Our 120-Hour AI Media Buying Test for Black Friday

[00:00:00] McKay: if your business was off, let's say 15% by noon on Black Friday, like what is the right lever to pull and how many more checkpoints do you have or is your team making to ensure that gets, uh, you know, course corrected and back on track?

[00:00:11] So that was, uh, really foundational for, you know, can we build some type of traffic light system, if you will, to look at hourly, hourly data and, and come up with the recommendations for what we need to happen.

[00:00:23] This episode is brought to you by Miracle Connect. When ads get more expensive, growth can't depend on spending more for the same audience. Miracle Connect helps brands reach new shoppers through leading retail marketplaces with AI-powered catalog tools, more than four hundred and fifty channels, and one place to manage your catalog and orders.

[00:00:41] Find out what's possible at M-I-R-A-K-L.com/ctc. That's miracle.com/ctc.

[00:00:49] Richard: Hey, folks. Welcome to the Ecommerce Playbook Podcast. I'm your host, Richard Gaffin, and I'm joined here today, uh, by McKay Rasmussen, who is one of our senior, uh, paid media buyers here at Common Thread Collective, or se- senior paid media manager, McKay, is that right?

[00:01:04] McKay: Yep, that's right.

[00:01:04] Richard: That's right, yeah. But really what, what I'm saying is McKay is the brains be- behind a lot of how we're thinking about meta media buying, specifically paid media buying in general, as we move into BFCM here.

[00:01:16] And so I've worked with McKay actually directly. We did a Black Friday boot camp a couple of weeks ago, uh, in person at our CTC offices. Uh, McKay was there breaking down paid media for everyone, being a rock star. It was awesome. And so we're gonna have McKay break some of that down, uh, for you all today, and really the way that we're framing this is in 2026, how is meta media buying going to look different than it did in 2025 and in fact any previous year?

[00:01:42] And I think there's some very specific reasons that it's gonna be different. But, um, let's talk overall first, uh, McKay, about... Well, actually, first off, how are you doing?

[00:01:53] McKay: Yeah, doing good. Excited to be here. Yep. Awesome. Uh, I mean, Black Friday's around the corner, so we're getting into crunch time.

[00:01:58] Richard: Yeah, that's right.

[00:01:59] Excuse me. This is gonna be the, uh, the hell week or hell couple of months here for your team. So, um, I salute you for that. But let's, uh-- Okay, let's, let's go ahead and, and jump into a little bit about how this particular, again, this period coming up is going to look different. So tell us a little bit about, yeah, how is 2026 gonna be a brand new beast?

[00:02:16] McKay: Yeah. Yeah, I mean, I think, uh, one of the, you know, the elephant in the room, if you will, is the fact that now we have, uh, Meta MCP access, and, and a lot of the other media channels are, are rolling out, you know, their, um, LLM connections, which is just a way for us to talk to Chat or Claude, um, and, you know, do things in the account, whether that's reading, uh, read access, which is pulling, you know, reporting, or it's write access, which is making changes on our behalf.

[00:02:42] And I think, you know, ultimately, that's going to be, uh, you know, a big, a big shift, uh, in how we operate moving forward, uh, because whereas most of, uh, what has happened in the past, and I have some specific stats in terms of, you know, uh, in terms of, you know, bid and budget changes. Last year, our team over Cyber 5 made over twelve thousand adjustments across the entire portfolio when it comes to bid and budget.

[00:03:06] And so now we think through, well, the decisions that are being made, can we evaluate the accuracy, precision, and timing of those adjustments and whether they're, you know, most optimal? So I, I would say that's probably the biggest change from a year-over-year perspective, um, and that sort of touches all aspects.

[00:03:21] Richard: Yeah. So I-- So I mean, to kind of unpack that a little bit, uh, as we've said many times before, Black Friday is a different beast in terms of how you approach media buying in that you need to make a ton of micro changes across the course of, of, like you're saying, the Cyber 5, in order to take advantage of the sort of waves of demand that happen over these unusual five days.

[00:03:41] So o-obviously, I think it stands to reason that an AI agent could make those decisions better and more swiftly than a human media buyer. That's, that's definitely like a, a drum that we beat a lot on this podcast. But, um, so let's talk about, a, a, a little bit about, um, kind of how we are executing or deploying this specifically at CTC.

[00:04:03] McKay: Yeah, that's a great question. I mean, it all starts with, uh, our belief system. Um, and so at CTC we have something called the Canon, which is just a roll-up of our methodologies over the last, um, you know, 10 plus years of operating at CTC. Um, and, and holding, you know, weekly, monthly Canon meetings where we're aggregating befor-portfolio level data, um, to sort of influence, you know, what we believe should be true for optimal performance.

