How to Build an AI Advertising System That Scales Without Losing Brand Trust

Hannah Parry: So thanks Evan for the the brief intro.

Hannah Parry: I'm Hannah Parry.

Hannah Parry: I head up performance creative at Wayfair. I've been here for a little around like a year and a half at this point.

Hannah Parry: And I've really been building the PC department from the ground up, so from scratch, both as a net new team, but also as a net new function. Uh my team sits at the intersection of our fantastic performance marketing department, but also our great creative studio team. So today I want to share with you how we approach performance creative, uh and of course AI generated ads at an enterprise scale. But the truth is for all of you in this room, enterprise or not, I know that every brand in 2026 is being trapped in a huge pressure cooker right now. So we're all being asked to produce more content, launch across even more new channels. You need to move faster, but of course, don't forget to never damage the brand equity. So today I'm really excited to show you how we broke that bottleneck at Wayfair by moving away from making just singular assets and moving towards engineering systems. So let's dive into it.

Hannah Parry: So when you're supporting thousands of assets on a monthly basis across 10 plus channels, you really can't treat every ad like a blank canvas or you're going to be feeling like this squiggly mess that I drew here where you're floating in the abyss, always having to start from scratch. We instead had to really transform our workflow architecture. So now our designer builds core swappable components based off of our top performing ad library. And then we let AI engines handle all of the low lift repetitive work. So like our versioning, resizing, translating, and even all of our aspect ratios. So to the right over here, I have a few examples of some of our uh heavy hitting statics. We have what we call is our uh customer testimonial static and then also our X plus reviews statics. And then I also pulled out two really fun Gen AI videos where we actually shot a very low fidelity version of this on set. So think of like a door frame with just some products on some paper boxes. And then we use that as reference frames in order to build out this fully generative AI uh video that I'll play for you that we versioned across a bunch of different personas for our uh last year's holiday campaign. So hopefully this behaves and I will play a couple of these for you.

Hannah Parry: So that one as you could see was all about that woodsy cabin persona. And then I'll show you this one. Oh. So that one as you could see was all about that woodsy cabin persona. And then I'll show you this one.

Hannah Parry: And that one doesn't want to play, but that one's all about the cozy cottage essentials. So you can see how we usually use the framing and the setup to create a bunch of different variants of this based around different aesthetics or different product curations. So AI hasn't just changed our speed as a small nimble team, it's also of course changed our architecture and how we work as a team.

Hannah Parry: So to protect our team's creative sanity, we really try to use what I like to call is our 80/20 operating model. And this is more of a mindset versus us literally crunching these numbers every time we do a campaign. But we really strive to have 80% of our output utilizing automation to turn out these necessary high volume variations that we know in partnership with our performance marketing team that we can depend on the performance. So over here on the right hand side, you're seeing a selection of some of our tried and true formats that we're currently running as part of our American Summer campaign. So everything from like a full bleed static with and without copy, our what we call is our collage format, and then even a static animated GIF. So each of these assets at this point, uh we've automated in order to scale this across any campaign. And really we're doing that with a myriad of tools, but really via custom Figma plugins as well as even AI assisted computer vision. And so this really frees up the rest of the team's time and that 20% of our bandwidth for pure human imagination. So this is where we're really allowing our designers and creative strategists to test new ad formats like this kind of fun Polaroid we did with a partnership with Kim Wolf, but also we'll dive into some highly experimental, more surreal concepts that, you know, used to be cost prohibitive or impossible to do before Gen AI. So we aren't just making more as a team, but we're unlocking creative possibilities that simply would have never existed before.

Hannah Parry: And as everyone in this room knows, Gen AI is incredibly powerful, but of course, it's absolutely not magic. So in our testing, we've really found that it excels at two extremes. So for lower funnel ads, it's been a massive win for rendering beautiful contextual environments around of course a real product. But conversely at the top of the funnel, this is where it's fantastic for scroll stopping, more surreal whimsical videos that we tend to really push during our peak holiday moments. So the biggest lesson we've learned though on the opposite end of where it's not as successful is truly about customer trust. So early on, we saw that customers were rejecting fully AI generated uh inspirational room scenes. So think about if you're looking for home inspo on Pinterest, all those very clearly uh Gen AI aspirational images. So we had tested some inspirational images that weren't necessarily tied to real furniture and did not see strong performance from that. And in a similar vein, we've also seen that art direction and product curation has become so, so important in our AI generated scenes because if you're providing inputs like for example, a specific color palette, like I really want orange and blue within this room, AI will really take that input and run wild with it. And before you know it, every, you know, art piece, throw blanket, uh throw pillow in the room has that same color palette, which then immediately makes that image or curation feel AI generated and doesn't resonate as well with the consumer. So really our job at the end of the day has been understanding that the product must always be the hero and take center stage. So that we really have to manage and silence all of those other uh variables or noise.

