Taylor and Andrew Debate Cost Controls (We Ran the Numbers)

[00:00:00] Andrew 3: Do you want to start there or do you want to, do you want to yell at me for being bad at business first? What do you

[00:00:03] Taylor: think- No, no, no. Um, we could start with that because I think, I think we're- We could start with

[00:00:06] Andrew 3: what? Yelling

[00:00:06] Taylor: at me or data? No, no.

[00:00:08] The research. 'Cause I think, I think this is a good one for the two of us, and I, that's why I even think it's a good place for us to... We have been waiting to publish this, but I think, you know, you've, you've probably, I'd say, taken over as the cost cap maxi lord. I think so. I think even more so than me.

[00:00:26] Andrew 3: I think so.

[00:00:26] Taylor: I think I've backed off of it so much 'cause I just hate the argument so much now.

[00:00:30] Andrew 3: I, I'm getting there.

[00:00:31] Taylor: But you've picked up the flag, and I think- The,

[00:00:33] Andrew 3: the reason why is that I'm so close to accounts that I see... Like, it just still makes me so crazy to look at somebody's business and look at the dollars being lit on fire.

[00:00:41] Taylor: Right.

[00:00:42] Andrew 3: And, and it just breaks me. I don't actually care about what is the best media buying method. That's not what I care about. What I care about is how to grow your business without lighting money on fire, you know? Right. And I think it's just a big thing. The, the actual position I hate the most is this doesn't matter.

[00:00:58] Yep. Unless there's a data-driven reason to say it doesn't matter.

[00:01:00] Taylor: Right.

[00:01:00] Andrew 3: Right? Then it's like, fine. But, uh, but, like, this doesn't matter, like, go make more creative, is just like, I don't understand why we can't care about both. Yeah. Right. So anyway, you're right. I do think I've been- become that to some degree.

[00:01:13] Taylor: And I, I'd be curious, like, what do you think the industry disposition is now? I feel like the Overton window moved a lot in

[00:01:22] Andrew 3: our direction. Like- Oh, yeah ...

[00:01:22] Taylor: most people conceded eventually to the idea that this makes- Manual

[00:01:25] Andrew 3: bids?

[00:01:26] Taylor: Yeah.

[00:01:26] Andrew 3: Uh- Or where do you think it is? ... I actually don't know if I think that. I still see some pretty loud voices.

[00:01:32] I think the auto bidders stopped arguing about it at some point.

[00:01:35] Taylor: Okay. Um, well, so what I want to say is that, uh, one of the things I try to do, despite maybe my more dogmatic public presentation, is that I just really want the idea to work. I don't actually care about the merits of the idea. Like, I don't want to run people's ad accounts if it's a bad strategy to do that, right?

[00:01:54] Andrew 3: Exactly.

[00:01:54] Taylor: Right?

[00:01:54] Andrew 3: Exactly.

[00:01:55] Taylor: So, yeah. So the, the, the, the way we got to this was not to begin with dogma. It was to begin with like, well, what would be the best way to protect their money and to be

[00:02:02] Andrew 3: effective? That's exactly my point- Yeah ... is like that's the reason I care about it too- Yeah ... is like, yeah, it, and it, yeah.

[00:02:07] Anyway.

[00:02:08] Taylor: So, so with that, like I, I think that one of the experiences, one of the things we've been struggling with, 'cause we've been working really hard to try and, we call it the canon, like operationalize a defined- Yeah ... set of ideology further than we've ever gone before.

[00:02:20] Andrew 3: Yeah.

[00:02:20] Taylor: And you get into really hard questions- Yeah

[00:02:22] about setting bids in

[00:02:23] Andrew 3: particular. It's funny. I'm, I'm in a stage at AJF Growth right now- Yeah ... where, like our big goal right now is document more things. Yes. Make them more repeatable.

[00:02:30] Taylor: Yeah.

[00:02:33] Andrew 3: And the thing, the th- one of the struggles I have with it is that in the act of writing something out-

[00:02:40] Taylor: Yeah ...

[00:02:40] Andrew 3: you start trying to explain what to do in a situation, and then you start thinking through, if you've got real s- experience and thoughtfulness here, you start thinking through all the details of what you would tell somebody to do.

[00:02:48] Taylor: Mm-hmm.

[00:02:49] Andrew 3: And then you're like, "Well, wait." Y- and now you get to, like, you get to details that you have to make a, you have to put, you have to put your pen on paper on. That's right. And there, and I know when I'm doing it that I'm, I'm more or less confident about parts of that- That's right ... at a given time. So it's just really, it becomes really tough, and where I'm like, "Oh man, I wish there was a study."

[00:03:08] Well, yeah, exactly. "I wish there was a piece of data."

[00:03:09] Taylor: And you, you force yourself into ... The further you go on this, the more you encounter that exact feeling you're describing, which is why do I believe this? Like, and how would I actually define it algorithmically such that somebody else could apply it consistently as well?

[00:03:23] So when we say, like, even something like use manual bids, well, there's three different types. Yeah, which

[00:03:28] Andrew 3: ones?

[00:03:29] Taylor: And then ... Right, which ones? And then, okay, let's say I pick one, how do I set the bid?

[00:03:32] Andrew 3: Yes.

[00:03:33] Taylor: How do you build this count structure in light of that, and what happens if it fails? Yeah. Well, what does it mean to fail?

[00:03:37] Like, you, you start pulling on the threads, and you realize there's so much language that we use that is actually absent of any meaning at all, or at least any shared meaning.

[00:03:44] Andrew 3: The other day I told somebody in my company, "Okay, you have to recognize there's gonna be a larger amount of noise in smaller sample sizes."

[00:03:54] This is the, the, the coin flipping analogy I use all the time, right? Yeah. If you flip a coin three times, it's very possible you're gonna get heads all three times. That doesn't change what you would expect if you flipped it 3,000 times. Yep. You'd expect 1,500, 1,500 basically. Right? So you need to build a larger sample size.

[00:04:07] Okay, how large of a sample size?

[00:04:09] Taylor: That's right.

[00:04:09] Andrew 3: Right? So, like, when I tell a media buyer, "You should care about, you know, before you change your manual bid," whatever, how big ... You know, it is just ... And, and, like, when I answer that question, I'm like, "Uh, 50 purchases?"

[00:04:23] Taylor: Right.

[00:04:23] Andrew 3: Exactly. That seems reasonable to me. So- And so that's what you have to do.

[00:04:26] The, the position I end up in is saying make a decision that you think is reasonable, because you're not going to get the perfect truth on all of these things all the time. Yep. But then, but then over time, my hope is that you could be more reasonable. So you guys did two studies on this stuff. Right. And I was

[00:04:39] When you told me you guys did these studies, I actually had two reactions, and I think they're both, they're both interesting. One of them is I was, like, grateful. Yeah. I was ... One of my thoughts was like, "Oh, this is what an agency should be doing." Yep. Theoretically, an agency, uh, the whole value proposition is that we have more access to more data than anybody else.

[00:04:53] Right. But a lot of agencies are actually not providing that good of a service, because they never do the work of trying to actually operationalize thing. Or even worse, they say it's different on every account, which is the worst possible thing you can say. Yep. Um, so, uh, so, so and that was my first thought was like this is, like, why I still look at CTC as the gold standard in the industry about, like, what it means to be a good agency and why I happily refer you guys business when I can't service it and when it's- Yeah

[00:05:18] well, or just when you guys are a better fit or whatever. But then secondly, another reaction I had that I think is worth noting, before we roll out these studies, one of them saying, "Does do cost controls work?" The other one saying, uh, "Should you ever turn off an ad?" Getting data-driven answers with real studies across billions of dollars spent on this.

[00:05:34] Mm-hmm. Okay? Um, I realized something. Before I looked at the, the stuff you sent me, was I thought, "What if I am wildly wrong?"

[00:05:42] Taylor: Yeah.

[00:05:42] Andrew 3: Have you ... Did you think about that when this happened?

[00:05:44] Taylor: For sure.

[00:05:44] Andrew 3: Because you, you and I both have had, have made serious public hay- Mm-hmm ... out of both of these positions. Yep. And I have, like

[00:05:54] It, yeah, like you said, it's, like, become a thing. Uh, like, people associate with me, the bid cap guy. Totally. You know, or whatever. Um, and so even when you told me, like, bid caps versus cost caps versus T-ROAS- Yeah ... like, I have made the statement a lot of times that I prefer bid caps to cost caps. Yep. Now, I've always tried to hedge that a little and say, like, it's actually not the biggest thing that I care about.

[00:06:10] But- I had this thought of, what if I am about to go on this podcast and Large Data says and say, "You are actually really wrong about this." Yeah. Well- Did, did you, did you have that- Yeah ... experience with this data at all?

[00:06:19] Taylor: So this is the weird thing. I, in some ways I feel freer than I ever have about the outcome.

[00:06:24] Uh, and like, and I don't mean that, I don't know how to make that not sound lame, but like I'm post-concern, like in some ways. That's part of what's- Is that

[00:06:32] Andrew 3: because you sold the business?

[00:06:33] Taylor: Yeah. Like, I, literally like if I'm wrong, the cost is actually almost negligible. Like I, if, if it turns out that highest value is like, then o- okay.

[00:06:44] But, but it's sort of beyond the pale of direct life impact where the thing I'm risking is like not that great, you know? Um, so that, and that's could, could just be a total narrative structure in my head more than it's anything pragmatic or real. Yeah. Maybe the cost is devastating, I don't know. But I just don't feel that concern.

[00:07:00] It's like even the way I feel like comfortable talking to customers now is this sense- ... by which I just feel like I can tell you what I genuinely believe and devoid of the constant concern that I know I carried for so long about what it might cost me if I do it, you know?

[00:07:15] Andrew 3: Yeah.

[00:07:16] Taylor: Um.

[00:07:16] Andrew 3: Yeah.

[00:07:17] Taylor: So that's just like a little thing.

[00:07:18] But also, I just genuinely care about being more right if I can be.

[00:07:25] Andrew 3: So, okay. You know? So, so, so I do too.

