Meta’s algorithm now rewards new creative volume, and last year the industry treated that as a race to publish more. Vincent Bourzeix, Founder and CEO of Superbolt, explains why that race backfired and what actually drives ad performance instead. He also breaks down why ad platforms have moved from buying audiences to buying outcomes, and the three roles his team built to make AI useful instead of confidently wrong.
Behind the Expert
Vincent Bourzeix built his growth-marketing instincts at Try the World, a direct-to-consumer food subscription box, where he served as CMO through a stretch of vertiginous growth — the company’s Facebook ad spend went from $5,000 a month to half a million within six months. That experience convinced him the winning organizational shape for DTC growth wasn’t a media-buying team bolted onto a brand, but a fully integrated unit of media buyers, designers, a copywriter, a front-end developer, and a data scientist working side by side. In 2017, as Try the World was being sold, he spun that team out into Superbolt, an agency built on two ideas: multi-expertise teams operating as one unit rather than handed off between specialists, and acting as a strategic partner to fast-growing consumer brands rather than a pure execution shop. Superbolt is coming up on its tenth anniversary, and Bourzeix is now steering it through what he calls its next chapter — leaning further into AI and consulting-style strategy without giving up the agency’s operating core.
The Quick Hits
- Creative volume chased Meta’s Andromeda algorithm past the point of usefulness last year. You cannot produce a thousand high-quality creatives a week, and Meta doesn’t even count a two-word copy swap as a distinct creative — it reads that as the same ad twice.
- What actually drives performance is diversity of concept and storytelling, not raw output. Superbolt runs AI-generated and team-generated concepts against each other, but every concept is finalized and every asset produced by a human — a written AI principle bars any unchecked deliverable.
- Ad platforms have moved through three eras: buying placements for advertising’s first two hundred years, then buying audiences once Meta made granular targeting possible, and now buying outcomes — you no longer tell Meta who to target, you tell it what return you want.
- The old playbook of granular manual control — single-keyword ad groups, hand-tuned exclusions — now performs worse, not better. Platforms reward consolidated campaigns fed with high-quality signal, not fragmented human micromanagement.
- A pixel is table stakes, not a differentiator. Most brands under-invested in first-party data for a decade because “add a pixel, watch customers fall out the bottom” was easy enough. AI can leverage first-party signal — it can’t create it.
- New channels bring a measurement problem before they bring a growth problem. Pinterest under-attributes itself relative to Meta, so a channel can look three times worse than it performs — the fix is triangulating with post-purchase and incrementality data, not comparing in-platform ROAS across channels.
- AI adoption inside an agency needs three roles to stick: a tinkerer who finds what works, a process builder who asks whether it scales to twenty clients, and a data “piping” expert who feeds it exactly the right inputs. Miss any one of the three and the effort stalls.
From $5,000 to half a million a month
Bourzeix’s founding insight came from watching a single brand’s ad spend explode. As CMO of Try the World, he built a fully in-house growth team — not just media buyers, but two designers, a copywriter, a front-end developer, and a data scientist, all working closely together. The reasoning was structural: Meta’s advertising tools had just given brands access to real data and real testing for the first time, and capturing that required creative, media, and user experience to move as one unit rather than as handoffs between specialists.
When Try the World was sold, Bourzeix took that same team and spun it into Superbolt in 2017, built on two ideas: multi-expertise teams instead of siloed agencies passing work down a waterfall, and acting as a strategic partner rather than a pure execution shop for brands that were past product-market fit but not yet at scale.
The volume trap
Last year Meta shipped a new algorithm, Andromeda, that rewards fresh creative volume — and the industry, in Bourzeix’s telling, overcorrected hard. LinkedIn turned into a bragging contest: fifty creatives launched this week, a hundred the next, a thousand the week after.
What actually moves performance, he argues, is diversity — different concepts, different ways of telling the product’s story, different people telling it — not raw output. Superbolt’s answer is a hybrid production model: AI-generated and team-generated concepts compete on equal footing, but every concept is finalized by the team and every final asset is made by the team. One of the agency’s written AI principles is explicit — no deliverable ships without a human checking it. Founder-led ads, he notes, remain some of the best performers precisely because that authenticity can’t be generated.
From buying placements, to audiences, to outcomes
Bourzeix frames the current moment in ad platforms as the third act of a longer arc. For roughly two hundred years, advertising meant buying placements — a magazine page, a subway ad. About ten to fifteen years ago, Meta’s targeting tools shifted the industry to buying audiences: you specified who should see the ad, and got far less wasted media in return. Now, he says, platforms are asking advertisers to buy outcomes instead.
That shift explains why granular manual control — the single-keyword ad group, the hand-tuned targeting exclusion — now hurts performance instead of helping it. Meta’s newer “value rules” and Google’s upcoming natural-language campaign briefs for Performance Max both point the same direction: platforms want consolidated campaigns and rich signal, not fragmented human micromanagement.
A pixel is table stakes
If signal is the new currency, most brands are still under-resourced to spend it. Bourzeix traces this back to how easy the first version of performance marketing was: add a pixel, put money into Facebook or Google, watch customers fall out the bottom of the funnel. That ease meant the industry never built the first-party data infrastructure that would let it feed platforms richer signal today.