Hannah Parry: This is my favorite visual. A little bit overwhelming, but also hopefully exciting to see a lot of these examples of what we've been up to with Gen AI. Uh but really this uh highlights that exact tension that I just went over. So on one hand, scaling beautifully in the lower funnel when it's what I like to call is invisible AI. So rendering really gorgeous contextual backgrounds around of course a real piece of furniture selection of furniture or products. So I'll play there's a few videos on the screen as well where you can see how we're even exploring Gen AI to create, you know, before and after time lapses of how a space is being uh built out with a variety of Wayfair products or even using AI to build stop motion videos to really showcase functionality of different products. So I'll play the first one here is really that time lapse. So this was a part of a um home renovation campaign that we were running. And dynamic localization, that's a little bit self-explanatory, no video here, but how we're using AI to create um custom product curations that speak to uh different areas around the uh states or even abroad, right? Because we have global marketing teams. So how are you creating curations that are speaking to where different people live based off of what they're purchasing and how they're decorating their space. And then product functionality, I'll show this other video. So this is like a stop motion. I don't know why some of these don't want to play. But basically the it's showing in stop motion the um doors are kind of opening and closing within this cabinet. And then kind of more of a static uh before and after as part of our home organization campaign. And then another invisible AI that we're very uh heavily experimenting in is within rapid trend activation. So as soon as you see a trend going, you know, viral in the home space, how can we immediately activate on it without having to do a fully fledged production. So if we go over to the right side of this page, in the mid to the upper funnel, it's really where again, we've seen the opposite. So we lean a little bit more towards what I like to coin as the term of visible AI. And this is where we have a lot more fun with, you know, surrealism and having more whimsy with some of the video assets to really stop the scroll and build awareness, especially during major seasonal moments or we've also been testing into it for emphasizing uh quality as a trust lever within one of our Wayfair verified branded campaigns. So I'll test hopefully all these videos will play for you guys. Um so for surreal product showcases, we're showing again a variety of curations. This one had a really fun day to night transition as well. Part of our outdoor campaign. This is a pretty fun quick gift for our Halloween campaign last year. And here is that Wayfair verified kind of quality to the extreme. Always love these cute little sleeping bears. And then I already played these two for you, but I know you don't really get to see one. So hopefully they'll both play now. You can kind of see the difference between them. So again, using kind of modular systems to build out curation and videos for different audiences. And then some really fun atmospheric scenes for our last year holiday and Halloween. All right. So at the end of the day in home retail, we really see that scale and materials must be true to life. So again, that's really the focus of everything we do when it comes to generative AI.

Hannah Parry: And on that note, any commerce, and I mentioned this a little bit before about customer trust, but customer trust truly is your currency. So our unyielding rule is one to one product integrity. So the environment can be synthetically enhanced uh to tell a story or like I mentioned before, quickly activate against a trend, but the product itself has to remain an exact true to life representation of material, texture and scale. And so within this really quick workflow I threw together, you can see how we go through the process of curating products based off of honestly a lot of different inputs. So something like style, which is what you're seeing here for a glam based living room, but we're also looking at inventory levels, right, of the products and even price, making sure that it feels a good uh value and quality item based off of value for the Wayfair customer. And then we're pairing all of those inputs with inspirational photography from in-house. So trying to again, match our branded visual direction, uh and then even pairing that with imagery from the wild in order to generate a net new lifestyle image, which is what you're then uh seeing here in the center. So once this is live, which then you're seeing a screenshot uh of the ad, I think this is a screenshot from Facebook. We really this was the top performing asset from this round of testing. And deeming this successful not only due to the performance, but also we received a lot of positive sentiment uh on the ad itself, which you can see on the comments below, uh versus the comments we were getting when we were testing more of that, like I said, Pinterest style aspirational imagery uh that wasn't tied to exact products and caused confusion once people were landing on the landing page, even though we communicated that it was meant to be inspirational.