[00:07:27] Taylor: Yeah. I know.

[00:07:27] Andrew 3: And I, I, this is just one of my things that I thought about was like, well, if it turns out I'm wrong, then I'll have definitive evidence and I will adjust my position accordingly, which I've always said I would do.

[00:07:35] Taylor: That's right.

[00:07:36] Andrew 3: Um, that said, I would feel a little bad about having like maybe misguided some people, though I, though I think the princ- anyway. A- actually, I found myself doing that a lot, which is like- Yeah ... I, I'm, as I started thinking about the possibility this would reveal me to be very wrong about a lot of things that I've been public about, I th- I started trying to figure out why the things I was saying were actually right, you know?

[00:07:55] Even if they were wrong. Right. Yeah, yeah. No. Like, I started kind of playing, like spinning them- Well, yeah ... positively in my head. Like, "Oh, well it still would've been good to like think financially about your business," you know? You

[00:08:03] Taylor: have to think that I've also navigated this feeling a ton as it related to the other parts of my life where I feel like I- Ha

[00:08:08] did this for a long time, right? Like so, so I-

[00:08:10] Andrew 3: Oh, interesting.

[00:08:11] Taylor: Yeah. So I've already worked through the idea

[00:08:13] Andrew 3: that I- Can you tell people what you mean? ...

[00:08:14] Taylor: that I misled people down the path of choosing a religion that I no longer ascribe to, and I, I like vehemently attempted to get them to commit their life to it, right?

[00:08:23] Yeah. Like, so I think there's- Yeah ... lots of ways I've worked through that feeling a ton for myself. And- What

[00:08:26] Andrew 3: is the answer? ...

[00:08:27] Taylor: uh, again, it always goes back to this idea that in the moment, was I acting in integrity with what- Yeah ... I truly believed? Yeah. I think deceit would be- Yeah ... I didn't believe it but I tried to make you do it.

[00:08:35] Yeah. I don't, I'm not- And so I think that- I'm not worried

[00:08:38] Andrew 3: about that. Yeah ...

[00:08:38] Taylor: yeah, this, exactly. So to me, I, I feel okay that what you can expect from me is that you will get an earnest- Yes ... attempt to present- Yes ... what I truly believe in the present moment, and that given new information, I would readjust my position to you.

[00:08:52] Um, that could mean, like, what I tell you is wrong, and you should filter that accordingly, which I think we all hold that obligation ourselves anyway. That I receive anybody's information with the obligation to verify it and fact check it for myself, so.

[00:09:01] Andrew 3: This is actually, this is actually- One of the things I've thought about here, even when I refer business I feel a little bit of this.

[00:09:07] I'll refer business- Yep, totally ... to you guys or to somebody else or whatever. The

[00:09:09] Taylor: outcome

[00:09:10] Andrew 3: is- And then I feel a little responsible if that experience doesn't go well. Totally. And then I have to actually release that and say, "That's actually ... I can make, I can make the best referral I can given the information that I had."

[00:09:20] Taylor: That person still has to make the choice.

[00:09:21] Andrew 3: They have to make the choice. And sometimes I'll even say to them, like, "This is as far as I know, but listen, I'm not in the day-to-day of their business." Totally. "And so you should still run a good vetting process for this." That's right. "And I would encourage that."

[00:09:30] And I, I think that is right, which is that you speak the truth as best as you can, and then you adjust. But the interesting thing also about that, what you're saying, is that, uh, when you are in count- Like, this is a real organizational problem. So, so first of all, if you were to give me this study-

[00:09:45] Taylor: Mm-hmm

[00:09:46] Andrew 3: and then I, like, denied it-

[00:09:47] Taylor: Yeah ...

[00:09:48] Andrew 3: or whatever, even though I was convinced it was true, in order to keep doing my thing, that would be, like, really bad.

[00:09:53] Taylor: Yeah.

[00:09:54] Andrew 3: Because that would be, like, sort of non-integrous. But what it has made me think about is something that I saw you point out a long time ago, and I remember I had a conversation with Olivia at House about this, um, which was, like, one of the problems with selling truth, um, is that people actually don't want it.

[00:10:10] Taylor: Right.

[00:10:10] Andrew 3: They don't want ... Like, many times they don't because the truth often is costly. Mm-hmm. Um, and so, so at the level of, like, incrementality studies, right, like, one of the things that can happen with these kinds of things is it can say, "You've actually been wasting hundreds of thousands of dollars." Yep.

[00:10:25] And nobody wants to hear that. Right. Or, "You should spend less ad dollars and grow slower." Right. "Nobody-"

[00:10:29] Taylor: Especially if you were the person who did it.

[00:10:30] Andrew 3: Exactly. Right? And this is the other, this is the other element of that. So, like, at the organizational level, I think this is a real problem. Um, and, and you have to, like, figure out how to solve that.

[00:10:38] But, but you pointed this out to me a long time ago with some times when we would go in to clients, and we would be the p- we would be the people going to, like, a VP person or whatever- Right ... saying, "Here's why all of this ad spend was done poorly." Maybe they were auto-bidding and they were foolish about the targets or whatever, you know, or their measurement was bad or whatever.

[00:10:57] Then somebody Like, it's really hard if you are going to the, to that VP who then has to go to their CEO- Mm-hmm ... and say, "Oh, we've been wrong about how we've been spending all of this money."

[00:11:06] Taylor: What's the incentive to do that?

[00:11:08] Andrew 3: And so, yes, you point out, like, that will action, that will kill deals. Yeah. Because that person actually can get away for a long time.

[00:11:13] One of, one of the things... So it actually has made me feel more sympathetic to that person. Totally. 'Cause I felt like, "Oh, this could be costly for me."

[00:11:19] Taylor: For sure.

[00:11:20] Andrew 3: And so, so it's made me go, like, "Oh, okay. I, I understand that that position's, that person's in a hard position. It's hard to commit to the truth." But the other thing I believe is that if you are the person...

[00:11:29] Let's say actually as we roll these out- Yeah ... that it, it goes against somebody, somebody listening to or watching this narrative. Mm-hmm. And they then have to turn around and be that person. Mm-hmm. My encouragement is the truth actually does always win- Yeah ... in good organizations, and you will, you will... It will almost always feel really costly in the moment to do that, but actually I think it is one of the biggest in, like, trust builders there possibly is.

[00:11:51] Taylor: Well, so, so I, I always think about this, like, what, what actually builds trust between humans, and I think about this with my kids, is that, you know, there's the classic thing that, like, your dream is that they make a mistake and they bring it to you proactively with honesty, right? Oh, man. Like, that, that is like the ideal state of trust relationship is you would do something wrong.

[00:12:07] You would break the glass in the kitchen, or you would, you know, whatever, didn't do your homework or lose your folder or water bottle, and rather than hiding you would come and say, "Dad, I broke the water bottle." And knowing that, and actually knowing that there would be a consequence, accepting the consequence, but trusting that the end result of that relationship, that would make it more productive not less.

[00:12:26] Andrew 3: You're, you're saying as a parent-

[00:12:27] Taylor: Yeah ...

[00:12:27] Andrew 3: you were, you were- That's the ideal state ... you were hoping for that. One of my f- one of my best, one, literally one of my favorite moments with my kids- Yeah,

[00:12:33] Taylor: you told me this.

[00:12:34] Andrew 3: Yeah ... who are four and six, right, is my six-year-old coming to me and telling me, like- Like, at first he, like, hid that he did something wrong- Yeah

[00:12:41] and then, and then he quickly turned around on it, and it was like I was so proud of him. That's right. I felt like he had been so courageous- That's right ... as a little six-year-old.

[00:12:49] Taylor: And so I think, I think that the truth is, is that as an organizational, as a leader, having a VP of marketing or somebody else who would do the same-

[00:12:57] Andrew 3: Yeah

[00:12:57] Taylor: would actually probably-

[00:12:59] Andrew 3: Yeah ...

[00:12:59] Taylor: build, reinforce trust in the same way.

[00:13:01] Andrew 3: 100%.

[00:13:01] Taylor: Now, you would, you would hope that they would scrutinize their decision and build some sort of structure around it- Yeah, that's right ... that makes them confident that they're not gonna make those kinds of decisions again. Yes. But at the same time, the willingness to admit being incorrect is actually, like, an impetus for change.

[00:13:15] It's, it's, it's a prerequisite to improving your thinking.

[00:13:19] Andrew 3: Yes.

[00:13:19] Taylor: Um, and so I think that the, the kind of people that are actually capable of that are, like, the kind of people you wanna be around.

[00:13:25] Andrew 3: Yep.

[00:13:25] Taylor: Um, so are we about to change our minds on cost controls? Okay,

[00:13:29] Andrew 3: yeah, let's roll it out. Let's, let's find out.

[00:13:30] So let's, let's, let's do it. Why don't you... It's your study. Yeah,

[00:13:32] Taylor: let me set

[00:13:32] Andrew 3: up the- Do you want me to show- ... question ... do you want me to show? I can screen share some stuff here.

[00:13:35] Taylor: Yeah, you could, you could do that. What do you, which

[00:13:36] Andrew 3: one do you wanna do? Cost controls?

[00:13:37] Taylor: Yeah, let's start with the cost controls. So- Okay, so,

[00:13:38] Andrew 3: so what is, what is the question-

[00:13:40] Taylor: Yeah

[00:13:41] Andrew 3: that this, that this study sought to answer?

[00:13:43] Taylor: So the headline is Do Cost Controls Work? Which is more clickbaity than anything. Yeah. But the question we're trying to ask- I, I,

[00:13:49] Andrew 3: I actually, I was like, "That's not really what this is about."

[00:13:51] Taylor: Yeah, yeah. The question it's really asking is do they deliver the outcome that you set?

[00:13:57] Andrew 3: Yeah.

[00:13:57] Taylor: So i- i- if you're trying to, the idea of a cost control is define the expectation of the outcome that you want, either in a return on ad spend or in terms of a cost per acquisition or cost per result goal.

[00:14:10] Andrew 3: Mm-hmm.