[00:04:27] So, um, in terms of, you know, how we've started building out the foundation for, um, you know, analyzing Black Friday spend, you know, we looked at, uh, tens of millions of dollars in, in spend, uh, which has generated, you know, hundreds of millions in, in revenue. Uh, and analyzed, you know, what happened last year, how can we, how can we find efficiencies within that process, um, and, and, and build upon that.

[00:04:49] So I think one of the things that stood out to me was when looking at the portfolio data is, uh, you know, when-- because we hold, uh, daily expectations, hourly expectations at CTC, uh, we have a really good understanding of where the shortfalls are by any given moment. Um, and just showing that, you know, if, if your business was off, let's say 15% by noon on Black Friday, like what is the right lever to pull and how many more checkpoints do you have or is your team making to ensure that gets, uh, you know, course corrected and back on track?

[00:05:18] So that was, uh, really foundational for, you know, can we build some type of traffic light system, if you will, to look at hourly, hourly data and, and come up with the recommendations for what we need to happen.

[00:05:30] Richard: Mm-hmm. Um, yes, and that touches a little bit on, on a kind of a broader question maybe, which is, which is, uh, why not just let M-Meta handle it, given that we have ac-access to the MCP?

[00:05:40] Obviously, we're not just running on autopilot. There's a certain amount of human intervention that is involved. So talk a little bit about like what, what that relationship is. Like what, maybe broader, like what are the media buyers doing on an hour-by-hour basis, actually?

[00:05:54] McKay: Yeah, it's a really great question.

[00:05:55] I mean, ultimately, you know, anyone has access to the Meta MCP. Um, now I think the caveat is how you tie in your ad account changes to what's happening on the business level. Uh, and so having expectations around, you know, things like MER by hour, um, and, and being able to even pull in like well what can we expect for revenue by hour throughout the entirety of Cyber 5?

[00:06:17] So there's 120 hours that we have data, uh, collected for in terms of, you know, how much of the revenue by this point in the day should we expect to have. And we can sort of overlay, you know, our business expectations into what needs to happen in the ad account to get us that result. I mean, I think in summary, you know, it's, it's always the context layer is, is, uh, super important because you can't just have meta go off and make changes without also measuring, you know, what's happening on the business side. Um, so you can think of it, you know, similar to how we thought about Statlas, which is the hierarchy of metrics, right?

[00:06:50] Like, uh, we're only making these changes if there's, uh, you know, there's like a decision framework that happens on the back end for, you know, is contribution margin in a healthy place? Is MER in a healthy place? And then, you know, we start evaluating incremental return on, uh, return on spend in relation to our targets and expectations.

[00:07:07] Richard: So, so then I, I, I think maybe the next way to help continue to clarify this or talk about it is just an example, right? So like talk to me about, okay, it's let's say, I don't know, it's 8 A.M. on Black Friday. What are, what is your-- like, what's an example of a situation in which, or maybe how would like the, the media buyer and the sort of the AI agent interact in a given situation to sort of solve or fix or something like that?

[00:07:36] McKay: Yeah. I mean, I think one of the, one of the benefits of leveraging, you know, things like LLMs and automations is the fact that they don't sleep, right? And so, um, where in past years, you know, obviously people are downing Red Bulls and, and trying to stay up as many hours as they can throughout Cyber 5. Um, but the reality is, you know, at some point there's, you know, times where the account's not being looked at.

[00:07:56] And those are the gaps that we wanted to identify and solve for. Um, and so, you know, traditionally, I think, you know, teams during this season are online. Um, however, we are gonna have, you know, 120-hour monitoring. So every hour throughout Cyber 5 is gonna be monitored against expectations. Um, and, and so when we think about how our teams interact with this, it is, uh, you know, presenting the information, uh, in, in a way that is very clear.

[00:08:22] In that way, like, anyone that's looking at the data will inevitably make the same decision. Because, you know, everyone can build dashboards on Chat and Claude, um, but the issue arises when, you know, three different team members are gonna make three different decisions. And so we can actually make it, uh, a bit more black and white, if you will, based on the business context and the levers we have in platforms to make changes for us.

[00:08:44] And so to go back to, to get back to your example, it's 8:00 a.m. on Black Friday. Uh, you know, we re- we know that based on our, our data from our portfolio that we can expect, you know, about 27% of revenue to come in the door by that time. Um, and so then we can just look at our, our target for the day, and if we're not at at least 27% of that target, we know that something's off.

[00:09:04] And then we just sort of identify, well, how off are we? Is it-- You know, think of like a traffic light system. Is it, you know, in terms of standard deviations, is this a, like, you know, 0 to 10% gap that we can close by taking this lever? Is this a more of a red light where like, "Hey, we're like really far off.

