot this thing with a smaller group of users is so valuable. And without plugging too much on CRM and RevOps and the rest of it, but being able to have the nimbleness in your system structure. To go and carve that out and be like, we're going to have these three users,
we're going to have them work in a, and you don't necessarily have to air gap it, right? But like, we're going to have these users do a different process that's segmented from what everybody else is doing. Your ability to experiment with that and iterate against that. I think, especially as things move really, really quickly becomes a massive competitive differentiator.
So kudos to pulling that off at 1200 person organization is a lot of things and nimble is not one of them is my assumption. So that's huge.
EVAN: Absolutely. And, and that's so yeah, I would say the two things I want people to think about with AI. One is like, what should I systematize? And two is how do I manage it? And that's where experimentation frameworks come in. And that this conversation I was having with that chief of staff basically is what was spawned
that experiment pod kind of concept. Basically, what I've started doing is I don't really use marketing plans. I use experimentation frameworks as far as pipeline growth and acquisition is, is concerned. And so I work with sales and SDR and, and, run the SDR team now at ServiceBell. But we basically set up like, here are the top hypotheses we have, right?
And AI could be in there. It could be in there for summarizing. And finding the right people, summarizing the company info and all that stuff, but at first it doesn't really need to be, we're going to go prove the concept and then we're going to figure out the system to scale, right? We'll see what has teeth, what has legs, whatever analogy of the body parts you want to use for longevity, right?
What has strong hearts, robust, no heart surgery needed and sticks around long enough to say, Hey, we should scale this up. Look, we got two customers, five open ops and 20 conversations that are positive out of this test in the last month. Well, let's put some more dollars there. Let's run ABM ads alongside the SDR outreach, right?
Like think in terms of like, what are your, like, maybe one, two, three scale of emotions. And the third one is like full on AI. Let's do this forever. Super scale, right?
CONNOR: I want to, I want to expand on that because what you just described and I think about in the beginning of A8 and then I was running all of our RebOps services work that the, it was just super interesting because we got so much exposure to so many different GTM teams and GTM leaders and the ways in which they would do things.
And what you just described is. In my opinion, the correct borderline only way to figure out new demand gen strategies. And I, I think so many people don't get it. And I know that, that hapily when we've gone through a couple of different, I don't know that we've churned a whole bunch of GTM leaders yet, but we've had a bunch of people come in and do stuff.
And I think that if you are anyone in any go to market function, and you are experimenting with new demand gen methodologies or frameworks trying to figure out what works the absolute best way to do that is what you just said. And it's like, you can get fancy and you can build CRM attribution. You can do all sorts of stuff.
But at the end of the day, you can have a sheet that's like, here is each thing we're doing in column a here is how much a very top of funnel generated and B like C D E as many and far out as you want to go. And if you do that, and you update that, and you present that to whoever your boss is on an ongoing basis, and you're like, we are learning stuff.
I think the whole conversation around like, well, should we be spending on this? And what should we be doing here? And like, does that make sense? Like that is the answer. And I I would just say for anyone who is a if you manage a marketing function, whether you're a CEO, founder, whatever, like ask for that.
That's the best thing that you can get. And if you are managing a demand gen function, do that thing. Like, and don't get distracted by all of the pieces that I think feed into and build that.
EVAN: Yeah,
CONNOR: It can be totally manual. That's fine.
EVAN: Connor, you're the first person I've spoken with who has described the sheet exactly as I have always described it literally I just recorded a podcast.
CONNOR: Every time I talk to you, I want to work with you. It's the reality.
EVAN: Well this that's the thing is is what you're doing, imagine this right you have five salespeople and you had to lay off 20 percent headcount last year, really common scenario, right? And you're like, well, where are our best customers? We kind of have info about them, let's pull out records and notes and CRM and like of our hundred, 200 customers, you know, even 500, what are the 50, 10 top ones look like?
What do they do? Okay. Well, we found a few different threads of attributes, let's go test those. And we will test in order to fail fast, meaning salespeople, this isn't like you're on the hook for making each of these work. The CEO and Founder can't come down on you and say, "well, you didn't, you just don't know enough about that segment."