[00:14:11] Taylor: Does setting the goal deliver the outcome? Now, there's a separate question about volume and all the things that you wanna, you could do as a supplement to this, but strictly we wanted to ask is does Meta deliver against the bid that you've set?

[00:14:24] Andrew 3: Okay. Yeah, okay. So I set a ROAS target of two to one.

[00:14:28] Taylor: Yes.

[00:14:29] Andrew 3: Do I get a two to one?

[00:14:30] Taylor: That's right.

[00:14:30] Andrew 3: I set a cost cap of $100, do I get $100?

[00:14:33] Taylor: That's right.

[00:14:33] Andrew 3: Okay. One of the things that's challenging in this conversation we should get out in front of is that, uh, it is, it is bad to perform below ROAS but above CPA.

[00:14:44] That's right. Yeah, we're gonna- So we're gonna for sure get confused there. Yeah. But right, the, the bad goes in two different directions. That's right. Yeah. You want your ROAS to be better than it says, you want your CPA to be... Yeah, anyway, you get the

[00:14:54] Taylor: idea. And so one of the reasons I think we're in a unique position to a- ask this question is because, uh, almost all of our spend is on cost controls.

[00:15:01] So-

[00:15:01] Andrew 3: Can we link this, by the way, before- Yeah, yeah. So this will go in the show notes? Yep. Okay. Um- So you can get this study somewhere in the show notes for this episode. You can read the whole thing, and there's another one coming as well.

[00:15:11] Taylor: Yeah. You'll, you'll have access to all of it. Um, so we're in a unique position because we have a really large set of spend.

[00:15:17] You can see that it reviewed 1.5 billion, the specific analysis on $200 million of spend. So, um-

[00:15:22] Andrew 3: Across 253 accounts.

[00:15:24] Taylor: Yeah, 253 accounts. Um-

[00:15:26] Andrew 3: It's a lot

[00:15:28] Taylor: And not only is it a lot of spend, but all of it's ... Because so much of it is on cost controls, I don't know that there's many samples that are much larger on specifically this issue.

[00:15:36] So I think that's why it's exciting- I

[00:15:38] Andrew 3: mean, yeah ... to go

[00:15:38] Taylor: after.

[00:15:39] Andrew 3: I don't- I've never seen any other agency publish it. Yeah. Especially 'cause I don't know if any other agency's buying so uniformly on cost controls- That's right ... like you guys are. And, uh, so I would think the only larger data set's Meta.

[00:15:49] Taylor: Yeah, right.

[00:15:49] I, I ... And, um, so, so with that then I think we can ask this question pretty effectively. And so there's an outline- Yeah, go ... here of what the report is trying to learn. Um, and there's a few different things that we're, like, after here, uh, in terms of looking at this. We're trying to understand all three different types of cost controls.

[00:16:09] And here's another acknowledgement is that the vast majority of our spend, if we look at what Andrea has on the screen now- Yeah ... is actually on min ROAS. This surprised

[00:16:18] Andrew 3: me.

[00:16:18] Taylor: And we're

[00:16:18] Andrew 3: gonna talk, we're gonna talk about why. Yeah. This surprised me a lot. So the- Yeah ... this is an in- interesting thing. So much of your spend is on, is on minimum ROAS, which of course like for people listening means it's actually on highest value optimization- That's right

[00:16:28] not highest volume optimization. That's right. I think that's very interesting, and I have something to say about this. Some people ... I, I got ... There's some attention on a podcast episode and a couple tweets about incremental attribution- Yes ... which we've switched to, which may materially impact this, and I do wanna talk about that at some point.

[00:16:40] Yeah. Um, 'cause I've, I've seen some different behaviors, a little anecdotally here, with incremental attribution than standard. But- Yeah ... but, um, but, uh, anyway, so much minimum ROAS spend. So anyway. And,

[00:16:51] Taylor: and, and not very much bid caps spend, which is your favorite. Yeah.

[00:16:54] Andrew 3: Yes. Yeah.

[00:16:54] Taylor: So, um, let's talk quickly about why.

[00:16:57] This is actually ... One of the things that's fun about doing these studies is you get to see the actual true behavior of your organization and how- Yeah.

[00:17:04] Andrew 3: Right ...

[00:17:04] Taylor: the, the, the cultural trends and narratives have shaped towards that 'cause this has not always been true. No. Yeah. I'd say that like two years ago this was- It was all cost caps, right?

[00:17:12] Yeah ... very disproportionately cost per result. Yeah. But what we believe, and I think it's part of what manifests in this study a little bit, is that the problem with cost per result goal is that it's incredibly difficult to actually set the bid.

[00:17:23] Andrew 3: Yeah.

[00:17:23] Taylor: Um, especially for brands with tons of SKUs where the pathways- Yes

[00:17:27] to purchase are many, that actually defining the correct target is so, so challenging.

[00:17:33] Andrew 3: Yeah.

[00:17:33] Taylor: Um-

[00:17:34] Andrew 3: Let ... Well, I'll give you an example. I have a client right now. Yeah. So because we've been using more cost cap- Even with bid caps, we've had this problem for forever. Yep. I've got a client that sells like, um, uh, jewelry.

[00:17:44] Taylor: Yes.

[00:17:44] Andrew 3: And so their account structure is like, uh, sort of a consolidated CBO. And each ad set I would say is what I would call an economic unit. So it's not always an individual product. Right. But if it's like, the- let's say there's three products that all have roughly the same AOV, and the same margin profile, and the same LTV profile, which this, in this case there's a, a fair amount of these products that do this.

[00:18:11] Yep. And do that. So now, so in the end you've got a CBO with like, right now it's like probably 12 or 15 ad sets, and it's kind of expanding all the time. Yep. And each of those ad sets represents a slightly different AOV or something. And so now as a media buyer, if you're setting a cost cap or a bid cap, you have to pay attention to the AOV- That's right

[00:18:28] in 15 ad sets. And this gets back to the sample size thing, where it's like, what is, at what point does an AOV become reliable? What happens if one of the first three purchases somebody buys five pieces of jewelry, and now your AOV is like twice what it normally is? Do you believe that's because something about this ad produces monster or...

[00:18:43] You know what I mean? So, so, so the, so the media buyers start doing this, and I've actually been frustrated at points with my media buyers who are going c- for not paying enough attention to this. And what, what I hear you reflecting is that it's 'cause it's really hard.

[00:18:53] Taylor: Well, a- and it forces you to guess and change, guess and change, guess and change.

[00:18:57] Constantly. And so, and then you run into these situations like I, I'm sh- I'm showing Andrew something that maybe we could pull screenshot later, but I'll just describe what I'm looking at. It's an AOV histogram for a brand- ... that sort of illustrates the complexity of the problem sometimes. So this is a brand with a very large SKU set.

[00:19:11] It's home goods, so lighting-

[00:19:13] Andrew 3: Oh, gosh ...

[00:19:13] Taylor: fixtures, everything. It's

[00:19:14] Andrew 3: everything from $20 to frigging $3,000 I bet. Exactly. Yeah, right.

[00:19:17] Taylor: So you can see the AOV histogram order distribution. A- and this is ... I encourage brands all the time to look at the three measures of central tendency, mean, median, and mode order values, right?

[00:19:25] We talk a lot about this with setting, uh, uh- Free shipping threshold ... free shipping thresholds and things like that. But what it illustrates to you is the actual, uh, distribution of order values such that you would think about in an ad account what you're actually trying to accomplish. Um, and so, so often what I see brands do is they take their average order value for their website, divide it by their ROAS, equals their CPA target.

[00:19:45] Andrew 3: Yes.

[00:19:46] Taylor: Right? But in many ways, that-

[00:19:47] Andrew 3: That's totally wrong.

[00:19:48] Taylor: It's totally wrong. It

[00:19:49] Andrew 3: will mislead you dramatically.

[00:19:50] Taylor: Yes, exactly.

[00:19:50] Andrew 3: You'll end up dramatically overspending on low AOV things and underspending on high AOV things. You'll ... The distribution of your dollars will be a mess- That's

[00:19:57] Taylor: right ...

[00:19:57] Andrew 3: doing this. That's exactly right.

[00:19:58] Because, because, like I've got a, a ... another client right now where we just did this, and it was like they've got one product that is like something, and it's like probably produces 3X the AOV of the vast majority of their catalog. That's right. And if you lump those together, you're gonna really underspend on that product even though it's probably the best product to spend on.

[00:20:13] You have to give it separate treatment.

[00:20:15] Taylor: That's exactly right. And, and to be clear, this is true without cost controls if you run on- 100% ... cost versus highest value. Yeah, yeah. Anyways, but so the example I'm looking at, this brand has an average order value of $1,000 ... $1,027. The median order value is 512, and the modal order value is 141.

[00:20:31] So all three measures of central tendency wildly disparate. And so the question becomes, if you have a limited budget, where do you set the cost control?

[00:20:38] Andrew 3: Right.

[00:20:38] Taylor: A- and I don't, I don't know that there's a right number.

[00:20:41] Andrew 3: Yeah.

[00:20:41] Taylor: Um, and, and in reality what VO min ROAS solves for is it solves for having to optimize to answer that question.

[00:20:49] Andrew 3: Because the AOV is baked into the control.

[00:20:51] Taylor: That's right. And so it minimizes the amount of work that humans have to do, and one of the preferences I have all the time- Yes, yeah ... is to reduce the dependency of lots of numbers of people making individual decisions about where to set things.

[00:21:04] Andrew 3: Now, we, we should say Meta's rolling out

[00:21:06] I don't know if you guys have used this yet.

[00:21:07] Taylor: Max conversions on VO.

[00:21:08] Andrew 3: Yes. They're rolling- Yeah ... they're ... Well, they're-

[00:21:10] Taylor: Which I don't even know what it means totally. And

[00:21:12] Andrew 3: I think the idea is that you can set a ... So the id- so they're rolling out highest volume optimization-

[00:21:16] Taylor: Yep ...

[00:21:17] Andrew 3: so CAC optimized instead of va- value optimized, with a ROAS control.

