We Killed the Growth Map! Here’s What Replaces It

[00:00:00] Taylor Holiday: It's really hard to, for any media buyer, me, I. Ferris. You know, pick your favorite media buyer on Twitter, your current employee, to prove that their work is better than everybody else's.

[00:00:16] Richard Gaffin: Yeah.

[00:00:17] Taylor Holiday: Because one of the sad things about our industry is that the actual relationship between the actions that we take and the corresponding response or the outcome, like it's never documented anywhere.

It's never aggregated. Like no media buyer brings their resume and is like, here's my change history across all the accounts I've ever managed, and here's what happened when I made all of those changes definitively and provably. 

[00:00:43] Richard Gaffin: Hey folks. Welcome to the Ecommerce Playbook Podcast. I'm your host, Richard Gaffin, Director of Digital Product Strategy here at Common Thread Collective.

And I'm joined today as I always am

[00:00:51] Taylor Holiday: on things

[00:00:52] Richard Gaffin: our CEO here at Common Thread Collective. Mr. Taylor Holiday. Taylor, what's going on today, dude?

[00:00:57] Taylor Holiday: Well, today we are saying goodbye to

[00:01:01] Richard Gaffin: Yeah,

[00:01:01] Taylor Holiday: sits behind me. The, life's work fading into the distance just like me soon, I'm sure. we are evolving,

[00:01:09] Richard Gaffin: I.

[00:01:10] Taylor Holiday: and I think it's gonna be a great thing for of the world.

Less so for my my artifact here.

[00:01:15] Richard Gaffin: Yeah, yeah. There. Yeah. In a, in a shocking turn of events. We're talking today about killing the growth map. So actually Taylor sort of moved outta the way here. If you can sort of duck for a sec, those of you on video can see. This is a, a printout of the growth map, which is of course, like the thing that defines Taylor's life.

But like any great artist. We're getting rid of the old thing and bringing something new in. So what we wanna talk to you today about, we'll kind of maybe take a second to remind you exactly what the growth map is and how it functions, and then the specific reasons why we're getting rid of it and replacing it with something new and better.

So let's just jump right into it. Tell, tell us about what the growth map is and then like let's start to get into why we're getting rid of it.

[00:01:53] Taylor Holiday: Yeah, so I think all of that intro probably is a lot of inside baseball. For

[00:01:57] Richard Gaffin: Yeah.

[00:01:58] Taylor Holiday: we're really describing, which is today's conversation is gonna be about AI is affecting our business and how it, we think it's going to affect your business in particular as it relates to media buying and execution. and what some of these tools and their evolution represent that I think is gonna be really important and good news for a lot of people. And I'm gonna coin a new phrase today that I'm going to hopefully use a bit more often, and it's the phrase circadian me media buying. And one of the things that I've noticed about the, the biggest distinction. In my interaction with AI that I feel how my humanness limits my usefulness as it relates to many tasks.

[00:02:41] Richard Gaffin: Mm-hmm.

[00:02:42] Taylor Holiday: And so the, the joke here about the idea of a circadian rhythm is that each day I go through a sequence of very human specific. Activities that are like sleeping and eating and waking and hormone release that affect my emotions and all these different things that are part of my circadian rhythm. And what happened over the last 10 years at CTC we built a lot of tools and processes that are related to that human rhythm. This growth map is an example of one of them. So the growth map at CTC sort of functioned as this centralized workflow, where every day at a specific time, our database automatically pushed into a spreadsheet through a really cool process we built called RSGS. A lot more exciting news coming on that later about how you can benefit from that and cancel your super metric subscription forever, but a little teaser, but it would auto-populate the data. teams would come in, they would look at their individual tab, depending on their role, check out what happened yesterday. Share a note into Slack to the client about Here's what happened yesterday and here's what we're gonna do about it. They would update their, the expectations of go forward, so change the forecast for every campaign in their media plan. Make sure that they're on budget and target. Discuss if they needed to launch any new pieces of creative.

And all in all, just ensure that you are tracking on plan. So the growth method is just a plan. It's a detailed view of a daily expectation of every metric and then how you performed against it. And there was a process where that data showed up at a certain time, which was intended to be right before you would get to the office, where you would look at the data and then your decision making flowed out of that interaction. None of that that I just described. any consideration for whether or not that's the right time to make a decision?

[00:04:28] Richard Gaffin: Right.