It's just like, no, we all agreed with this was an experiment. The point was to figure out what not to pivot into. The first time I did this with this was a convoy. LinkedIn ads for summer running and built a whole business case used rice scoring to argue for prioritization of different test initiatives, and we ran a spreadsheet of the spend and the opportunities created and conversion to customers for each of the experiments in LinkedIn ads, then it air wallets, no product market fit in the U.
S. They hadn't done any research before launching in the U. S. Back in, 2021 Midway. And, the, the founder CEO would not let me win the argument that we didn't have private market pit cause he wanted to have private market pit. Well, so I said, okay, eight industry hypotheses. Bought validated phone data for one SDR, two months of this work, he's shaving into three industries that are really working and three that are really not, and two that are ambiguous.
He's booking a meeting a day. And then he's booking two months later, five meetings a day. This is what experimentation can do for you, is give you the, the, finally the lens to focus on the growth avenues, the critical paths that you know are there, you just don't know how to find them.
CONNOR: And I think to your point,
EVAN: It's gotta be a sheet.
CONNOR: Yeah, I mean, I think to your point, and I, I, I don't know, I think at one point I probably had a title of growth hacker at some company sometime, but I think to your point, right? I think, I don't know if it's that origination of, of approaching a lot of this, because my, my way into ops was, was entirely through sales and then marketing and then into the ops piece.
And I think that the thing that like, people miss is that ultimately, I think this is especially true in marketing. I'm thinking about whether or not it's true and everything else, but at minimum it's, it's true in demand gen and marketing is like, it goes into two functions. You are either experimenting and trying to find
new channels, new methods, new strategies, new ways of, of doing things and, or like does anyone want this at all? It's like maybe even an
earlier point, right? Or you're, you're optimizing and building a lot of those channels. And I think that when I look at the, and marketers have historically low tenures, right?
And I see a lot of, and I think there's a lot of founders and CEOs that don't have a marketing background. And I think what ends up happening is, you are looking for somebody who is an experimenter and then you, you want to judge them by their ability to manage and optimize and or you're hiring somebody who is a manager and optimizer and they've never done the experimentation framework.
And I think the most seasoned and long tenure and successful marketing leaders that you're going to hire are people have a lot of experience in managing and optimizing existing things. And they don't know the first thing about building and testing and managing all of those components. And I think that's why you see a lot of executives, but it's like the VP marketing loop, right?
EVAN: Yeah. Well, and I think that's why demand gen has stuck around is because of that, that agility and experimentation is pretty native to demand gen type.
CONNOR: Do you think that's a different function? Do you think that there's like something like a a framework I talk about a lot when I think about leaders on our team and where it's been and in 2023 A8 was sort of used to like doubling every year and 2023 it was like a little bit better than flat, but like, definitely not at the same pace
EVAN: It's a tough year.
CONNOR: Tough year.
And I think I think a lot of our leaders are builder leaders like they love to build and adapt and create something new. And like, that's where they thrive the most in that, like, structured chaos environment. And this, like, manage status quo was just so much an antithesis to what they love to do and want to do and can do.
And I think I've started to think a lot about people that I hire and work with and leaders that we build as, are you somebody that's going to go and be the builder of the thing? Or are you the person that is going to be the manager of the thing? And there's a different time and place for each one of those profiles on any individual function. Do you think that that applies to like marketing and demand gen? And like, these are different things and it doesn't mean one person can't. Do or be good at both, but are, are they fundamentally different practices?
EVAN: Yeah, they totally are. I think because experimenting is like. Lab coat, black, dark room, you know, chemistry by yourself.
CONNOR: Might explode in your face. Who knows?
EVAN: Totally. But you gotta be given permission to go hide in a wormhole and come out with meaningful stuff. And you gotta be given budget to do that. Now, now, at bigger companies it should still be 80 percent of your budget is like the stuff you know works and 20 percent is always experimenting or the stuff you know works stops working in two years.