[00:21:21] Yeah. And I think the idea is that the ... is that it's like- See, I

[00:21:23] Taylor: thought it was the other way. I thought it was still value optimized- Nope ... for max conversions.

[00:21:27] Andrew 3: It's volume optimized. It's volume optimized, but you can set a ROAS target. So, uh, I think the idea is that they are solving this problem, which is like it- Yeah, but

[00:21:37] Taylor: it, but it doesn't, doesn't really make sense to me.

[00:21:38] I don't really understand what the input would be

[00:21:42] Andrew 3: I think the input would be that Meta just reads the AOV and says like, "We're still gonna try and get you a certain, uh, a s- a certain ROAS even though we're optimizing for the, for the highest volume of purchases."

[00:21:52] Taylor: But the highest volume of purchases would, if you just make- It's

[00:21:54] Andrew 3: not trying to actually- All of that-

[00:21:56] get higher value spenders, which is the problem with highest value.

[00:21:58] Taylor: Yeah, but that's, but I, I go, if you think about this distribution, if I'm pursuing, if I have a fixed budget-

[00:22:04] Andrew 3: Yeah ...

[00:22:04] Taylor: and I want the highest number of purchases, I'm always gonna end up with the cheaper product.

[00:22:09] Andrew 3: Yes.

[00:22:10] Taylor: There's just no way around that.

[00:22:11] Yes. So in that sense, it doesn't really seem to make any sense to me that it would solve for this problem other than it would allow you to run- Towards a different distribution port part of the order set

[00:22:23] Andrew 3: That's, that's it. With a ROAS target I think the point is it would allow you to reach a different customer by, uh, the customer who's gonna sp- like in your example of home goods, right?

[00:22:32] Right. The reason the modal order value is 141 and the, even though the mean is one... $1,000 is because of course there's a lot more people buying $140 objects than $3,000 objects or whatever it is, right? Just because it just, it just is the reality of how like money gets spent in the world, right? So, um, so in that case, like you'd, you'd be able to sort of target that customer without actually optimizing up the value chain- Yeah

[00:22:54] which is what the highest value optimization does. It's the reason why I recommend running both- Yeah ... volume and value, right? Exactly. Yeah. Exactly. Is in, in every account. We ... Literally in our setup, we duplicate every ad and run it in a value setup and a volume setup. Yeah. So do we. And in some accounts we've gotten more value spend, in some accounts we've got more volume spend.

[00:23:08] It just depends on the products and the customer and all those kinds of things. That's right. So yeah. So that's the reason I recommend it, is 'cause you're actually reaching b- people on different parts of a like, um, behavioral distribution.

[00:23:17] Taylor: Yeah. And if you, and if you actually look at the reach, it'll, it'll be And the

[00:23:20] Andrew 3: conversions and all that.

[00:23:20] Yeah. Right.

[00:23:21] Taylor: So, so I totally agree. Okay. Yes. And our bidding structure is start VO, we, we call it bid surface expansion to get more volume, you would expand it. And actually right now, partially because of the results of the study, which we're jumping ahead, we, we actually are now the default is bid cap, not cost per result call.

[00:23:34] Andrew 3: Really? Uh,

[00:23:34] Taylor: yeah. Uh, we'll, we'll talk about-

[00:23:36] Andrew 3: Wait a minute, wait a minute. What just happened, Taylor?

[00:23:38] Taylor: We'll, we'll talk about why. Like I, I think that, um, I don't- Wow ... actually know that the answer is- Yeah.

[00:23:42] Andrew 3: You don't have enough bid cap spend for that to-

[00:23:43] Taylor: Yeah. But, but I feel fairly confident in the CPR goal being like just not as useful as I would like it to be.

[00:23:49] We'll get- Yeah ... so jumping a bit ahead. But so this gives you a bit of the distribution of the spend is just where are we spending the money. Um, and then the next is like how bid targets were set. So I think this is just an interesting illustration of what that weighted average goal was, uh, what the median was-

[00:24:06] Andrew 3: Yeah

[00:24:06] Taylor: what the middle 90th percentile, how many data points in the spend. So this just shows you where we're setting our targets. Yeah. So this is always kind of interesting because I think people, um, are always wondering sort of what's the, the, you know, average result on meta kind of thing. And the other, uh, thing we'll get to, and if you scroll just quickly down, um- So

[00:24:23] Andrew 3: you're

[00:24:23] Is there anything you wanna pull here? Like-

[00:24:25] Taylor: Let me come back to this- Okay ... 'cause this next part is really important. Yes. But go down to the attribution settings split- Yes ... 'cause this is something I think you brought up. So-

[00:24:31] Andrew 3: Yeah ...

[00:24:31] Taylor: so we'll show you by placement, and then we also show you how much of the spend is on each attribution setting.

[00:24:37] Um- Yeah ... so our default most often is min ROAS seven-day click. Yeah. And you'll see that that houses the vast majority of the spend. Yeah. Second most would be cost per result, seven-day click only. Yeah. But there is some spend on seven click, one view- Yeah ... one click only, um, et cetera, et cetera. So you can see how much spend is on each attribution setting as well, and how those delivered to target individually.

[00:25:00] Andrew 3: Yeah.

[00:25:00] Taylor: So, um, the, the TLDR if you scroll back up, or actually w- we can just stay here.

[00:25:05] Andrew 3: Yeah. This might help. This is helpful. Sure.

[00:25:06] Taylor: Is, is that min ROAS, uh, is an incredible product. I am actually like, I think we need to stop and the whole world needs to acknowledge Meta's unbelievable capacity to deliver this outcome across a broad enough data set.

[00:25:23] Andrew 3: It's-

[00:25:23] Taylor: Um-

[00:25:23] Andrew 3: It's astounding. Yeah. And, and it is, the single dumbest counterargument to all of this to me is like, "Meta's out to get you," or like, "Meta's..." Or what- or, "Meta's not reliable." It's like, it's, it just like, not only is it obviously wrong- Yep ... but, like, for a bunch of reasons theoretically, I think, uh, it, it has always seemed fairly clear to me that was the case.

[00:25:44] But when you point this data set out and saying that it is 97% actual to target-

[00:25:49] Taylor: Yes. So across a hundred-

[00:25:50] Andrew 3: I'll, I'll say something. This result actually really surprised me. Yeah. Because what I have found is that we find that our 2 ROAS bids are, uh, typically Uh, less reliable

[00:26:03] Taylor: So you should run it. Like you should, you should do it

[00:26:04] Andrew 3: Well-

[00:26:05] Taylor: You should check

[00:26:05] Andrew 3: well, I, I'm going to again- Yeah ... for exactly this reason. And, but, um, well, anyway, I ... So this is the most surprising result in this whole thing to me. Yeah. The idea that tROAS actually performs really close to target because- So I think that- ... in our experience, it's the case. Do you think this is because you're spending so much money here that essentially the model- No, I

[00:26:22] Taylor: think we got here-

[00:26:22] gets really good ... because we experienced this. Like, I actually think that- It's so

[00:26:24] Andrew 3: different than our experience ...

[00:26:24] Taylor: that what, what happened is, is that it just worked a lot better more often. Um, and so we ended up with that result. Spending more money. Yeah.

[00:26:31] Andrew 3: Interesting.

[00:26:31] Taylor: So, so, now I don't know that for sure, but I, I know that we didn't have some sort of like organizational pressure to put people into VO.

[00:26:37] But, um, so you can see that the target was 236. The average delivered was 228. That's a 96 and a half percent of target. Seven-day click, one-day view is actually at 101% of target. Uh, one-day click, um, missed target more often, which I think is like, uh, not totally surprising either. It's

[00:26:51] Andrew 3: more ... It's gonna be noisier on- Yeah.

[00:26:52] That's right ... a one-day window, right? Yeah.

[00:26:53] Taylor: Uh, and then where, where, though it is ... I think the thing that I was actually s- more surprised by was how poor cost per result goals were- Yeah ... at delivering to target. Yeah. And this is actually much more consistent- This is surprising to me ... with my experience as well.

[00:27:06] Andrew 3: Yeah.

[00:27:06] Taylor: So seven-day click was at 140% of target. So- This

[00:27:09] Andrew 3: is part of why I have not used cost caps very much historically. Yep. Because I have ... When I have, when I have tested this in accounts, and I use tested in a general sense, like th- this is real testing, what you're doing here. Yeah. What I'm doing was like anecdotal testing, but it's part of the reason why is it because of this is- Yeah

[00:27:25] Taylor: yeah. So set 100 ... So a- an average bid of $78 delivered $108 as an outcome, so 140% of target. Um, now what ends up happening is the way that that gets managed in an account, and this is my experience too, is that- Yeah ... you just keep changing the bid, right? Yeah, you- You just keep lowering it and lowering it and lowering it and lowering it- Yeah

[00:27:39] until it ... You drag it into the performance that you need.

[00:27:43] Andrew 3: Right. If you have a baseline idea that it performs about 50% worse than you expect, then you just set the bid 50% worse, and this has always been the case with cost controls to me- That's right ... which is like generally speaking, like it doesn't matter to me if they're actually accurate as long as they are relatively consistent.

[00:27:57] That's

[00:27:57] Taylor: right.

[00:27:57] Andrew 3: And in an account, you just set ... You just adjust. We do this all the time.

[00:28:00] Taylor: Yeah. Yeah. And, and so w- what happens there, um, is I think about this not too dissimilar from the way we're gonna treat incrementality factors, which is that what you want- Yes. Yes ... to do is have a starting point of expectation of how- Yes

[00:28:12] you should teach people to set the bids.

[00:28:13] Andrew 3: Yes.

[00:28:13] Taylor: And you want it to, it be algorithmic in the sense that there's a formulaic input to- Yes ... okay, what is the cost control that I want? And so we can use these factors to begin the process of analysis. Now, bid cap was better than cost per result, 123%- Yeah ... of target.

[00:28:27] It's a small sample. It's only $3 million. So to me, the reason we've moved to say, okay, well, I have a signal that bid cap is more accurate in delivering to the expectation that it wants-

[00:28:37] Andrew 3: Yeah

[00:28:38] Taylor: It's not a large enough sample. I'd like to increase the sample there- Yeah ... to see if I can grow my confidence.