[00:04:29] Taylor Holiday: happens to be when we would wake up and get to the office and view the information and make a decision as a result of our humanness. And I think this is like a, it's a understood, I haven't heard people really talk about this much, but that like our work revolves around that human reality.

We wake up and sleep and our work fits in between those things.

[00:04:45] Richard Gaffin: Mm-hmm. So then let's talk about then what, what do you believe media buying should look like? This is something that's going on 24 7 or like Yeah, unpack that a little bit.

[00:04:54] Taylor Holiday: so if you think about like a, a simple extreme example is that errors in meta system where there are like sudden changes in performance bad or good, necessarily connected to that circadian rhythm. They may happen at one in the morning, they may happen at. 10:00 PM and if that's the case, then there is no human present in observing the information. and so that obviously can be handled with rules and it's sort of a very basic example, but the idea is also inside of the daily rhythm. It's not just sleep. I have to use the bathroom. I go to lunch every day at 1230. I have other responsibilities. I text with my wife. I cannot and never will be omnipresence and attentive to the information to take advantage of every opportunity that exists and prevent error, every error as it's happening. What AI represents is the potential for omnipresence. Let's set aside omniscient because we recognize that it's built on a, a sort of collective knowledge set that is limited. It doesn't yet possess all the cures to cancer. So it's not omniscient yet, but it is omnipresent and I actually think this is the underappreciated attribute about it.

So much of AI's attention gets focused on how much it knows its capacity to process large amounts of information. You'll talk, hear all the conversation related to how many GPUs are there and how big are the clusters. it's actually the presence to the data that I find to be most interesting in that it can be always watching the ad account.

And

[00:06:22] Richard Gaffin: Mm-hmm.

[00:06:23] Taylor Holiday: been a transformative thought process for us to think about.

[00:06:25] Richard Gaffin: Right. So obviously, presumably the right way to mediate buy has always been this, and so let's talk like a little bit about some of the changes just beyond, I mean, I guess, I guess it's ai, but maybe the specific AI changes that have happened within Meta that makes now the time to get rid of this old circadian human workflow and replace it with the AI one.

[00:06:44] Taylor Holiday: Yeah, well, I don't, it, it changes in meta are part of it for sure,

[00:06:47] Richard Gaffin: Mm-hmm.

[00:06:48] Taylor Holiday: In that they are moving more consolidated meta based approach. And if you listen to, Zuckerberg on Ben Thompson read his interview with Ben Thompson and Eckery, you'll hear Mark sort of. Describing this future, which is basically just tell us the objective and get outta the way we're gonna do everything that's targeting the creative, all of it.

[00:07:06] Richard Gaffin: Mm-hmm.

[00:07:06] Taylor Holiday: definitely like a movement towards that for sure. But it's also just the like, workflow of a media buyer. So if I think about those jobs that we just described, even the process of like sending a daily message about performance yesterday,

[00:07:19] Richard Gaffin: Mm-hmm.

[00:07:19] Taylor Holiday: if you go look at CTC right now, every calendar of a media buyer has a block in the morning of their work called One Map.

Here's the problem with that as like, again, a very human limitation is if I work in San Francisco, okay, I can do my best at CTC to try to ensure all of our clients. Are in San Francisco

[00:07:39] Richard Gaffin: Mm-hmm.

[00:07:39] Taylor Holiday: on the West Coast. But the reality is that like you never get that perfectly mapped. So that means that 9:00 AM for them might be 12:00 PM for somewhere else, which means that they receive that note after he comes into work, assuming he or she is on time or isn't, and sends that message.

And it may or may not be useful relative to where it's received. So the ai, let's just now, let's just take that and say, all right, rather than a human going into the growth map, looking for the information. Sending a message, depending on the time zone, you can eliminate all that.

[00:08:09] Richard Gaffin: Mm.

[00:08:10] Taylor Holiday: omnipresence of the information can make it so that we can ensure each individual and recipient receives a perfectly tailored message every day in the same structure relating to the same information at the exact right time all the time. Okay. And now that's just the communication part. Okay. Another big thing that we made media buyers do was we. Build these media plans that we call trackers. Okay. So when a, CTC, the way the rhythm works is if you, once the growth strategist builds the budget, so we build a budget off of our spend models that set the total budget.

We use an MMN to work on the channel level allocation. So then a media buyer will receive four meta this month your spend goal is $200,000 at a 3.4 Ross, whatever. Okay. Let's just assume that it, they receive a set of inputs. Then their job is to map that spend to all of the campaigns that we plan to run. The ones that are already live, as well as the ones that we plan to launch in the future, which come out of the creative concept log so that they can get a view of where every dollar in their channels gonna be spent every day. Okay. So what did we ask them to do? We asked them to do some things that, again, humans are really bad at.