Hannah Parry: And now that I've shared that rule, I'm going to immediately break it. But I I bet you didn't notice from the slide before that there is actually a minor hallucination uh at the edge of the couch cushion. So within the lifestyle image, the edge of the cushion has like a piping uh type of end to the cushion, whereas the actual product that this was modeled after has more of a downward slope uh that goes into like the um studs uh within the bottom uh of the cushion. So I'm saying this because if we go backwards in time before where, you know, my team and the company is today, when we first started testing these room generations at scale a little over a year ago, we had to develop our own AI quality compass based off of what is still passable to the human eye. And that's really how you maintain brand standards without slowing down the pipeline. You need to stop chasing uniform perfection and start matching your polish to the placement. So a temporary paid social asset, such as what you're seeing here, does not need to require the same level of retouching as something we're putting on our website or, you know, something within a video that we're putting on our broadcast TV placements. What it does require, however, is speed and relevance. So that's where you want to save your human first heavy lift resources for the enduring high stakes brand storytelling uh and placements that you're doing. However, if we now zoom back forward in time to the current day, we do now have one to one product accuracy locked in. A lot of that is due to AI assisted QA pipelines that we've now built. And so our quality compass as of today has become much more nuanced, right? And what we deem as passable to the human eye is usually more about the increasingly complex contextual backgrounds that we're now generating.

Hannah Parry: So for everyone that's listening, I know you're like, oh, this all sounds great, but how can I really get started? Uh so if you're trying to jump in as soon as tomorrow or Monday morning, what I always say is to stop, you know, thinking about building files and start thinking about building your flywheel. And so this is a very boiled down, but exact representation of the infrastructure that we're using to marry the art of design with the analytical science of business growth. So I like to call this the three easy A's. So architecture, audience and analysis.

Hannah Parry: And we'll dive into more detail on each of these in the next slide. All right. So first you want to make sure that again, you're architecting your layout files into modular swappable component layers. Uh you're building both a scalable but also a performative design system that can then be accelerated and automated via AI systems. And then second, you want to make sure that you're testing your narrative angles early against uh real audience behaviors, either via real or synthetic market research platforms. So that you're validating your concepts and messaging uh early. So I have an example here. These are all from our current back to school campaign uh that's live at the moment. And we had a dilemma where we were working with our go to market team on wanting to launch assets separately for parents and for students. So speaking to two different demographics. And instead of guessing what they wanted to uh hear from us and what they wanted to see from us and related to going back to dorm or dorm shopping, we actually used a um AI assisted market research platform where we were able to uh go through messaging and concepts with real parents and real students uh that were in our uh core demographic. That way we had a exact direction that would then limit, of course, the amount of assets we would need to do because we knew the exact messaging that we needed to go after. Um but it also then enabled us to have this locked in before we even put a single dollar behind the marketing campaign. And then lastly, creating a tight data loop is so, so important. So using tools like Motion, shout out to Motion, uh to decide which specific modules to double down on and which to iterate on, which then feeds right back into your architecture step. And I always like to say the biggest misstep I see here is not including your designer as an integral piece uh of this step. They should be provided with, but also understand the performance of the creative assets, not just assigned the next design ticket. Uh you want to make them be autonomous and making confident design decisions and provide net new concepts, ideas grounded in the whys behind uh what has performed well in the past.

Hannah Parry: And that will make your team feel very much in lock step. So I'll leave you all with this. This is where I always love to say that working in the field of performance creative takes the heart of a designer and the mind of a growth marketer. So when you build data driven systems that value and protect the core creative craft, creative can become the most powerful lever for growth in your entire organization.

Hannah Parry: So thank you so much everyone for listening and thank you to Motion for having me. If you're interested, I have a QR code here if you want to scan and add me on LinkedIn or follow up with any questions. And also because I know there's so many amazing creative strategists in the room, I also wanted to shout out that I actually currently have a role live on my team. So I'm hiring for a uh senior creative strategist, more so focused on a longer form video channels. So think uh YouTube, CTV, even upper funnel paid social. So if you or someone you know is interested, I believe we might be dropping the link to the JD. So please check that out. Um we are ideally looking for someone who is willing to be based out of Boston or someone who will at least be able to travel uh quite a bit, definitely for all of our productions. So, yeah, check it out if you're interested and thank you so much. I know I was hoping we'd have a little bit of time for Q and A, but we only have a few minutes, so maybe one question, but thank you.

Evan: From Lake Como to the motion stage, blessing everybody with the amazing content. And we have roles for people. Everybody, big round of applause in the chat. Let's show some love. Let's drop the emojis, all that good stuff. Hannah, we don't have a ton of time to get to questions here, but I think the final thing I'm curious about is like, is there any final words of advice that you have for our audience today?

Hannah Parry: Final words of advice. Honestly, again, I think it would go back to making sure you have a a strong foundation of that, but it becomes very clear what AI tools to focus on when you identify, you know, where you kind of need some efficiency gains. It's really understanding where your pain points are in that flywheel and then addressing it with AI tools that can support you with that.