[00:28:42] Rerun this study

[00:28:42] Andrew 3: in however

[00:28:43] Taylor: many months. Cost per result I feel fairly confident is not going to deliver to the exact outcome, and I'd have to make it a factor adjustment- Yeah ... to my bid.

[00:28:50] Andrew 3: Yeah.

[00:28:50] Taylor: Um, and I think that that is just related to the difficulty related to the distribution of possible end nodes of purchase.

[00:28:56] Yeah.

[00:28:56] Andrew 3: That's right.

[00:28:57] Taylor: Um, it, it's just a- This is a- ... more complex task.

[00:28:59] Andrew 3: This is a great point because if you take the AOV out, of course the cost cap is going to work worse in the sense that, to use your home goods example- Yeah ... there's just a broad range of potential purchases. That's right. And so, like- So many

[00:29:11] when people, people get mad at, about this sometimes as if there's like, as if Meta's being like sh- like weird. It's like I, I actually saw somebody from Meta point this out in a tweet a while back, Kristen Kupke, shout out. She said like, "You're asking Meta to predict human behavior across your entire SKU set, across every brand with all of these things."

[00:29:30] Yeah. And it's actually incredibly good at it, and but, but it's an inherently volatile exercise. Yeah. The point is like the forecast is inherently volatile, so yeah.

[00:29:38] Taylor: Yeah. So, um, a- and this is the thing is that there's, there's aggregate data here that shows the power of a modeling product in large samples.

[00:29:48] Andrew 3: Yes.

[00:29:49] Taylor: Okay? And this is, goes back to the coin flip examples, is that this is what you should expect to be true and, and if we scroll down now and we look at the actual distribution of individual results-

[00:30:01] Andrew 3: They're all over the place ...

[00:30:02] Taylor: we can recognize that any individual set of flipping the coin-

[00:30:06] Andrew 3: Yeah ...

[00:30:07] Taylor: can produce- Yeah

[00:30:08] a wide array- Yeah ... of potential outcomes. Yeah. So what's on the screen now is the distribution of the individual accounts whereby the dotted line in the scatter plot represents the t- delivering exactly to target.

[00:30:24] Andrew 3: Yeah.

[00:30:24] Taylor: Okay? So you have the delivered result on the Y-axis and the target on the X-axis.

[00:30:29] Andrew 3: Yes.

[00:30:29] Taylor: And so you can see that distribution. Uh, same thing for the bid cap, or sorry, cost per result goal, and then you can see the distribution plotted a different way below. Now, what's important to recognize, um, if you look at the dots, you'll see that each of them, more of the dots- Coalesce below target in the min ROAS sample and above target in the CPA sample.

[00:30:56] Yes. So if you look at the... It's easier to see maybe visually in the distribution below where, um-

[00:31:03] Andrew 3: Yeah,

[00:31:03] Taylor: yeah, yeah. Right here ... it is, it is also fair to say that m- m- the majority of accounts deliver below target on ROAS goal- Yeah ... and above target on CPA target. Yeah. And in fact-

[00:31:15] Andrew 3: So by below and above, you mean worse than?

[00:31:17] Taylor: Yeah, so in fact, specific data, 67% of accounts on min ROAS, up the other way, um, right there. Yeah. 67% of accounts are greater than 5% below their own target.

[00:31:30] Andrew 3: Yeah.

[00:31:31] Taylor: And 75% of accounts are greater than 5% above target on CPA bidding- Yeah ... or cost per ROAS goal. So again, the average in the distribution is different than the majority.

[00:31:42] Yeah. It's different than the individual outcomes. Yeah. They're all different ways to consider the potential result.

[00:31:48] Andrew 3: Yes. So what we have always done about this, and what I think would still say, is run manual bids. Be conscious that they, they are, especially the smaller the sample that you're dealing with, the, the more likely it is that they're gonna deliver a range of outcomes.

[00:32:01] And you ... And the way we've handled this is we've just, like, looked at it with clients where it's like, is this client, is it better for this client to err on the side of overspending or underspending? That's right. Like, and so, like, your super high LTV supplement brand might be like, "It's actually more important for us to get the spend out than it is for us to protect the CAC target exactly."

[00:32:19] That's right. "Because if pushing out our, our, um, our, uh, break-even point to month four instead of month three, we're financed enough. It's not actually that big of a deal. We, as long as we get more customers." That's right. That's, that's one way of handling it. Whereas your brand with low LTV needs to make their money on first purchase might be like, "Mm, we're gonna constrain this a little bit more.

[00:32:36] We're not gonna push the growth super hard here," et cetera. Um, the other thing I'll point out here that I think is really interesting is this, like, the clustering of data points on the lower end of the ROAS and the, and the- Yeah, right, right ... CPA thing. Some of that is because the percentage is maybe the same when you're lower, like the percentage range is similar.

[00:32:56] Like, uh, you know, what I'm trying to say is, like, a 55 dollar cap- Uh, our f- $50, uh, cap and a $55 outcome is a 10% difference. Yep. Right? Uh, whereas, like, at, at a $200 cap, it like... A two, a $20, it's gonna... Anyway, it's gonna, there are gonna be a, a f- it's gonna look further from the line. But what I also wonder if the thing is happening here is that the larger the sample size for Meta-

[00:33:18] Taylor: Mm-hmm

[00:33:19] Andrew 3: the, the closer, in an individual account, the closer you're gonna get to actual target. And there's a reason that I... And I would actually be curious to see if there's a way to follow up on this for you guys. Um, like essentially, more conversions and less ad sets might produce results closer to target. And the reason that I think that, in part, is partly 'cause it's logical that, like, Meta having more data inputs would make it better at predicting.

[00:33:39] But also because when you th- understand how these products work, what they do is that, um, minim- target ROAS bidding and c- and, um, cost per result goal, right, cost caps, they both do the same thing actually, which is they, um, they are, they are able to bid dynamically above and below your target to explore audiences more aggressively and then come into your average in the aggregate.

[00:34:03] Yep. But when you look at the way this works, and I'm gonna use my hands to illustrate this point, they start by bidding really widely. Like, uh, my understanding is they can bid, like, like 9 to 10 X your target or something like that, like, in the earliest stages. But then they condense over time as, as more data comes in.

[00:34:18] And it would therefore be logical that, like, if you could sort of isolate the, you know, the first 10 grand and spend on an ad set, I bet you the range of outcomes is wider, uh, on a c- on a cost cap or a, or a min ROAS. Uh, I bet the out range of outcomes is wider than the next 10 grand, and then the next 10 grand, and the next 10 grand, and so on.

[00:34:37] Especially if you're getting, like, especially the more purchases you're getting per ad set, right? So, um, because this is all happening at the ad set level. So it would make sense to me that what you're saying is true here, which is that especially considered in the aggregate for minimum ROAS or for target ROAS, like that would, that they would all come together.

[00:34:52] But, but that maybe in an individual account, you would expect some condensing of the outcomes the longer the campaign exists. And I wouldn't be surprised if that's the reason target ROAS works so well for you guys, is you guys are spending, you guys have a lot of accounts spending big time dollars. Um, you know, and that, that, like, if you're spending three grand a day or two grand a day, you might expect a wider range of outcomes.

[00:35:11] And one of the problems with bid caps actually that I've experienced and what people sometimes get frustrated with, is that they don't dynamically bid, and therefore, you can sometimes overconstrain your spend because Meta will sort of lose confidence, especially with low conversion volume. And so that's the place we haven't used it.

[00:35:26] So that's a, that's one thing to consider here. The other thing is, that I would point out, is that, um, I am really curious what happens if you rerun this with incremental attribution. Um, and I, I'm sure you will in six months or whatever when you guys- We're working on it right now. I'm sure. Um, because what I've experienced with that is that so far, performance has been much closer to target, uh, for us.

[00:35:45] But, but

[00:35:45] Taylor: target is an interesting question here because you're not setting a bid.

[00:35:48] Andrew 3: Well, no, no, no. We're running incremental attribution with a cost cap or a target

[00:35:51] Taylor: ROAS. Okay.

[00:35:52] Andrew 3: Yeah.

[00:35:52] Taylor: Okay. So yeah, I think the qu- but, um, I think the, the problem with the- Um

[00:36:03] The question, one is, um... Well, so we're designing this right now. We only have a little bit of, of spend on this. We right now have s- We're running CLS on all of these to try and answer the incremental effect to determine if it's worth rolling out broadly enough to answer out to the target. The question is, what I'm trying to say is, what is the target?

[00:36:22] Meaning I don't know the incremental attri- or the incremental, uh-

[00:36:26] Andrew 3: Value

[00:36:27] Taylor: creation ... factor of IA.

[00:36:29] Andrew 3: Yeah, I think you would just- And so- ... run it g- given- 'Cause these are all relative to the d- optimization method you set, right?

[00:36:33] Taylor: Yeah, but I'm saying the, the target is set with a consideration for the incrementality factor.

[00:36:38] So right now the struggle we have is- So

[00:36:39] Andrew 3: you're saying

[00:36:40] Taylor: the- Where do you set the target?

[00:36:41] Andrew 3: So you're saying the actual here in this report, like the actual CPA is- Is- ... is incrementally adjusted?

[00:36:48] Taylor: No, I'm saying the way that they set the bid with- Yeah ... was the consideration for it. But right now-

[00:36:52] Andrew 3: I see. So you can't- So you don't know where to set the IA

[00:36:53] you're just saying you don't know where to set the bid.

[00:36:55] Taylor: So right now the f- step one is we're running CLS studies of s- of, of individual campaign- I see ... holdouts to try- So- ... and understand what the factor is on IA because in theory-

[00:37:02] Andrew 3: Then you can set the bids- Yeah ... and then you can see.

[00:37:04] Taylor: Exactly.

[00:37:05] Andrew 3: Okay, great.

[00:37:05] Taylor: Then, so there's like a step process to getting- Okay ... to it, um, to, to get there- Yes ... because-

[00:37:09] Andrew 3: Your early result, your early studies on this, and I think, uh, uh, I think houses as well-

[00:37:14] Taylor: A couple. The- Right.