We're we're gonna ask them to make predictions about the future. where they would look at the historical spend of every campaign. They would predict it out every day for the future, using day of the week effect, and we started to introduce some tools along the way to help them with that. Then they would look at what day's new campaigns are launching.

So there's a go live date in our concept log that says, okay, the campaign A is going live on this day. much spend do you think the new campaign's gonna get? That's a really hard question to answer, especially in a cost controlled environment. Well, as we killed this spreadsheet where a human was going in and literally making those adjustments every single day to what happened yesterday.

Am I wrong or right? How does that change my prediction in the future? AI gives us like this sort of better ability to computationally abstract a way that individuals need to make those predictions. And now in our trackers inside of Compass where we're building this in stats, every campaign spend is predicted and updated every day on a machine learning feedback loop that improves it over time.

[00:10:19] Richard Gaffin: Mm-hmm.

[00:10:20] Taylor Holiday: So, and that's not just for like the campaigns that are already live, it's also for the campaigns that are going to go live in the future. We can understand the marketing moment, we can understand how many assets are in the campaign, and then we can make predictions about the spend level, the, the efficacy of performance against that efficiency target, where now we can the, like one of the biggest critiques of cost controls, which is just like utter nonsense.

But setting that aside, is this like a sort of ability to predict the volume? Well, we have. Thousands of inputs every day where we're guessing and being right or wrong, guessing and beat, being right or wrong. And so our ability to improve on a machine learning learning basis, each of those inputs over time is just gonna go exponentially up from where humans were doing it, where you had one media buyer siloed to their individual ad not gaining any of the collective knowledge of the ability and accuracy of that across a broader spectrum. And again, all of this is now enabled because all this data sits in a clear database that funnels into a clear process that gets executed again. These

[00:11:14] Richard Gaffin: Yeah,

[00:11:15] Taylor Holiday: more

[00:11:15] Richard Gaffin: I was gonna say, yeah, one of the, the drawbacks of the old system, like, as I understand it anyway, was that particularly like when you were gonna say like, Hey, we're gonna pull this campaign from last month into the current month, and I. At that point, the media buyer sort of had to guess essentially like how much ROAS do I think that this is gonna get?

And it would have to be sort of a ballpark, like, well it got 1.9 last month, it'll probably get 1.7 this month, or something like that. So that there, there's this sort of like that flimsiness is just inevitable when you have a human being having to make those sorts of decisions without the proper tools.

So that's at least one way that this is changing. But it sounds like to me that like the AI's ability to. At least mimic decision making is the type of thing that's transforming this. It's like the decision doesn't have to be made by the person anymore.

[00:12:03] Taylor Holiday: Well, and the, and the broader context of us

[00:12:06] Richard Gaffin: Yeah.

[00:12:06] Taylor Holiday: just hand the data to an LLM and say. this versus, and then to work through that interaction versus having to come up with the algorithmic expectations ourselves and be great at modeling and all these things. So it just makes you a great data scientist faster, right?

Like we have

[00:12:21] Richard Gaffin: Right.

[00:12:21] Taylor Holiday: scientists on our team, but also it just gives us a partner in collaborating by giving them the data and working through it. So, so if you think about like the sequence here, there's planning. Okay. Which is a big part of what we do at ctc. Then there's like day-to-day updating and execution against that plan.

So tracking against the progress, but then the end tail of this is still the set of corresponding decisions that a person has to

[00:12:40] Richard Gaffin: Mm-hmm.

[00:12:41] Taylor Holiday: And this is where all of us should be way more hopeful because here, here is a truth. And anybody who tries to reject this man,

[00:12:48] Richard Gaffin: Sure.

[00:12:50] Taylor Holiday: it's really hard to, for any media buyer, me, I. Ferris. You know, pick your favorite media buyer on Twitter, your current employee, to prove that their work is better than everybody else's.

[00:13:06] Richard Gaffin: Yeah.

[00:13:07] Taylor Holiday: Because one of the sad things about our industry is that the actual relationship between the actions that we take and the corresponding response or the outcome, like it's never documented anywhere.

It's never aggregated. Like no media buyer brings their resume and is like, here's my change history across all the accounts I've ever managed, and here's what happened when I made all of those changes definitively and provably. everybody can give you an anecdote. Everybody could tell you they were on an account that scaled up a bunch of money.