[00:09:20] Something probably needs to change, uh, from a larger swing," or potentially evaluating, you know, things like offer. So it's, it's all about, uh, ease of use and, and providing clarity for the team. That way there's continuity in decision-making, and we can be precise, uh, and, and, uh, consistent with them.

[00:09:38] If your ad costs are climbing and growth is getting harder, you need more than another campaign. You need more places for shoppers to discover your brand. Miracle Connect makes multi-channel growth simple with AI-powered catalog tools that turn your product catalog into optimized listings across more than four hundred and fifty retail channels while keeping your inventory, pricing, and orders in sync from one place.

[00:09:59] Expand beyond your own site without building a whole new operation. Find out what's possible at M-I-R-A-K-L dot com slash CTC. That's miracle dot com slash CTC.

[00:10:11] Richard: So what's the difference in, in that example then?

[00:10:13] What's the difference between what we would've just done in any other year and this year with, with the actual agentic aspect of this? Because presumably, if you're doing an hourly check, you can just do an hourly check and discover that you're 10% off. So what's the, what's the difference here?

[00:10:28] McKay: Yeah, I mean, I think the difference is really about the, the consistency and efficacy of the, the changes that are happening.

[00:10:34] So I referenced earlier in the call, you know, last year we had over 12,000 changes throughout Ciber Five. The question is, were those 12,000 changes the, the right levers to pull in platforms? Um, you know, can we, can we re-rely on every single one of those decisions to drive the best, you know, efficiency against target?

[00:10:50] And so trying to standardize that in a way that, you know, again, we're, we're referencing the business, uh, the business targets on top of, you know, what we're seeing in platforms to, to really tie it all together. So if I had to summarize like the differences, the fact that we have, uh, you know, unified decision-making and the fact that we also don't have to wait for someone to make that decision.

[00:11:13] And so if at, you know, uh, you know, 4:00 a.m. potentially before, uh, you know, your, your morning person wakes up to go check on results, if the, you know, if we know that revenue's off and we have an opportunity to lean into one of the campaigns based on the IROAs, uh, and our expectation of where that should be, uh, then, you know, giving the autonomy to an agent to go ahead and make that increase for us.

[00:11:36] Richard: Um, right. The primary thing being like given, the agent does have the authority in some cases to execute if you choose to give it that, and particularly like, oh, in, in the midnight hours or whatever. Somebody internationally is buying your product and you wouldn't have realized it or been able to jump on it, um, previously, you're actually able to do that now.

[00:11:54] Um, okay. So let's talk a little bit about then, and you've obviously alluded to this a few times, but the, the type of preparation that needs to happen before Black Friday. So clearly parameters have to be given in a certain way in order for you to create a situation where you can trust the agent to make those decisions at 4:00 AM.

[00:12:12] So talk a little bit about what those look like.

[00:12:15] McKay: Yeah, I mean, ultimately this is the greatest responsibility of, you know, potentially handing your LLM access to your ad account where, you know, you're spending real dollars. And so, um, just as important as building the system were the guardrails around it.

[00:12:29] And in terms of analyzing the types of changes that occurred last year and the bounds of, you know, reasoning that we can sort of set for an LLM to make a decision. And, you know, for a, a, a very, uh, drastic example, there's no reason why an agent would want to or should ever take, you know, a target ROAS and, you know, move it to a 0.0, right?

[00:12:50] Like there, there are expectations that we have around, you know, the levers we can pull, and so building a guardrail system in place of, uh, in place of that to make sure nothing goes haywire and, and even on top of that, sort of like an alert system, if you will, to just flag an actual human to check things, um, is-- it was crucial during the build-out of this.

[00:13:09] And on top of that, I think, you know, as people move in this direction of autonomous, you know, buying, et cetera, is to test it ahead of time, right? Like you, you wanna get a good understanding of how it works, uh, before you go into peak. Uh, and so that's, yeah, that's, that's a-

[00:13:24] Richard: Okay. So part of what we're, um, what we're doing here is sort of testing a hypothesis, right?

[00:13:31] Like this idea that the incorporation of the MCP or agentic buying really on BFCM is going to make an impact. So, um, talk to me about like how, what, what it is, what the experiment looks like right now. Like what it, what it is you're hoping to sort of prove, uh, or the hypothesis you're testing.

[00:13:51] McKay: Yeah, I mean, I, I think, uh, Cyber 5 is such an interesting time to, uh, you know, overlay this, this methodology ultimately because there are such drastic differences from a day-over-day perspective.

[00:14:02] Um, I, I have a hypothesis, at least at this point, that it actually requires more human intervention than other moments throughout the year, right? Yeah. And so-

[00:14:11] Richard: Yeah ...

[00:14:11] McKay: if in evergreen, you know, period, we, you know, are sort of, you know, have really good understandings of what our target, uh, efficiency expectation is, you know, we're essentially trying to prove out whether more intervention that is, uh, you know, unified under a clear framework is going to lead us to a better business result than, let's say, you know, just letting Meta, you know, do what it does in terms of setting the target.