[00:37:15] Andrew 3: It,

[00:37:15] Taylor: it's- The house has more. The-

[00:37:16] Andrew 3: It's very early, but I think the idea is that the, it's a higher incrementality factor on them.

[00:37:19] That would

[00:37:19] Taylor: be the goal, right? I would hope so.

[00:37:21] Andrew 3: Yes. Yeah. And, and, and logic and what you had said, I think what you first said in your, and I think it's called like, let's call it a beta or something like that, right? Yeah. Like, it's, it's like grain of salt on this. But what you first reported on this was incremental attribution was producing a higher incrementality factor.

[00:37:35] So you have to adjust the actual- Right ... performance up relative to what Meta reports. Meta is under-reporting more on incrementality, but even net of that, standard attribution was beating it. Yeah. That was what you first rolled out. Yeah, and- And what I believe is that this has all gotten better in the last few months.

[00:37:47] Taylor: Which is what the house study showed- Yeah ... and why we're, why we're willing to dive back into it. Yeah. That what I have seen, and it's funny, is that, like, incrementality, both the geo holdouts as well as if you pr- decompose incremental attribution, and you were to sum up seven-day click, one-day engage view, one-day view, is that it's all sort of coalescing around seven-day click plus one-day engage view- Yeah, I know

[00:38:07] which was the historical seven-day click. Yeah. Yeah. So it's like we all pressed messa- Meta into this seven-day click thing. I did not. Yeah. Fair. Um, and I think that the end result is incremental attribution might be very similar to that. Yes. But we'll- Yeah ... we'll see. We'll-

[00:38:20] Andrew 3: Yeah. The engage through thing, I, I think this, this has been a lot of my take on this, is that, like, the switch of click attribution really screwed up a bunch of our accounts.

[00:38:27] Yeah. Like, it really, really... Even, even when we adjusted our optimization method. Yeah. And so the reason we tested incremental was because of that. Yeah. And it seems to have been a really good solution to the

[00:38:35] Taylor: problem. I think on videos, uh, like campaigns that are video-heavy or... This is another thing, theory I'm working on is this- Interesting

[00:38:40] the, the, the change that, that when you... a- and if you break apart, like, the amount of, uh, engaged view that you get- Yes ... now, now it's hard 'cause they added a different definition, which is the watching five seconds- Yeah ... doesn't require any clicking at all. So it's not exactly apples to apples. Yeah. But anyways, I just wanna co- Wait

[00:38:54] I wanna wrap up a couple things. Is there

[00:38:55] Andrew 3: anything else you wanna say on this? Yes. 'Cause we're gonna run out of time, and we're gonna let-

[00:38:57] Taylor: I do. I wanna wrap up a coup- one last thing. Okay. So, so the end result-

[00:39:00] Andrew 3: Do you want me to reshare something?

[00:39:01] Taylor: Yeah, the end result is just that, um, we're using a factor when setting our bids.

[00:39:06] So just like an incrementality factor. Yeah. For Minross it's 95.9, and for cost per result it's 154. And part of that is it, because of... If you wanted to get to a 95% confidence range, so if you take that distribution- Okay ... and you wanted to try to create a bound where the predictive value was 95%-

[00:39:22] Andrew 3: Yeah ...

[00:39:22] Taylor: then you can see what that bound would be in each case.

[00:39:24] Yeah. It's 90 to 101% or 145 to 164%. Yeah. Covers a pretty large, uh, percentage of the potential viable outcomes. Yeah. So using that factor is a good starting point for people before you run that. But then the last thing I wanna show is just this table below, 'cause it kind of touches on this thing, which is, like, to give you some real examples- Yeah,

[00:39:41] Andrew 3: yeah, yeah, yeah

[00:39:41] Taylor: to illustrate. Like, we have this one account as an example that is just... I, it might be brand 14 in this example. It just, it is so off all the time- Yeah ... no matter what. Yeah. And it drives me insane. Yeah. And I can't figure out why. And I'm going through the events manager and the EMQ score. Yeah,

[00:39:56] Andrew 3: yeah,

[00:39:56] Taylor: yeah.

[00:39:57] Yeah. Is it the subscri- Like, we're trying to figure it out. But we recognize and acknowledge that there are brands- Yeah ... for which the individual outcomes will deviate from the mean, of course. But

[00:40:04] Andrew 3: the interesting thing about that, right, is you're not abandoning cost control for those. No,

[00:40:06] Taylor: not at all.

[00:40:07] Andrew 3: No.

[00:40:07] You just, you just make an adjustment. Right. This is what, this is how I've always handled this. And I've never known how to say this to people exactly 'cause it gets tricky. But this is also why, by the way, all of this is... Now, maybe you're gonna take the opposite stance here, but it's also why I just don't think this is gonna be very easy to automate anytime soon and why media buying I think is still really hard.

[00:40:23] People will say to me, like, like, something like, "Ah, come on, media buying is just commoditized," or whatever. It's like I- I- I just think that's, that's just not- Well,

[00:40:30] Taylor: I think that's, yeah, it's very, very wrong. Yeah. Now, I actually think this is the foundational piece that allows us to automate because-

[00:40:34] Andrew 3: Yeah

[00:40:35] Taylor: like if you think about- Yeah,

[00:40:35] Andrew 3: yeah,

[00:40:36] Taylor: yeah ... how to set a bid- Yeah, yeah ... it's just a, a, an operationally challenging idea.

[00:40:39] Andrew 3: Yeah.

[00:40:39] Taylor: Um, and so you have to have some starting point for it. But, um- Yeah ... so those factors, just like i- incrementality factors, are important components of automation. What

[00:40:46] Andrew 3: is the spend per- When it says, like, spend $6.9 million for brand six on the top of this, over what time period?

[00:40:51] Do you know?

[00:40:51] Taylor: I think the d- the date range sampled here is March 12th of 2025 through July 11th- Okay, so a

[00:40:55] Andrew 3: little over a year ...

[00:40:56] Taylor: of 2026, yeah.

[00:40:56] Andrew 3: Yeah. Interesting. Okay. Great.

[00:40:57] Taylor: Um, so these are just examples of, uh, yeah, accounts sampled and the deviation between them, so it can be pretty wide.

[00:41:02] Andrew 3: Okay. We're so low on time.

[00:41:04] You're- We're not gonna have any time for you to yell at me today.

[00:41:07] Taylor: That's fine. We're

[00:41:07] Andrew 3: gonna have to save that. We can do it right now. No, but you wanted to also talk about... Do you wanna save this for something else? Yeah, let's, let's just- The ad stop loss?

[00:41:12] Taylor: Yeah, yeah. Okay. 'Cause this one's, like- This one's simpler

[00:41:13] it's more convoluted. It's just, it's harder to... What I, what I wanna say- Does- What I wanna say is- Should I just turn off

[00:41:18] Andrew 3: an ad?

[00:41:18] Taylor: No. The answer is no. The answer is there is no discernible impact to making ad-level decisions. It doesn't make things worse, it doesn't make things better. It has no impact. You think you're doing something, but you are doing nothing.

[00:41:29] And so it is, it is a giant waste of time to spend a bunch of effort trying to manage things at the ad level. Um, so-

[00:41:36] Andrew 3: Turn- When you say manage things, you mean turn off- Turn- ... and on ads. Yes,

[00:41:38] Taylor: yes. Uh, and we have, like... And, and to be very clear, I know this because we've done it 16 trillion times or something.

[00:41:44] Like- I know ... because we do this endlessly, and it drives me nuts, and we do it at the compulsion of clients, we do it at the compulsion of ourselves, we do it for so many different reasons. Yeah. So, um, uh, now, uh, the, the, that, the tail-end research that we're trying to get to is a stop-loss rule that says, like, "Well, okay, but Taylor, what about the outliers?"

[00:41:59] 'Cause everyone tells me about the story of an ad that spent $20,000 at zero- I know.

[00:42:02] Andrew 3: Can I tell you how I handle those stories?

[00:42:04] Taylor: Yeah.

[00:42:04] Andrew 3: I think they're wrong. Yeah,

[00:42:05] Taylor: yeah. Well, so I, and I just go-

[00:42:07] Andrew 3: Most times ...

[00:42:07] Taylor: o- okay, but you have no counterfactual, um- Yeah ... to what would've happened- Yes ... 'cause it, it's impossible to create.

[00:42:13] Exactly,

[00:42:13] Andrew 3: yes.

[00:42:14] Taylor: And, um, you have no broader set of, uh, information about other circumstances like that. 'Cause the beautiful thing about if anybody gives me a rule, so if you're out there and you think you have a stop-loss rule that makes accounts better, so I see it all the time. Like, there's people that say, like, "At 3X CPA, if there isn't one purchase, we turn it off."

[00:42:29] The beautiful thing about if you're willing to stand up and put your hand up about a rule like that, is you can analyze- You

[00:42:34] Andrew 3: can test it ...

[00:42:35] Taylor: every ad that crossed that threshold over time- Yeah ... and what happened after.

[00:42:40] Andrew 3: Yeah.

[00:42:40] Taylor: So I can put any rule like that to the test across a very large sample, and what I'll say is that we had a bunch of internal people with rules.

[00:42:47] It used to be 5X the CPA at zero purchases, auto kill. It doesn't work. It is, it is a non-beneficial structural effect on the ad account. Um, and if you have a rule, send it to me. We'll run the study, and you can analyze it. We can even talk about it on the pod. Um, so w- w- well, the point is, is that this kind of stuff, like, we have a research directory that, that we're gonna be publishing that has things like this all the time, um, that we're trying to do.

[00:43:12] The MCP at CTC and the database structure that we have with Statlas combined with Claude has made this just, like, unbelievably helpful to do. Yes. Like, as an example, you just gave me a question which was, "Does, uh, does an increase of spend increase the accuracy of bids?"

[00:43:27] Andrew 3: Yeah.

[00:43:28] Taylor: Right? Oh, yeah. So while we're sitting here, we can just, like, take that research study and build on top of it across the database and answer the question.