They can, they can Corolla, they can create a correlation between their presence and their participation in the outcome. But the causal factors are basically non-existent. Nobody has them. And so what we're trying to do is to, and what makes me sad about that is that I, it matters to me. It actually matters to me that our work is good and right and I don't wanna do something that's just a made up fugazi moving of buttons around. And candidly, that's what we're engaged in most of the time.

[00:14:05] Richard Gaffin: Mm-hmm.

[00:14:06] Taylor Holiday: people are engaged in most of the time. And so what this offers us the opportunity to do is to now tie together qualitative action so we can bring in the change log into our database. This is really important. all of the actions, bid up, budget down, launched new ad, changed targeting, adjusted optimization setting from seven day click to one day view and to say like, okay, across all of these actions, what happened after? Was the action good or bad? And then the, what is the cumulative value of all of the actions that we're taking and when? And then that can help to inform which actions we should take more often. And we can look at the frequency of actions. 'cause I've even talked about this. I have a sort of a hypothesis that. Well, my hypothesis is really that there's no way we're taking the right number of actions right now.

[00:14:51] Richard Gaffin: Mm-hmm.

[00:14:52] Taylor Holiday: way too many or way, way, way, way too few. That's like, sort of what I believe. And, but we can, we can sort of start to test this idea and to understand how much interactions with the machine there should be. But it's not a human just going in there and randomly doing it relative to their circadian rhythm or when they happen to look at the ad account, which is what happens right now. like someone, I gotta know, an argument on Twitter with, with somebody that was like. Tried to use this example where they said like, well, what about this example where we had an ad going off one day and we went in and we raised the budget from here to here, and then two hours later we raised the budget from here to here, and we got all this more volume out of it. it's like, well, you're actually illustrating the exact problem to me, which is that you went in, you picked a random tranche of increase from A to B, and then you saw that that was working. And so you went, oh, increase again from B2C. que like, what if you could have just gone from A to F. Automatically without you having to ever show up. How do you know you picked the right levels along the way? You don't, you're, it's always this like sort of guess and check guess and check against relative information in a way that's not aggregated. It's not clear. And so this is like, it's a real, real problem that of this we're trying to take out of people's hands, out of a spreadsheet where people go in and try and analyze the information themselves to a set a, a database structure that's being constantly monitored by these agents.

[00:16:10] Richard Gaffin: So you've already obviously hinted at a lot of this stuff, but let's talk about then what the, the replacement for the growth map is. What the, the tool looks like, at least right now to kind of start to bring this non circadian media buying into life.

[00:16:24] Taylor Holiday: Yeah. So we're, we're building all of these things into, technology, right? So if we think about everything from the messaging, so we built an agent that we call Scout, that is a blue panda bear. There's a inside joke about why he's a blue panda bear or it's a blue panda bear.

I don't know that we've gendered our panda bear yet, but that now provides the messaging and that our clients are gonna be able to interact with in Slack that you're, that sits on top of our data and provides that regular recurring messaging. Then we took that spreadsheet, that this tab right here, That, that we called the Facebook tracker that shows every campaign in projected spend. And we automated it into stats. We, it's inside of our tool called Compass, where there's now the connective tissue between the creative planning process, the ad predictions and the daily forecast of spend to be able to see like, okay, are we gonna get to our target or not? And then the last step, the place we're going. Eventually is to actually allow it to make actions on its own.

[00:17:21] Richard Gaffin: Hm.

[00:17:22] Taylor Holiday: or I would say there's two steps. First is like we've automated the information out of a spreadsheet into technology automated the forecasting from a And given that the next step is to, rather than asking the human to analyze this large set of information, to derive the insight to say campaign is capping out on budget. recommend you change it to make sort of a recommendation set of insights. That's the next step. So go from the human analyzing the data to the human reading, the insight, and then choosing to act on it or not to the final step, is it just taking the action on its own and reporting back the sets of actions that it's taken.

[00:17:59] Richard Gaffin: Yeah.

[00:17:59] Taylor Holiday: so I think there's a sequence of behavioral steps here that we'll go through as we sort of kill the spreadsheet and enable the AI as we go.

[00:18:08] Richard Gaffin: Yeah. So what then is just sort of like, this is thinking more long term obviously, but like what's the future of the human being? I. Within this world, like what, what decisions are left to be made?

[00:18:19] Taylor Holiday: yeah, so the, what's the role of the human one is, I think we are still discovering that.