That's everyone's paid budget in B2B SaaS. We don't talk about that enough. Like what happened to paid budgets? Like they vaporized for smaller B2B SaaS. First thing to go. Cause then they didn't have to cut headcount as much. But, the, the manager leader, the, the, the person who can see, we've got some working things and I need to be able to coordinate across working things and non working things.
And a lot of different people is, is a, is a soft skill, people, person, project management mindset. And that's just different and you need it. You need stabilization.
CONNOR: The chaos to those people, right? Is the antithesis of like, oh, man, if we're trying to do a lot of unstructured things in our system, like, the system's not going to work anymore. And that's like, very bad.
EVAN: Yeah. Like a good, a good analogy for the question you're asking, I think is like, at what point do you need a CRM when you are right? Like when you're starting out a company. Obviously once you've got some amount of like customers to, to manage and relationships, to manage, but it's not right away, right? Right away is I need to go have a lot of conversations and I, I could keep a spreadsheet of them or I could use a free CRM or something like that, but at what point do I need like HubSpot level? Salesforce? I know HubSpot is a free...
CONNOR: but no, I so here's what's what's interesting that you just highlighted for me as I think about. In the very, very, very early days of HubSpot CRM product before I think it was not even called. It was the original sidekick product. And then they started building a little bit of that initial CRM stuff.
I was doing tons and tons of early stage startup consulting. And what we would basically, I would always tell people like, "Hey, you need something, go set all this up". And I would talk to these CEOs that are like, "Our, our investor told us we need Salesforce". And I was like, "no, no, no, no, no, you don't, you don't know anything,
you don't know anything, you don't know your business process. Like you don't know anything." This is not a good plan. It's like, go put everything in HubSpot. It'll track your emails. You'll kind of know what's going on. They have this little deal board. Like, that's all you need. You don't need anything else.
And then at some point they'd be like, "Oh, wow. We have like a manager and we have a business process and we could like explain it to you" and I was like, great. Now we'll go move everything to Salesforce. And that was like the main thing that we did in our original exposure and why we were so prolific at the beginning of HubSpot sales product is like, we were, it was basically just like Salesforce starter for everybody is what we were using it as.
And then I think that HubSpot and I, to give credit, credit and credence to their product team, I think that people think about. HubSpot starter professional and enterprise as being like for different size and scale of, of companies or sort of a function. I think that's actually the wrong way to think about it.
And it's much more like, where are you on your life cycle of like your GTM maturity? And I think the best part of like, HubSpot starter and some of the basic things is like, look, you don't need a lot of automation functionality. You don't need a lot of customization. Like you just need a place to dump all your data and manage it. And we have a product for that and it's, it's free or it's basically free.
EVAN: Here's what people miss, right? Is if you deploy something like Salesforce, when you don't have a known and delivering GTM with some established processes and people to admin and, and, and operate it. You actually do the inverse of helping you hurt everyone go ask any demand gen marketer Who's had to work with Salesforce in a company in a unicorn?
I've worked at two unicorns both had infuriating Salesforce instances and in both we were slower than we should have been because of Salesforce, right? And sometimes that happens with HubSpot, but I can actually go change stuff in HubSpot myself I can't do that in Salesforce. Like I
CONNOR: No, I think to your point, the thing that makes people, when they talk about usability or they talk about something, I think what you just described is, is the actual component, which is, do you need an engineer or a technician to make an adaptation and make something change. And if you don't, then you and I, we're going to loop it all the way back to AI.
I think that this is what a lot of the AI functionality does is it actually decreases the level of skill required in order to administer and manage and do things in different business systems. And then as a result, as you lower that barrier to entry, you increase the speed at which things can happen because the person who feels strangle holded by the system wants it to change, wants it to adapt, doesn't need to go and like find somebody with expertise who knows what they're doing, who can manage the thing and like actually edit it. Instead
they're just like, yo, I want this to be different. And they do not need a high degree of technical acumen to actually make that happen and shift it.