[00:43:33] It's

[00:43:33] Andrew 3: awesome.

[00:43:34] Taylor: So it's, like, so fun. So, uh, more to come. If you have more research questions-

[00:43:37] Andrew 3: So

[00:43:38] Taylor: maybe the- ...

[00:43:38] Andrew 3: we'll answer them publicly ... maybe the thing here is do you, are you gonna do a separate podcast about this?

[00:43:41] Taylor: Yes. We have a whole, we have a whole set of things that we're gonna be publishing about this.

[00:43:44] Okay,

[00:43:44] Andrew 3: so here's what you should do.

[00:43:45] Taylor: It just doesn't

[00:43:45] Andrew 3: get the Andrew Ferriss exclusive. We'll, we'll link,

[00:43:46] Taylor: we'll

[00:43:47] Andrew 3: link this in the show notes- Yeah ... so you can go check it out for yourself. Um, go s- make sure you're subscribed if you're, if you're listening to or watching this on my feed. Let's, go, go to Taylor's feed and subscribe.

[00:43:56] What is, which, how do people go find you for this?

[00:43:58] Taylor: I, I mean, uh- Is this common? Is it- Yeah, you can do the Ecommerce Playbook podcast. That won't be me in most likely, but I'll, I'll be publishing on X a little bit. I'm in the middle of a incrementality study myself there, but, uh But, uh-

[00:44:09] Andrew 3: I have noticed you're quiet

[00:44:10] yeah,

[00:44:10] Taylor: yeah. Yeah, yeah. So, but, um- How

[00:44:11] Andrew 3: does it feel to be quiet on X?

[00:44:12] Taylor: Uh, uh-

[00:44:13] Andrew 3: Is it fun, or is it annoying?

[00:44:14] Taylor: It- It's a little bit of both. Yeah. There's lots of days I'd like to, to jump in there but- Jump

[00:44:17] Andrew 3: in. 'Cause it's fun. It is. X is fun.

[00:44:19] Taylor: It is. I agree.

[00:44:19] Andrew 3: Yeah. You, you told me how bad you wanna go away, but it's fun.

[00:44:22] Um, I go, I go through weeks where I don't post much, then weeks where I'm like, "Oh, this is fun." It is. Um, okay, so, so links in the show notes. Follow up with Taylor's stuff. Go get the Commentar Collective podcast, play- Ecommerce Playbook podcast, formerly hosted by Andrew Ferriss.

[00:44:36] Taylor: Yep.

[00:44:36] Andrew 3: So-

[00:44:36] Taylor: Formerly hosted by me.

[00:44:38] Andrew 3: Right, yeah. Um, and go do that. The, um, the, uh, the thing I wanna point out here is so- I think sometimes people think that I am, like, a- I don't know, and I think people think this especially about you, but that like, it- I'm like, um, dogmatic. I'm unwilling to be moved on things. Right. And the answer is not that I'm unwilling to be moved.

[00:45:01] It's that, it's that, uh, this is the kind of stuff that actually moves me. Right. It's historic. And, and the vast majority of like... This is the thing that kills me. The vast majority of like, "Oh, we've tested this- Yeah, yeah ... in media buying land," is nonsense. Right. It is nonsense. It is nonsense. It is nonsense.

[00:45:16] Yeah. It, it was not tested and measured and studied and none of those things. It's like, where I appreciate this, um, the CPMR guys- Yeah, yeah ... who did- Phil Collin, those guys. Yeah ... yeah, Phil. Right? Phil, you just, you ended up disagreeing with Phil on some of that stuff, but he made a real attempt

[00:45:28] Taylor: and- But publishing allows you to scrutinize- Yes

[00:45:29] in a way that's helpful to everyone.

[00:45:30] Andrew 3: Yes. I loved it. That's why I invited him on the podcast. Yeah. It's, I, it allowed somebody to go and say, "We're actually trying to discern truth here. It's not just some media buying," be like, "Oh, I tested this," you know? Yeah.

[00:45:39] Taylor: And it's risky. You subject yourself to scrutiny.

[00:45:40] Publishing data means that someone is gonna take it, and they're gonna try and tear it apart with- Yeah ... intention. That's right. And I, I actually- They're gonna

[00:45:46] Andrew 3: do that with these ...

[00:45:46] Taylor: it, yes, exactly. It is beneficial to the community that we do so, and that we can refine that. Yeah. That's kind of what we do internally.

[00:45:51] It's like, "Here's a proposal," then we go through and read it, and it's like, "That doesn't make sense. What about the..." Like, so all for it.

[00:45:55] Andrew 3: Yeah.

[00:45:56] Taylor: Um, five minutes. One thing, um- Uh, about you. So-

[00:46:01] Andrew 3: Okay, hold on ...

[00:46:02] Taylor: I've talked about, like- Yeah, go ahead ... Andrew and Patrick's podcast themselves that they're doing, that they're just, like- Oh

[00:46:06] subjecting themselves to the world of, like, just-

[00:46:09] Andrew 3: Build in public. Yeah,

[00:46:10] Taylor: yeah

[00:46:10] Andrew 3: Try to, try to talk about

[00:46:11] Taylor: it. In a way, I, I- The, what, what

[00:46:12] Andrew 3: Taylor- I love it so much ... what Taylor's talking about is that Patrick and I, we record, my, my business partner Patrick and I, Kadau, record every, every month. We do a podcast.

[00:46:20] And what we're just doing is airing conversations that we're having internally about what to do at our agency. Yeah. Sometimes it's, like, cool stuff we're rolling out. Uh, sometimes it's, like, our last one was about, like, I sort of feel like we should go chase a creator business, but I'd- Yeah ... also feel like it would be secondary to our thing.

[00:46:34] So we do it- It's just so much better, like- We do it every month.

[00:46:36] Taylor: So- Yeah ... there's this, there's this brand of podcast right now. I've said this about the SaaS Operators too, 'cause it, it, it feels and sounds similar to me. It's very dissimilar to the Operators podcast, and this isn't a knock on those guys.

[00:46:44] Yeah. They're very smart, but they're just telling you what to do the whole time. Yeah. And they're giving you advice from a place of knowing the answer. Yeah. And, and that is super helpful, don't get me wrong. But I love being invited into the wondering. Yeah. Um, the curiosity, the, the lack of answers. It makes me wanna jump in the conversation way more-

[00:47:02] Andrew 3: Oh, cool

[00:47:02] Taylor: than when I feel like you're just talking at me.

[00:47:05] Andrew 3: Yeah.

[00:47:05] Taylor: Um, uh, like, I want to be in the room with you guys- Yeah, yeah ... when I listen to you.

[00:47:08] Andrew 3: You should do one with us.

[00:47:09] Taylor: Yeah. It'd be... Well, yeah, I mean, I, I think that in some ways your guys' journey is your own, and- Yeah ... I, I think it's really cool. So I, I, I really appreciate that.

[00:47:17] But I think-

[00:47:17] Andrew 3: Okay. Go ahead ...

[00:47:18] Taylor: I get to ask of you now, 'cause I

[00:47:20] Andrew 3: listened to- Okay,

[00:47:20] Taylor: great, great, great ... so you have, you brought up this whole thing of, like, is our goal the right goal?

[00:47:23] Andrew 3: Uh-huh.

[00:47:24] Taylor: And you and Patrick, Patrick is so, like... He's funny. He, like, you, you, uh, you, I think, would be so much more willing to throw things out and start over.

[00:47:32] Yeah. And he is, like, seems, like, ruthlessly committed to just know this is- Stick to

[00:47:36] Andrew 3: the plan.

[00:47:36] Taylor: Yeah, yeah. Stick to the plan. Yeah.

[00:47:37] Andrew 3: Yeah.

[00:47:38] Taylor: Um-

[00:47:38] Andrew 3: And we have, like, a EOS, VTO, we've done the whole thing. We've got, like, a five-year... We did it on a five-year window, not a 10-year window. But basically that whole thing of, like, here's how much money we wanna get to EBITDA-wise.

[00:47:46] Yeah. We've got all of our goals really clearly defined. We have this year, here's the, here's the services we wanna have, like, rolled out, here's what we care about, all that kind of stuff. So we got it down on rock. Yeah. So, so yeah, we have a cl- we have a pretty well-defined goal.

[00:47:56] Taylor: I, yeah. And I think the plans are awesome because they will increase the likelihood of achieving that specific thing.

[00:48:02] Yes. Um, I think they are also limiting in that they will help you to achieve that specific thing.

[00:48:06] Andrew 3: Yes.

[00:48:06] Taylor: Um- Yes ... and, and so if you wrote down a number twice as big You would probably increase- That's interesting ... the likelihood of accomplishing that specific thing. Yeah. Um, and so the number you write down becomes very power- powerful.

[00:48:17] Yes. It's a very powerful driver of your behavior. It's

[00:48:19] Andrew 3: really helpful.

[00:48:19] Taylor: Um, and so, and the other thing that I've learned, uh, is that the macro environment in which you exist is a unique variable that should be a consideration in the plan. Um-

[00:48:31] Andrew 3: Interesting ...

[00:48:32] Taylor: that is hard to, to predefine.

[00:48:34] Andrew 3: What do you mean by that, the macro environment?

[00:48:36] Taylor: I think that it is a very good moment for service businesses.

[00:48:39] Andrew 3: Yeah.

[00:48:39] Taylor: Um, and I don't know that that will be true three years from now. Yeah. I don't know that it won't. I, I think there were years that were very hard for CTC, and there were years that were very good for CTC. Um-

[00:48:49] Andrew 3: That were not about you guys, it was about the environment.

[00:48:50] Taylor: That's right. A- and I think that you could potentially accelerate the end result by recognizing those windows-

[00:48:56] Andrew 3: Yeah ...

[00:48:56] Taylor: acknowledging- And trying

[00:48:57] Andrew 3: to capitalize ...

[00:48:58] Taylor: yeah, acknowledging that in the, in the good or bad spectrum of environments, if you're in a good one, that that should at least, uh, beg the question: Could we gather more now while the weather is nice- Yeah

[00:49:11] um, to potentially recognize that that won't always be true?