Again, I, you've heard me maybe say that the

[00:18:25] Richard Gaffin: Sure.

[00:18:25] Taylor Holiday: area between these things is, is hard to see, but, but one of the things I know for sure is that. There's still the qualitative understanding of the client's goals, desires, and outcomes that inform the inputs of the system. So the reason, again, I think cost controls are such a, we have such an odd relationship with the idea.

All they represent to me is my input into the algorithm about what I care about. That's all

[00:18:50] Richard Gaffin: Mm-hmm.

[00:18:51] Taylor Holiday: just me letting meta know this is my desired result. And I want that in consideration of your delivery. And I think that's a lot of how we'll interact with systems is like. This is my desired outcome. Can I represent that you're progressing towards that? Can I give you feedback of that? If you read, again, going back to that Ben Thompson and Mark Zuckerberg, he, he, he's like, the end state is like, you just connect me to your bank account. Tell me what you want to do and I just grow the bank account. Like it's all, but, so if you think about that as like an end state, what you represent there is like, what is the desired outcome?

[00:19:22] Richard Gaffin: Mm-hmm.

[00:19:22] Taylor Holiday: and then communication with people because they're still very human. And then a lot of creative work still sits in this process. That then becomes a more and more of the focus where if we look at the change history, one of the goals that we have. Is if we built a relationship between, let's say, changes related to budget bid and setting versus number of new ads created, could we alter that ratio to where way more of the changes are more creative volume going out and way less of the changes are about bid and budget?

As an example. So that's like, an a way in which more in the short term, more human comes to life. What happens long term with creative is still TBD, but I think these are all steps along the way. But, but the intel is still, we have to get on with the customer. We have to justify what happens. We have to be accountable.

A lot of what I hear, like the human job is, is accountability. It's ultimately you are responsible and these are your tools. So your job is to stand there and accept the result on behalf of the system.

[00:20:17] Richard Gaffin: Mm-hmm.

[00:20:18] Taylor Holiday: like that's literally the job. About it sometimes. Like what? What's my job in my child's like schooling.

It's like I have to go to class and hear the teacher say, this was the result of the report card, and like be there and be accountable to the interaction. And in some ways there's like a bit of that, which is that this is my system, I represent it. You hired me a human. And if you're disappointed, you're not gonna place the disappointment onto a non sentient emotionless.

Computer, you're gonna place it onto me and I will absorb that and I accept that. And so I think there's a piece in there that is the, the human elements of truly what we are in client service that will always remain

[00:20:52] Richard Gaffin: Yeah. All right, cool. So I mean, I think, I think that kind of covers it. Is there anything else that, that you wanna hit on this or like draw out maybe as sort of like a broader

[00:21:00] Taylor Holiday: just say is

[00:21:01] Richard Gaffin: insight?

[00:21:01] Taylor Holiday: to you as brands is that you need to begin to demand demand. Omnipresence, like the idea that you wouldn't have 24 7 monitoring of your ad

[00:21:10] Richard Gaffin: Mm-hmm.

[00:21:10] Taylor Holiday: unacceptable. It's gonna be unacceptable, which means a human can't do it. Like that's just not going to be possible.

This is too dynamic of a never ending marketplace, especially as you're global. Like if, if your brand is distributing ads globally on many platforms, it's just impossible without it. So I think that's the case. And then the question is. will be the standard by which you determine the causal effect of the actions the person in the people in your organization are taking, how, how high can we raise that bar of obligating, all of us to the truth that the work we do is impactful? And I think that's, that's the system we wanna create. We wanna create the most accountable system to the most value and, and. I hope in hearing me say that what, what that means is that you get to come and just like we're publishing our results of our forecasting, we've done that on the last episode.

We do it with a global accelerator. We want to hold ourselves accountable because what that does is that means when we're wrong, we'll fix it. And, and I think that is the highest value of the system. And so just like why does forecasting every campaign, every everyday matter, it's because we get way more tries.

We get better, faster. And it gives us an opportunity to continue to improve the system at, at a rate that will outdo everybody else so that we can continue to be the best possible partner.

[00:22:16] Richard Gaffin: Awesome. All right, well, I think that'll do it for this week. So of course, if you wanna be a part of this part of this change that we're making a part of, kind of the new system that we're talking about here, you can always go to common thread code.com, smash that hire us button, let us know that you wanna talk.

We would love to have a conversation with you. All right, until next week, everybody. Take care.