The same logic governs channel diversification. Bourzeix isn’t a believer in diversifying for its own sake — Meta on discovery and Google on intent remain the best-measured, highest-quality channels available, and a brand spending under roughly $100,000 to $500,000 a month has plenty of room left within them. Diversification becomes necessary only once the next dollar is genuinely better spent on TikTok, Pinterest, out-of-home, or podcast advertising — and each of those brings a measurement problem before it brings a growth problem, since a channel like Pinterest under-attributes itself relative to Meta and can look three times worse than it performs. Superbolt was one of the first agencies TikTok badged, and among the first things that badge unlocked was measurement-partner credit — a tacit admission from TikTok that its own attribution was the thing most likely to make brands walk away.
Precisely wrong versus approximately right
Superbolt’s defense against AI-generated overconfidence is structural, not vigilance-based. The agency runs what it calls quarterly learning roadmaps — tests built like scientific protocols, typically four to eight weeks long, with the decision rule written down before the test starts: here’s what we’ll do depending on which outcome we see.
AI enters only after that framework exists. It can’t build the framework itself — that stays human — but given a well-designed test and clean data, it turns interpretation from a once-a-month event into a near-weekly one, and the number of hunches the team can chase went from two or three a week to ten or fifteen. Weekly client updates are now roughly ninety percent AI-drafted. What Bourzeix is careful to separate is where AI sits in the pipeline: a data-engineering team still uses code to extract and organize data from the platforms, and only then does AI analyze the result. Skip that step — let AI pull the data and build the dashboard itself — and it will fill in missing context by inventing it.
The three roles AI adoption can’t work without
Getting AI to actually stick across an organization, Bourzeix has found, takes three distinct people working together, not one enthusiastic champion. A tinkerer finds a workflow that works well for one client. A process builder asks the harder question — can this scale to twenty clients, or does it only work because one person is babysitting it? And a “piping” expert makes sure the system is fed exactly the right data, since Superbolt holds its reporting to 99.99% accuracy with effectively no tolerance for error.
That framework is also how Bourzeix describes Superbolt’s next chapter. Historically the agency has run roughly 80% growth agency, 20% consulting — built to produce sharp strategic insight on top of media execution. The next version pushes further: about 50% growth agency, 30% more like consulting, and 20% more like a technology company, with an internal function built specifically to keep leveraging innovation rather than falling behind peers who don’t.
Sound Bites
- “You needed someone very good at creating creative concepts and someone very good at understanding the evolution of these ad platforms. And you wanted continuity of the user experience from the ad to the website, from the first time they see you to the conversion.”
- “You can’t create a thousand high quality creatives a week. It’s not possible. Or you’re doing a change of two words in the copy above the image — which Meta doesn’t think is diverse. It thinks it’s twice the same creative.”
- “We’re in a phase where we’re going from buying audiences to buying outcomes. You’re not saying, I want to show ads to that person. You’re saying, I’m trying to sell those products with that return on ad spend.”
- “I love this idea of signal — how can I coordinate the delivery of signal to these platforms such that they can do a better job? Certainly it can leverage it — you cannot create it.”
- “A lot of people look at — oh, our performance is up today, what happened? Performance is down today, what happened? Oftentimes it’s nothing. This is noise, not signal.”
- “If it doesn’t have the data points, it’s going to make it up. It will give you an answer. It will be very confidently wrong.”
The Chord take
The volume race and the signal gap look like two separate problems, but they’re the same mistake in different clothes: mistaking an easy, visible action for the thing that actually drives performance. Publishing more creatives is easy and countable, so the industry over-rotated on it the moment an algorithm rewarded it — even though Meta itself doesn’t count near-duplicate ads as distinct. Adding a pixel was easy and countable a decade ago, so brands stopped there instead of building the first-party data infrastructure that would let them feed platforms genuinely useful signal today. In both cases, the platform quietly moved the goalposts — from placements to audiences to outcomes — while a chunk of the industry kept optimizing for the previous era’s scoreboard. Bourzeix’s answer in both cases is the same discipline applied twice: separate the noisy, easy-to-produce signal from the one that actually predicts performance, and don’t let AI paper over that distinction. AI can draft ninety percent of a report, but only after a human has built the test framework that makes the report meaningful. AI can generate concepts, but a human decides which ones ship. The pattern generalizes past advertising — anywhere a system can produce a confident-looking answer faster than a human can verify it, the answer is either a framework good enough that AI can safely fill it in, or a team disciplined enough not to trust it until it has one.
Put it to work
- 1Audit your last quarter of creative output for diversity, not volume. If the “new” creatives are a copy tweak on the same concept, you’re not getting the algorithmic credit you think you are — and you’re spending production time you could put toward genuinely different concepts.
- 2Write down your AI principle before you need it: what has to be human-checked before it ships, and what can run unsupervised. Superbolt’s rule — no deliverable goes out unchecked — is a floor, not a constraint on speed.
- 3Treat “is my pixel firing” as the bare minimum, not the finish line. The differentiator now is how much structured first-party signal you can feed back to the platform — audit what signal you’re actually sending versus what you could be.