EVAN: Yes. As long as we're talking about the AI tools that spoon feed you the like inputs and outputs, right? I totally agree. I do think it's interesting. There is an analogy here with AI, like clay. com is actually pretty complicated once you get it going and doing high powered stuff. So there are people, Jacob Tuiner, who's part of a service bell, but also his own consultancy.
Shout out to him. What's it called? Sculpted. Because clay. Where it's managed Clay implementations. And so if you need really powerful end to end, like account research with AI and then deploying emails out with custom messaging,
CONNOR: Wait, this is a guy who's doing like a clay agency thing. Okay. I would introduce me to him. I would love to talk to him.
EVAN: Yeah. Yeah. He's, he's great. He'll definitely help. There's Eric Novoselovsky. He's doing stuff like this. Kellen K Spear, who's big on the experimentation frameworks. Brigitte Ruha, Scott Martinez. There's a lot of these people who are doing it, but, but they will be the first to tell you, like, It's probably less expensive to go buy their retainer than it is for you to like spend the time on your best people and learn
CONNOR: I think, so I think to your point. And this is what I've told the people when I get asked of like, how are you worried about AI? Like you run the services business, like it's expertise, like that's what you're selling. And I think what you just said is right. I don't think that there's, and I think that this goes back to AI AI labor and what it does to economies and everything else is like, you do not vaporize the value of expertise.
You just move the threshold for expertise up. You move the requirements of domain expertise to achieve anything down and what you end up actually unlocking on the other side is just net higher productivity for both. I'm going to use the word class, which feels wrong, but maybe also really prevalent.
Both classes of people that are, that are doing sort of the high level expertise, high level of customization work and the folks that aren't. And I think one of the things and, and that A8 is really looking at is, is how do we go and build a lot of those pieces? And, and that I think is becoming more and more core to the work that we're doing because.
Almost everyone is trying to figure out how do I pull AI into the overall strategy that I'm doing and how do we do that? Well, which is not the same thing as AI, making it easy for people to do basic functions, which I think. The value sort of impacts both.
EVAN: Absolutely. HubSpot, I've got hooked up to Clay. I run my exclusion lists through it. So I can check when I create new account lists, like, are there already customers that they're out at hops, exclude those domains and then move on with my clay prospect. Yeah, these worlds are colliding absolutely.
And I think the biggest theme that I'm seeing from our conversation here is like, you have to think about yourself as a system builder, and in the early days of systems, you want the rawest, simplest form of experimentation that just gets you insights quickly to to validate or invalidate hypotheses you have about growth this year.
And everyone should be doing it. I talked to a lot of founders in my last job hunt about a dozen or so, and would ask them, like, is your marketing plan gonna work? They'd all say "no" I was like, "why do you have it?" And we'd talk about experimentation.
CONNOR: What are you doing with it then? That seems really silly.
EVAN: It's very politely obviously, and then, you know, we talk about experimentation frameworks.
They all love this idea. Everyone loves this idea, right? Like simple sheet projects to manage growth experiments, right? But, but then when you get to the point of like, okay we've got some traction. Now I've got some customers. I need to make this CRM native data. I need to make this a system where AI can solve some of the gaps.
A crawler can solve some of the gaps and just automate some of the pieces so I don't have my demand gen marketer spending 60 hours a month on, you know, building lists basically, or, or finding data and contact info. Yeah, that's, that's, I think the name of the game for any like, seed series a and B2B SaaS this year is you're probably struggling a little bit with where are your clean growth lanes.
You know, what we're talking about, I think is the way to, to figure that out.
CONNOR: Evan, I, just as the last time and every time that I speak to you, could spend hours and hours and hours talking to you, but I only reserved so much of your time. And so, I will do more of this, and thank you so much for coming and sharing super practical insights. Check out Clay. Check out, ask CSV.
We'll put a whole list of of tools that Evan sort of plugged in the notes here. But Evan, thank you so much for joining us. And I'll catch up with you more soon.
EVAN: Thanks Connor.