[00:49:15] Andrew 3: Yeah.

[00:49:15] Taylor: Um, now again, I, I, it's really hard to tell you for sure the weather will be bad later or it won't be. Yeah. Like, but I just, there's this moment, um- I think you're uniquely i- in a position to, to do that. Yeah. Um, X matters in the world, podcasts matter in the world.

[00:49:32] Andrew 3: Yeah.

[00:49:32] Taylor: The, the specific service function is still in demand. Yeah. AI hasn't disrupted it in any- No ... substantive way. Uh, if anything, it's been additive to the expectation. Yeah,

[00:49:40] Andrew 3: we've, we've only hired more with, uh, with more AI.

[00:49:42] Taylor: Yeah. Yeah, yeah. So, um, all of those things, I think should ask the question, like, when we set the goal, did we, um, underestimate, uh, any of the inputs?

[00:49:52] Were we wrong about them in any way that if we s- if we were recreating that plan today, knowing what we know now, would we create the same plan? I

[00:49:59] Andrew 3: think that's a really good framework. Yeah. If I was to recreate the plan today, would it still be the same plan?

[00:50:03] Taylor: That's right. That's a really- And if the answer is yes, then stay the course.

[00:50:04] Then

[00:50:05] Andrew 3: stick... That's a very good, that's a very helpful way to answer that question. But

[00:50:07] Taylor: if you said, we, if you're, 'cause like when we build our input for nuke- for our business, one of the inputs is, like, new customer acquisition expectation. And it, it, if, if that was just an input because we had a model that said we were gonna do $10 million in new business, and it turns out 15 was possible, then sticking to the model would be dumb- Dumb

[00:50:23] because i- it wasn't about some lifestyle thing that we wanted. It's

[00:50:26] Andrew 3: just- That's funny. You've, you've gone all over the place on this question- Yeah ... over time, and I, and I, that's fine. This is exactly what we were talking about earlier, right? Like, you just a- a- answer from the- I

[00:50:32] Taylor: just think creating a plan is a really powerful shaper of behavior.

[00:50:35] Yeah.

[00:50:35] Andrew 3: But at one point you said, you actually said you think brands should spend less to achieve the same amount of revenue, take the profit if it's gonna meet their plan.

[00:50:43] Taylor: Well, so, so because I think that when you define the plan, you should behave within the context of it.

[00:50:48] Andrew 3: Okay.

[00:50:48] Taylor: I think there should be moments where you scrutinize or evaluate the plan.

[00:50:52] I don't think it should be every day.

[00:50:53] Andrew 3: Yeah.

[00:50:53] Taylor: I don't think it should be every week. Yeah. I don't think it should be every month.

[00:50:56] Andrew 3: And I, I, I do have a tendency to almost be every week with it. Like, I like, like, we're, and it's because I'll see an opportunity. It's, one way of looking at it is being opportunistic.

[00:51:03] Another way of looking at it is- I think

[00:51:04] Taylor: it

[00:51:04] Andrew 3: would help you- ... is shiny object

[00:51:05] Taylor: syndrome ... it would give you freedom if you actually defined a space at a midpoint or a quarter point to say, like, "Andrew's allowed to change the plan here." And you could ch- scrutinize it- You

[00:51:13] Andrew 3: know what's funny about this is I remember sitting on the other side of this conversation with you as a leader.

[00:51:16] Taylor: Yeah.

[00:51:17] Andrew 3: I'm realizing right now you're, you're doing this to me.

[00:51:18] Taylor: And I changed, I changed, I said, "Don't change the plan"?

[00:51:20] Andrew 3: No, no, you, I just rem- I remem- you've heard me say this as a, as a critique- Yeah, yeah ... but I don't really think even a critique is right, is that like- I, I think about this with your questions around faith.

[00:51:29] Yeah. I think, like, this is what Taylor believes today. Right. But Taylor believed a really different thing about a lot of things two years ago. That's right. Yeah. And, and so I'm... So, and actually once I accepted that about you- Yeah ... then it made it really fun to work with you, 'cause it was like, okay, I, I never think you're lying to me.

[00:51:41] Yeah, yeah. I always believe you're telling me the truth about what you believe today. Yeah. I- I also know-

[00:51:44] Taylor: It may

[00:51:45] Andrew 3: change ... it may change. Yeah. And that's okay. We're gonna try and do what we think is best in the, in the- Yeah ... in that moment. Uh, but- And- ... but, but you also didn't make those swings every week. It

[00:51:53] Taylor: was like- Well, and, and I think- It was like, it was like, I don't know, a year- I think if there's anything that I've tried to do, and this is probably what has created a little more stability for CTC, is to try to have at least some horizon to the set of behavior.

[00:52:01] Yeah. Like, to say like, "Okay, quarterlies is the business rhythm, where for this quarter these are the KPIs and we're not gonna change them." Yeah. "And we're, we're gonna behave with these." Yes. "And so we're gonna do this." Yeah. A- and like right now, we just did this at CTC, is that, in, in part because we got off planet in like this, in, in a portion of it, but it was like, okay, I wanna try, for this second half we're gonna add this specific set of actions as an alternative opportunity to see what it unlocks, um, and we're gonna behave within the context for a while.

[00:52:27] I think it's very hard if you change it too much, but I do think that the scrutiny of the inputs in some period of time. This is where, like when clients come to us, I, we, we do every month forecasting at CTC. Yeah. If they tell me they have a budget that they wrote at the beginning of the year and they force me to behave within it, it drives me insane.

[00:52:42] Yeah. I don't actually know how it's possible if you're-

[00:52:45] Andrew 3: No ...

[00:52:45] Taylor: have a $2 trillion new customer acquisition gap for me to go next month and act like your August forecast is right. Right. Like, I, it's very, uh, demoralizing. Right, right. Um, now there's certainly a balance. Like, I, I don't know- Yeah ... I do think, and I've gotten this critique from people that like the biggest limiter of CTC is actually the plan that I wrote.

[00:53:04] Andrew 3: Yeah.

[00:53:05] Taylor: Um, in the sense that it's just a number. Now it's a really good number, but why didn't I write 5 million more? And it's just because I didn't have a mathematical modeled pathway to it- Yeah ... that I knew the answer to. But rather than it being a discovered frontier, I just, I, I operate within the bounds of a thing I can build a bridge to in my brain.

[00:53:24] Andrew 3: Right. I'll say, we built our plans basically on like month one of really building a business. Yeah. And I, so I think it ought to be a wider set of error bars on that. Yeah. Right? Absolutely. Like, there's just, you just know less about how it's gonna go, and we've just learned a lot in the last- Right ... seven months or whatever.

[00:53:38] Taylor: Well, and, and part of it is I hear you saying things, even the conversation you had with Matthew, and it just feels like your resourcing models- Like, I just don't, I don't really understand why you want to be in a circumstance where you're doing all this value creation on the front-end demand side and not being able to satiate it.

[00:53:56] Andrew 3: Oh, I don't want that.

[00:53:57] Taylor: Yeah.

[00:53:58] Andrew 3: I'm working on the resourcing problem. Okay. It's just, I'm just not there yet. Okay. It's just the problems are hard.

[00:54:02] Taylor: Yeah.

[00:54:02] Andrew 3: You know? So, like, my hope, I, I never... I want to stop referring you business as soon as I possibly can.

[00:54:07] Taylor: Yeah.

[00:54:08] Andrew 3: But right now, I, I'll take the referral fee-

[00:54:11] Taylor: Yeah

[00:54:11] Andrew 3: over nothing.

[00:54:12] Taylor: Yeah.

[00:54:12] Andrew 3: You know? And so- The last thing I would say- And I'll build a wait list ...

[00:54:14] Taylor: I just think you're insane for not offering more services. Like, I just... The fact that you don't do Google media buying is just absolutely crazy.

[00:54:20] Andrew 3: We will.

[00:54:22] Taylor: Um-

[00:54:22] Andrew 3: We just suck at it still.

[00:54:23] Taylor: But, like, spend five minutes on the internet.

[00:54:26] Come on. You're, you're too intellectually capable for that to be the excuse. Like, be good at Google media buying is literally, like, a three-hour task.

[00:54:32] Andrew 3: That's not true.

[00:54:33] Taylor: Yes, it is. You have a foundational set of knowledge about media buying. It

[00:54:36] Andrew 3: is for me.

[00:54:37] Taylor: Yes.

[00:54:37] Andrew 3: Yes. But what I'm saying is that- That's all you need

[00:54:40] like, y- y- y- yeah. I mean, I... Yeah. We, I think, yeah, we, we will offer more services. It is all a part of it, and right now we're above our plan doing the things that we're doing right now, and our attention is elsewhere. It's just not, it's not a, it's not a valuable enough thing for us right now to go put the time into building a service and documenting and all that kind of stuff.

[00:54:58] But it will be, and it won't be that long before we do it.

[00:55:00] Taylor: Yeah.

[00:55:01] Andrew 3: You know? So.

[00:55:01] Taylor: I think it's a weird, that's a weird definition of value. I think it would be literally financially valuable

[00:55:06] Andrew 3: to do it. It would be. Yeah, okay. But it wouldn't be as valuable as other things.

[00:55:08] Taylor: I don't know.

[00:55:09] Andrew 3: It wouldn't be. You, you... It was way more valuable for me to go find more growth strategists than to go build a, than to-

[00:55:14] Taylor: No way

[00:55:15] Andrew 3: than to build a Google No

[00:55:15] Taylor: way.

[00:55:17] Andrew 3: Yeah, okay. Oh, that's interesting. Okay, well, next podcast. Um- You gotta go. You have to work out. Yeah, yeah. You're not gonna get jacked if you don't

[00:55:21] Taylor: Hiring new growth strategists is, like, the lowest value task you could possibly do.

[00:55:24] Andrew 3: That's-

[00:55:24] Taylor: You should never do it again. How's that?

[00:55:27] End, end, end on that. Okay. All right. Don't, do not do that.

[00:55:30] Andrew 3: Thank you. Bye.

[00:55:31] Taylor: Uh, bye.