- 4Before adding a new ad channel, price in the measurement problem, not just the media cost. A new channel will often look worse than it is because it attributes itself worse than your incumbent platform — plan to triangulate with post-purchase or incrementality data rather than trusting in-platform ROAS at face value.
- 5Before trusting an AI-generated insight or dashboard, ask whether it was built on a predefined test framework or generated on the fly from raw data. The latter is where AI fills gaps by inventing them.
- 6If your AI rollout has stalled, check which of the three roles is missing: someone finding what works, someone testing whether it scales, or someone making sure the inputs are actually right. Most stalled rollouts are missing exactly one.
Full transcriptShow ↓
Vincent Bourzeix
You can't create a thousand high quality creatives a week. No, like, it's not possible. There is this idea that came up already two years ago, which was the idea that creative is the new targeting. $5,000 a month on Facebook ads at the time. Within six months, we're spending half a million a month. Today they're 90% automated.
Bryan Mahoney
Yeah.
Vincent Bourzeix
Compared to 10 years ago. And it's missing context. And if it doesn't have a data point, it's going to make it up. A lot of people end up being precisely wrong versus approximately right. AI doesn't help you set the framework — the human sets the framework. But if you give a very good framework and data to an AI.
Bryan Mahoney
All right, everyone, welcome back to another episode of the Brilliant Commerce Podcast, where I get an opportunity to sit down with some of the brightest minds behind what I normally say are iconic — or next-to-be-iconic — brands. Joining me today is Vincent Bourzeix, who is not behind an iconic brand himself, but behind what I would like to call an iconic, or next-to-be-iconic, agency: Superbolt. Thanks for joining me here today.
Vincent Bourzeix
Thanks for having me.
Bryan Mahoney
Normally my first question is, where do I find you today? Because more often than not, I do these remotely. But today I know exactly where you are — we're here in Manhattan, New York City, where the World Cup is. One of the host cities is right here. And if my intel is correct, you may have been at the football match last night.
Vincent Bourzeix
I was there yesterday at the France game.
Bryan Mahoney
Yes. Congratulations, congratulations. Yeah, yeah. So let's not talk too much football, but let me get your bold prediction. France is considered one of the favorites. I don't want to jinx it for you.
Vincent Bourzeix
The favorite. Okay.
Bryan Mahoney
We are not afraid of jinxing it. You're going for it right there. Yeah, so good. Good start to the World Cup.
Vincent Bourzeix
Good start to the World Cup. Don't want to get too much into the game of yesterday, but honestly, first half, nothing was going on, and they got the magic back in the second half. So awesome. Yeah, good prospect.
Bryan Mahoney
All right, well, congratulations. I'm excited that's off to a good start. I'm also super excited — I've been looking forward to this conversation. We had a chance to catch up last week. I've been a fan of Superbolt's work for a really long time. We've had some customers in common, but also just agency in general — I ran an agency for nearly 20 years, and you started your agency in 2017. So you're coming up on that 10-year anniversary. But let's rewind the clock for just a sec. Before Superbolt — tell me a little bit about your background, your sort of history.
Vincent Bourzeix
Absolutely. So before Superbolt, I was CMO of a direct-to-consumer company called Try the World. In the food subscription box world — that was a very popular category, and the company grew very quickly. When I joined, we were spending $5,000 a month on Facebook ads. Within six months we're spending half a million a month.
Bryan Mahoney
Incredible.
Vincent Bourzeix
Very fast growth for the company. And because we were at the early era of these DTC brands and we had raised quite a bit of venture money, we'd build a team fully in-house to work on these new growth levers. And one of the core insights from that time, from an organizational design perspective, was that it was important to have on that team not only people with expertise in marketing and media buying, but also two designers, a copywriter, a front-end developer, and a data scientist — all working very closely together in that growth team. The why behind that is that with the arrival of these new forms of advertising on Meta specifically for that brand, you had access to a lot of data for the first time. You were able to test a lot of variables.
Vincent Bourzeix
And so you needed both someone very good at creating a lot of creative concepts, and someone very good at understanding the evolution of these ad platforms that were moving very quickly. And then you wanted to ensure continuity of the user experience from the ad all the way to the website — from the first time they see you to the conversion. So that was the role of the front-end developer. We saw this regrouping of expertise being very key. In parallel, I became an advisor to earlier-stage companies where it didn't make sense to hire a team of 10 people in-house. So at that time I looked at the agency landscape as an obvious alternative.
Vincent Bourzeix
And for these fast-growing, high-potential consumer brands, I found only agencies that tended to be either good at the media side, or good at the creative side, or good at the front-end side — but not one agency that had regrouped these expertises in a way that would holistically, vertically integrate what became the core of the growth engine for these brands, which was the paid media side.
Bryan Mahoney
Yeah.
Vincent Bourzeix
And so when the company was about to be sold, I went to see five people who were working with me on the team and said, hey, I have these brands that are asking me to solve that problem for them — how about we do a sort of spin-off of our team into an agency? And really with two big ideas: one, this idea of multi-expertise teams. And the second point — I had seen it sometimes be a challenge for these early-stage brands that they wanted help running the media campaigns, doing the execution, creating the creatives, but they also had a lot of questions. They wanted a strategic partner, not just an execution partner.