[00:14:35] Now, we do have, uh, we do have data to, to point to that, right? In saying that, again, our team made 12,000 changes last year. Um, so we know it is a time of year where there are, uh, many, uh, I would say it is a time of year where the volume of changes is increased, and it's trying to, you know, find efficiencies within those changes, right?

[00:14:54] Because there are so much-- uh, there's a lot of variability in, in how things get enacted based on what needs to happen. And so ultimately, the hypothesis, you know, again, is like hourly decision framework around where are we at, where do we need to go, and what levers should we pull is, is what we're after.

[00:15:11] Richard: Yeah. So then let's kind of-- I think we can kinda wrap up talking a little bit about practicals. And, and one thing I like to ask is like, if somebody doesn't have access to our tools, what's one thing they can do to get closer to that and start preparing for it? Obviously, I would say if you want to execute this in the way that we're executing it, the best way to do it is to become a client if you're a seven or eight-figure business.

[00:15:31] So commonthreadco.com, you hit that hire us button. You let us know you're interested in us building something like this out for you. Um, but talk to, for somebody who's not in that position or not currently a client, like what, what is the practical, tactical approach? Like, what's the first thing that you should do to get set up for executing something like this?

[00:15:49] McKay: Yeah, it's a great question. I mean, ultimately, it all starts with, with your beliefs, right? And so I would say because anyone can connect, you know, their, their MCP to their ad channels and their Shopify, uh, I would, I would start by forming a belief about what your expectation is for this Cyber 5. Uh, and you can even reference, you know, what changes you made at what times last year, um, and start to dig into some of the, the questions as to like, well, was this change, you know, impactful for the business?

[00:16:15] And so it all starts with the forecast. Uh, you know, start there, and once you have your expectation for what should happen, which is influenced, you know, by, by your marketing moments, et cetera, um, then you can start to overlay on, you know, the decision framework that you can try to automate. , And so I think once you have the forecast and you can reference even last year's data, having a good understanding of what the reality looked like last year, uh, and this is assuming, you know, your marketing moments are staying the same, uh, in terms of, you know, w- what-- uh, how much revenue can I expect per hour throughout all of Cyber 5, um, is, is a good starting point because you can overlay this year, you know, again, 8:00 a.m.,

[00:16:52] uh, Black Friday morning. Well, what happened last year by this hour? How much revenue was in the door? If you're not there, like should start to raise some alarm bells for you, uh, a yellow flag if you will, uh, and then just determine the weight at which, you know, you, you need to course correct to, to close that gap.

[00:17:07] So it all starts with your beliefs and the foundation of, of your forecast, um, and then you layer in the action layer from there.

[00:17:15] Richard: Right. So yeah, I think that that's a, a good way to sum it up. That like there's... As we talked about before, the initial sort of forecasting prep belief part, which is determining what you think is likely to happen based on historical data, and then running on an hour-by-hour basis against that with sort of human oversight.

[00:17:32] So, um, okay, cool. Well, anything else you wanna hit, Mikah?

[00:17:35] McKay: Uh, no, I think that really covers it. I mean, ultimately, I, I would say like dig into, again, like with the MCP, meta MCP, um, you'll be able to have more clarity around like what was-- what happened last year. And so we analyzed a lot of things in terms of, you know, from, from post-peak day, from Friday to Saturday, from, uh, Cyber Monday to Tuesday, like what were the observed changes, because we know that after the peak, that conversion rate drops and, you know, how can we best, uh, protect our, ourselves and the accounts from, from overspend or inefficient spend.

[00:18:06] Um, so having again, again, a good understanding of what those peaks and valleys are will help you overlay some, some decision-making framework for what we think needs to be true. Um, and similarly, you know, even on the tactic level, when it gets that into like, you know, what, what budget, uh, adjustments are we making?

[00:18:22] What bid adjustments? I think we did a lot of analysis into, you know, the hierarchy of, you know, what actually produced better efficiency in terms of, you know, 24-hour windows and whether it was, uh, bid or budget. And so, um, I would, I would tell, you know, anyone looking to replicate something like this, uh, is to have a really good grasp on what has happened in the past.

[00:18:40] Not that it will happen in the future, but it at least gives you a starting point to, to build some foundational work.

[00:18:47] Richard: All right, folks. Well, I think that's gonna do it for us. So, uh, if you're interested in us building this out for you, uh, you can work with us as, as a seven or eight-figure brand. Hit us up, cometthreadco.com.

[00:18:56] Hit the Hire Us button. We would love to chat some more. But all right, appreciate it, McKay, and thank you everybody out there for listening. We'll see you next time.