Vincent Bourzeix
And so that was the second pillar of Superbolt — we're going to bring multi-expertise teams so we can be vertically integrated and help holistically, but we're also going to advise on growth strategy and be a strategic partner to the clients we work with. Because at this stage, it's often post-product-market-fit but pre-scale — that's what they need.
Bryan Mahoney
I mean, that's fascinating — we didn't get into that during the warm-up call. I started our agency back in 1998. Maybe you were in school, but you certainly weren't running an agency. It was this concept of waterfall — you would design something to be pixel-perfect, one agency would do that, and then you'd hand over the design to another agency that would ultimately program it. We thought that was fundamentally broken. So we had this crazy idea to put product designers — they weren't even called product designers back then — alongside front-end developers, and we would work in a way that was iterative.
Bryan Mahoney
Let's sit beside each other, let's work through these iterations, let's cut down that feedback loop. And it felt so innovative at the time — companies just weren't doing that. They were so used to something that was waterfall: we approved this design, and now the design has changed, how did that happen? We were like, well, as we're using it, we found opportunities to make it better. And it seems like you stumbled upon a similar thing, and then you were able to work across a lot of brands to bring that vertically integrated mindset to market. Is that fair to say?
Vincent Bourzeix
That's super fair. And I think what this recruitment of expertise gives you is speed of execution.
Bryan Mahoney
Yeah.
Vincent Bourzeix
And coherence in the strategy. We saw that when you had one agency doing the creatives and passing it off to another agency doing the media — when the two needed to be so integrated — the feedback loop between a team, versus the feedback loop between two agencies, is a million times better within a team.
Bryan Mahoney
Yeah.
Vincent Bourzeix
And for me, I was always obsessed with having the team in the office. Even to this day, where we're a bit more remote — about 20% of our team is remote — we're planning our onsite where everyone comes to New York and spends time together. Because I always believed that one thing that made our creatives so good was that every Thursday we have a happy hour, and the designers talk to the marketers, and they build a relationship. Because it's such a fluid relationship, we have insights flowing very well in both directions. That's really what makes our creatives, at the end of the day, really good at the intersection of what works.
Vincent Bourzeix
But still being on brand, which is really hard to do. I really think it's this sort of flow between the teams that makes that possible.
Bryan Mahoney
Yeah. And I think one of the advantages agencies have always had is that when you work across a portfolio of brands, you can share — not necessarily a playbook, but the ability to see what has worked across a platform. Sometimes when you're just working from within a brand, without that perspective, you only know what you know. And I've always found that being able to partner with agencies — whether I was in-house, or whether I was the agency —
Bryan Mahoney
finding that partner that understands your brand can really get embedded there, but also bring that expertise and perspective you're just not going to have because they've been able to work across other domains, other verticals, and to see those channels emerging.
Vincent Bourzeix
I think that's so important. And, you know, there is a lot of conversation around how much AI is disrupting the agency world.
Bryan Mahoney
Right.
Vincent Bourzeix
And one of the things it is not disrupting is this portfolio view — this much broader spectrum of understanding what works for different brands, especially once you start building category expertise within the agency. That can't be replaced by AI. I think AI is replacing a lot of the execution. When I was starting in 2014, even in my previous role, there was a lot of manual work to be done — a lot of what I sometimes call day-to-day trading.
Bryan Mahoney
Right.
Vincent Bourzeix
It was a lot of day trading on the ad platform — constant optimization, because it gave you a lot of space to operate on the platform. But the optimization was all manual. Today they're 90% automated compared to 10 years ago. That's been changing a lot with AI, but it happened over the last five years — actually that's pre-LLM, more like the machine-learning ad products Google and Meta came up with. Advantage+ on Meta, Performance Max on the Google side. That's changed a lot of the work we do on the agency side. But it's put more importance on the strategic partnership side, and more importance on the creative side.
Bryan Mahoney
Yeah, yeah. There's going to be these hype cycles for sure — AI is going to replace all agencies, AI is going to replace all engineers, AI is going to replace this, AI is going to replace that. And I think, as I've learned over time, you can't overreact to some of these things. I've been on this three-week learning tour, sitting down with companies to understand how they're using AI, and there's this — on one hand, I'm hearing an awful lot that the pace is insane, we can barely keep up, and there's a lot of individuals doing AI within the company, but very little coordinated effort.
Bryan Mahoney
And so, being a strategic partner, I'm curious how you and the team are advising some of the brands you're working with — how they might want to leverage AI to keep them close to their customer, to put emphasis on creative diversity, which I've heard you say and I 100% agree with. How do you advise them on using AI to keep you closer together, without falling into the hype cycle, but also not being defensive about it — like, we're here because AI isn't replacing us — but instead saying, hey, we're learning new ways of working? Would love to hear your thoughts on what that looks like.
Vincent Bourzeix
First I'll start with how we're using AI as an agency, and what our philosophy is around it. We're very much embracing AI. Since the beginning we've been trying to find ways it can help our teams drive more value for clients. And it hasn't changed the main areas where we think we can drive value — again, it's strategic partnership and the quality of creatives, because that's the biggest driver of performance, especially on creative-first platforms like Meta. AI can have a big role on the strategic partnership side. Where it really helps us is saving time on manual reporting and the search for insights. We're very proud, historically, of the quality of our slides and our reporting.
Vincent Bourzeix
It was really focused on driving insights for clients — insights that helped improve performance. Sometimes those insights were given almost directly to their head of growth, so they could pass it back internally, to product teams, to finance teams. We're very proud of that. That took our team a ton of time.
Bryan Mahoney
Sure.
Vincent Bourzeix
And the frequency at which we could drive great insights — it used to be once a month that we got a really strong insight. Now it's almost once a week, because we have so much more firepower in analyzing data and doing drill-downs.
Bryan Mahoney
Yep.
Vincent Bourzeix
Into the hunches that we have — we could do two or three hunches a week before. Now we can do 10, 15 a week if we want to. It's very powerful for the strategic partnership. It's a bit the same story on the creative side. You were talking about volume and diversity — there was a lot of conversation, especially last year, after a new algorithm on Meta was released called Andromeda. All of a sudden there was a big focus on this notion of creative volume and diversity. There's this idea that came up already two years ago — that creative is the new targeting. Basically you used to find people by giving Meta the classics — other brands your customers might like, plus an age demographic, and so on.
Vincent Bourzeix
So, whitelisting groups of people. And what Meta is telling you now is: don't give us a lot of parameters, we'll do it through the targeting, through the creative circuit. That came up two years ago, and we've seen it be the most effective. On top of that, last year they released this new algorithm that rewards new creatives. At that point there was almost an over-interpretation of this, where it became volume at all costs. It was a race on LinkedIn — one week it was, I launched 50 creatives this week; the next week, someone said they launched 100; the next week, someone launching a thousand.
Bryan Mahoney
Yeah.
Vincent Bourzeix
And you can't create a thousand high-quality creatives a week.
Bryan Mahoney
No.
Vincent Bourzeix
It's not possible. Or you're doing a change of two words in the copy above the image — which, by the way, Meta doesn't think is diverse. It thinks it's twice the same creative.
Bryan Mahoney
Yeah.
Vincent Bourzeix
So I think there's been an overemphasis last year on volume. What really matters is diversity — different concepts, different ways of telling the story of your product, different people telling the story of your product. This is why Meta is pushing so much creator content. I was at the performance marketing summit they did a couple of months ago in San Jose, and it was all about creator content. Last year, when they talked about creatives, they were showing off new AI tools to help your team create products from scratch. Not a word of that this year — now it's all about creator content. So I think there was an overemphasis.
Vincent Bourzeix
But even a very diverse set of creatives still takes a lot of time, because coming up with creative concepts is time-consuming for the team. So now we have a hybrid approach where we use a mix of AI-generated concepts and team-generated concepts. Both are welcome to compete, but all of the concepts are finalized by our teams, and all of the final assets from these concepts are also made by our teams. Our teams use AI to augment themselves — our motion designers, graphic designers, copywriters — but there is no replacement. We've written some AI principles, and one of the rules is: nothing that comes out, no deliverable we produce, can go unchecked by a human. Everything has to be human-approved.
Bryan Mahoney
Yeah. We did the same thing recently at our company — it's an incredible thought partner, whatever tool you're using, but you have to own your work, you have to own the output. And you said something at the beginning that was really important — there's no better indicator of success than the quality of the creative. We can fall into this volume trap, and my bold prediction is that as humans, if you think about the way you want to be marketed to, you want to have a connection with the brand, a connection with the creative. That's what brings me back to this notion of an iconic brand — a brand that really knows its customers.
Bryan Mahoney
It's a brand that can create a movement you ultimately want to follow. I just don't think a race to the middle of generic content or generic creative is going to compel me to want to be friends with that brand. But I want to know that the creatives behind the brand are next level — that they're always coming up with ideas, always finding those efficiencies — and I want the final product to be something that really resonates with me. How would you recommend we balance out this idea of creative output, where you're on the record about diversity and humans in the loop being the final arbiters of something that should be impactful?
Bryan Mahoney
If we're creating content that's hyper-personalized, how do you balance that against the desire to create something viral? Because for me virality is more of a community movement — we all saw the same thing, we're all excited about that thing. How do you thread that needle?
Vincent Bourzeix
I think what we're seeing is that the message from the brand needs to be authentic — resonant, sometimes viral, but also personalized. One of the best-performing creatives we see for several of our brands are founder-led ads — ads where the founder tells the story.
Bryan Mahoney
Yeah.
Vincent Bourzeix
And that's the identity of the brand — you can't replace that. You can't have AI do founder content. So what it tells me is people still look for authenticity in the message. Where we see a lot of opportunity for personalization is in cases where we have, for example, brands with large portfolios of products, where we know some products are better suited to certain people — there we're going to look for a lot of personalization, and really try to find the best product-to-audience fit. That's even how we organize campaigns for some of our clients — around the product category, because we know different categories of product go after different people.
Vincent Bourzeix
And in this moment, when we're talking about creatives as a big input for Meta, I think that's part of a broader concept in the evolution of ad platforms. The way they've framed it, which I think is really interesting, is that Meta kicked off a revolution a little over 10 or 15 years ago. For the last 200 years or so, advertising was buying placements — an ad in a magazine, an ad in the subway in New York. Then you go from buying placement to buying audiences. All of a sudden you don't have all that waste of media, and you have a lot more possible attribution. It totally changes the game.
Vincent Bourzeix
And now what they're saying is that we're in a phase where we're going from buying audiences to buying outcomes. So now you're basically not saying, I want to show ads to that person — you're saying, I'm trying to sell those products with that return on ad spend.
Bryan Mahoney
Yeah, this is the outcome that I'm—
Vincent Bourzeix
This is the outcome I'm looking for. And for us, how we're digesting that internally at Superbolt, we think about maximizing performance for our clients. What's the framework? High consolidation, high signal. For a long time the way to perform best on Google was single-keyword ad groups — one ad group per keyword. Extremely granular, extremely detail-oriented optimization. It was a similar story on Meta — highly fragmented, lots of control, lots of manual optimization. If you do this today, you have terrible performance, because that's not how the ad platforms work anymore. Their platforms have become very smart. The signals that go in have changed — on Meta, there was the whole iOS 14 shift.
Vincent Bourzeix
They get a lot less purchase data, so they adapt, they look at other signals. Now the name of the game is to provide as much signal as you can to the ad platform. There was a bit of panic in my industry three or four years ago when Advantage+ arrived, when Performance Max arrived on the Google side — everyone was like, oh man, they're taking away all the controls, and it's all black box.
Bryan Mahoney
Yeah.
Vincent Bourzeix
You don't know what's happening. And now these black boxes are evolving quite a lot — they're becoming a lot more transparent, and they're letting you give a lot more signal into them. An example on Meta: for a while, in those campaigns you couldn't say, I don't want to target anyone below 25. Then at some point it was called bid adjustments, and now it's this thing called value rules — same idea. You can say, for someone below 25, bid 80% as high as someone 35 to 45.
Bryan Mahoney
Yeah.
Vincent Bourzeix
And on Google they've done similar things. At the last Google Marketing Live, another interesting example — soon, for your Performance Max campaigns, you'll be able to give an AI brief. The way you'd tell a story to ChatGPT about who you might be targeting, that sort of blurb, you'll be able to put as an input on your ad campaign, in natural language. So we're going more and more toward high consolidation, really high signal. Today, brands ask me, what can we do to improve performance? To me it's an era where you look at your signals and try to improve them. And people think of signal as, is my pixel working — yeah, that's important.
Bryan Mahoney
That's table stakes, though.
Vincent Bourzeix
But you have all of these other signals you can now provide, and that's so impactful on performance — those—
Bryan Mahoney
Those signals now — you're speaking my love language, because those signals require... that's where the hard work is.
Vincent Bourzeix
That's right.
Bryan Mahoney
Adding a pixel to the site is not that difficult. And I think, as a commerce industry, whether it was 10 or 12 years ago, we fell in love with this idea of adding a pixel, putting a bit of money into Facebook or Google, and having customers fall out of the bottom. It wasn't all that difficult. So we didn't make an awful lot of investment in our first-party data, which is where I think a lot of the signal is. Brands that want to optimize their performance — I love this idea of signal: how can I coordinate the delivery of signal to these platforms so they can do a better job? That's my own sort of secret sauce, and we are going there.
Bryan Mahoney
I don't think AI is going to necessarily do it for us. It can find efficiencies.
Vincent Bourzeix
Certainly it can leverage it. Yeah — you cannot create it.
Bryan Mahoney
Exactly. And I think just understanding first-party data so we can amplify those signals — that's one way. It's going to be a black box that comes and goes. But if we understand who our customers are, and how we can amplify those signals across those different channels, it's a nice way of winning. I'd be curious — you've talked a lot about Meta, Google's been thrown in there as well. Superbolt is known as one of the first agencies recognized as a go-to resource by TikTok. As we shift into more of this outcome-driven economy—
Bryan Mahoney
when I think of these ad platforms that have their own incentives — they also want certain outcomes — how do you think about balancing the impact across these channels so you get the singular outcome you want for the brand, while navigating these multiple channels? Tough question.
Vincent Bourzeix
Maybe it's a good question. One of the big topics since the beginning of Superbolt is this idea of channel diversification.
Bryan Mahoney
Yeah.
Vincent Bourzeix
Because everyone feels over-reliant on Meta. When iOS 14 happened, everyone was like, we need to diversify. Even if you think about a portfolio investment, that's the rule — diversification is everything, and a lot of people feel it's intuitive to look for that. What's been hard in paid media is that the volume of ad placement, the quality of ad placement, and the quality of the ad platform have really only had two tiers for a long time: Meta on what we'd call the discovery side, Google on the intent side, and the rest of the world. That said, at some level, as you start scaling, you should diversify.
Vincent Bourzeix
If you're spending less than $100,000 for a brand with a focused demographic, or maybe $500,000 for a brand with a broader demo — I don't think it's a good strategy to diversify for the sake of diversifying that early. You can diversify within those platforms already. But at some point it becomes necessary, and your next dollar is better spent
Vincent Bourzeix
somewhere else — TikTok or Pinterest, depending on your product; out-of-home; podcast. What that brings is a measurement challenge, because those platforms don't have the same measurement capabilities. Let's say you were advertising on Meta, and all of a sudden you start advertising on Pinterest.
Vincent Bourzeix
Pinterest is not as good as Meta at attributing itself credit. So all of a sudden, if you were looking at your return on ad spend within the channel, Pinterest looks three times worse.
Bryan Mahoney
Because Google takes credit for Pinterest anyway.
Vincent Bourzeix
Google takes credit for everyone.
Bryan Mahoney
Yeah.
Vincent Bourzeix
But that's the point where it requires a conversation about the measurement stack — really creating a measurement approach built to improve decision-making, separating day-to-day decisions from month-over-month decisions. That's very important, because you want to keep looking at your ROAS on-platform to make day-to-day decisions within Meta or Google — those are apples to apples, and that helps you optimize that channel. But you can't use that data the same way when comparing month-over-month budget allocation across channels. When you're deciding whether to keep investing in Pinterest, you need to look at a broader set of signals.
Vincent Bourzeix
If you compare Pinterest ROAS on-platform to Meta ROAS on-platform, you're comparing apples to oranges. So you need to introduce triangulation — start looking at post-purchase data, MMM data, or if you have an MTA platform, start triangulating and understanding the moving pieces. That's when data becomes messier to manage, and a lot of people prefer
Vincent Bourzeix
precision. A lot of people end up being precisely wrong versus approximately right.
Bryan Mahoney
Right.
Vincent Bourzeix
That's one of the challenges. Once you become more sophisticated, the way you solve this — once you have more spend — is by running incrementality experiments to help calibrate your triangulated framework. That's the hardest part. When you go into TikTok, when you go into Pinterest — we're one of the first badged agencies with TikTok. What's funny is that one of the first things you get access to when you do that is measurement-partner credits, because they know the biggest problem they're going to face is measurement.
Vincent Bourzeix
TikTok's not going to get enough credit on its own, so they sponsor — they pay for six months of using this tool or that tool — so you can look at it as apples to apples with your other channels.
Bryan Mahoney
Yeah, yes — they're tilting the scales in their favor. But that's interesting. So on measurement, this idea of being precisely wrong — I'm on the record saying sometimes LLMs, depending on how you ask the question, can be confidently wrong. And as much as building dashboards can be inefficient, sometimes you build them and people don't look at them. I'm of the opinion that dashboards oftentimes don't change people's minds no matter how you've asked the question. And because we now have super-rapid access to answers, if we don't get the right one, we can make decisions with a real negative impact — you're talking about dollars out.
Bryan Mahoney
So I'm curious — understanding Superbolt's approach to QA, being vertically integrated on the creative side, and being really well known for being close to the data — what advice are you giving your team to make sure the same level of rigor you apply to creative QA also applies to insights in the era of AI? Are you applying the same rigor to making sure the answers you're getting are right, not just how fast you can give them to the customer?
Vincent Bourzeix
I think one of the very important parts of how we look at data is how we've set up different tests that let us extract insights from the data. A lot of people just look at — oh, our performance is up today, what happened? Performance is down today, what happened? Oftentimes it's just nothing — nothing happened, nothing changed. If you're not spending a ton of money, one day you'll have 12 conversions, the next day 8. There's no insight there — that's just day-to-day normal variation. This is noise, not signal. So for us, what we spend time on is what we call quarterly learning roadmaps. A bit of a mouthful, but it's basically creating structured tests. My background is engineering,
Vincent Bourzeix
so I like good framework thinking, and protocols like a scientific protocol. For tests where we're trying to test big assumptions, we don't run them over days or weeks — usually over a month or two, four to eight weeks is the average duration for these roadmaps, and they're very well structured. We define, based on the outcomes we're going to see, here's the set of actions we'll take. If we're running incrementality experiments, for example, this is usually the timeframe we use. If you have a good setup upfront on your test, interpreting the data becomes very easy afterward. And that's where AI is helpful.
Vincent Bourzeix
AI doesn't help you set the framework. The human sets the framework. But if you give a very good framework and data to an AI,
Bryan Mahoney
it's going to do a great job.
Vincent Bourzeix
It's going to do a great job at doing the analysis every day. And so, on a weekly basis, we give a little update to our clients on how this big test is performing — and that's written about 90% by AI. We check on it, but that's not where the value is. The value is confidence.
Bryan Mahoney
You have confidence in the setup, confidence in the quality of the data. AI is not going to come in and
Bryan Mahoney
clean up a mess that you created accidentally. If you've got that discipline and that really strong foundation, I think you can look at the numbers more confidently — is my understanding.
Vincent Bourzeix
Exactly. And I think AI shouldn't replace — to your point on dashboards — we shouldn't have AI do what code can do. We still have a data engineering team that uses code to extract data from the platform, format it, organize it, and then have AI analyze the result of that data. But sometimes now AI is used to extract the data from the platform and build the dashboard directly — and it's missing context. If it doesn't have the data points, it's going to make it up.
Bryan Mahoney
Yeah, because it wants to give you an answer.
Vincent Bourzeix
It will give you an answer. That will happen. It will be very confidently wrong.
Bryan Mahoney
Yeah, yeah. I find that over and over again — sometimes it's like, well, it gave me an answer, and I can squint at it, it seems reasonable I guess, and I'm just going to pass this along without stopping for a second to wonder — my intuition tells me that can't be right, but the AI has to be right, and it's not often right. So I think it's about just slowing down for a sec.
Vincent Bourzeix
And I think we need to spend time setting up AI for success. We've spent a lot of time over the years thinking about how to set up our team for success — the growth path, mentorship, and so on. For AI, it's a bit the same. If you don't give AI a lot of context — what's your approach, how do you think about things — it's going to default to the average way of thinking about this, which is not the way we think about it.
Bryan Mahoney
Right.
Vincent Bourzeix
So we've spent a lot of time in the last six to twelve months thinking about how we set up AI tools for success. And once we've set up success at the company level, the next question is how we give those tools, already set up for success, to the team.
Bryan Mahoney
Right.
Vincent Bourzeix
We also try to move away from — we have a few great AI people on the team who form a great AI organization, because the setup sometimes isn't super intuitive. We're not at the level yet in the AI journey where, if you're not a tinkerer, your AI output might be bad — and you've got to want to be a tinkerer. Not everyone wants to be one. That's okay. You're a growth marketing manager, you're interested in the marketing mix of creative and analytics — you might not be interested in AI technology, how to set up prompts optimally, how to extract data the best way. That's okay. So it's a lot about setting AI up for success, and then setting up the people who use AI for success.
Bryan Mahoney
I mean, that — I think more people need to hear that, honestly. I think there's this feeling in the industry, whether in commerce or adjacent industries, that everyone is doing AI and we're somehow left behind. What I'm hearing, and what I'm seeing firsthand spending time within these companies, is that we're very early in that adoption curve. There's an awful lot of experimenting and tinkering happening, but not a lot of coordinated effort to roll it out across the organization. You're coming up on your 10-year anniversary of founding Superbolt, and you've been through multiple cycles or chapters as a company. So as you look ahead to this next chapter, is it fair to characterize it as this?
Bryan Mahoney
Yeah, I don't know how I'm going to characterize it — I want you to. But tell me your vision for the next chapter, and how it's AI-enabled, and then how your customers and your brands are going to benefit from that.
Vincent Bourzeix
I think, yeah, we're calling it Native Enable internally — it's all the same, but indeed, the way we've historically positioned ourselves is that we're 80% a growth agency, 20% operating more like a consulting company. That's why the slides were good-looking and focused on insights — we saw that as a way to represent our strategic partnership position. In the next version, we want to double down on that. So we think Superbolt will look 30% more like consulting, but also 20% more like a tech company, because you need to build a function internally that's good at leveraging innovation. Innovation these days is mostly characterized as AI, but you can expand it a little more broadly.
Vincent Bourzeix
If you don't leverage innovation during this period, there's a high risk you'll be left behind compared to other service providers in the same industry. We'll still be at least 50% a growth marketing agency, with basically the same approach we've had since day one — that close multi-expertise focus and strategy beyond execution. But the 20% working more like a tech company is new, and, to your earlier point,
Bryan Mahoney
We are.
Vincent Bourzeix
At a moment that's still early in the era, you're going to have a few tinkerers. But what we've found is you need a few people working with those tinkerers. You need a process builder — someone who looks at what the tinkerer got working for this client and asks, can it work for 20 clients? How scalable is that process? And you need another person who's the piping expert — filling in the right data. That's really hard. It's a classic bad-input, bad-output problem, and figuring out how to give it exactly the right data is extremely important, because for most of our reports, we want to be 99.99% accurate.
Vincent Bourzeix
We don't have a 10% margin of error we can tolerate. So the piping has to be really good. Once you have these three people working closely together, it accelerates adoption and makes operationalization across the company possible. But if you only have one of the three — if you just have the tinkerer, it's never going to spread. If you just have the process person, they'll create processes that no one uses, which we've done before, by the way — it's a classic, and I think a lot of people get stuck there. And if you don't have the right data in, you can have the tinkerer and the process, but people won't trust the data, so no one uses it. If you don't have all three, it doesn't work.
Vincent Bourzeix
But I think in the future you need your agency to be AI-enabled. And to be enabled, you have to look a bit more like a tech company — at least part of the organization has to.
Bryan Mahoney
Turns out data is important — always has been. When I think about the through-line for all these different cycles or chapters you've led the team at Superbolt through, the recurring themes are curiosity, honesty, and rigor — maybe that speaks to your engineering background — and that you continue to innovate, as opposed to chasing trends or just declaring, this is the channel, this is the thing. What I've always appreciated about your work is this measured and honest approach. And even the way you talk about AI, how transformative it can be, there's a measured approach. It's a measured and honest approach.
Bryan Mahoney
And I think that's the message the innovation industry needs to hear. This is exactly what I thought we'd get out of this conversation, so probably a great place to wrap us up. You've left us with a ton of insight, and I really appreciate you being in the hot seat with me today. This has been fantastic. Thank you.
Vincent Bourzeix
This has been tons of fun. Thank you so much for the invite again.
Bryan Mahoney
Yeah, awesome. Anytime.