← Marketecture: Get Smart. Fast.
Marketecture: Get Smart. Fast.
Jackie Paulino on How Pixability’s “Pixie” Uses AI Agents to Transform YouTube Advertising
00:17:48
Loading summary
Transcript65 lines
- [00:10]
A
Welcome to marketecture where you can get smart fast with in depth interviews of leading technology executives. I'm Ari Paparo. I am joined today by Jackie Paulino who is the chief product officer of Pixability. Jackie, thank you so much for being here.
- [00:24]
B
Thank you for having me. Super excited to be here.
- [00:27]
A
Yeah. So Jackie, you gave me a little preview of your last product before it came out. I appreciate that. Love to see things before other people. Makes me feel like I'm special.
- [00:37]
B
You are. You are. Yep.
- [00:39]
A
So cute name Pixi. Tell me about Pixie.
- [00:43]
B
Yeah, it's our first agentic product that we've brought to market. So super excited about. Helps brands and agencies and our teams at Pixability curate inventory on YouTube easier. So we've seen it's a challenge for brands to understand where their ads are running and curate inventory specifically on YouTube and know which videos you're running on, which channels you're running on and how to curate that inventory. So we built a curation tool years ago and we've been using AI to curate to contextualize YouTube. And so now what we brought to market about a month ago now is an agentic technology that helps do that. Right.
- [01:26]
A
So the core business of Pixability is a lot like that. It's just now you've AI enabled it. We skipped ahead a little bit. For those who aren't that familiar, what is kind of the bread and butter for pixability?
- [01:37]
B
Sure. Yeah. So we are basically all things YouTube. So we help brands and agencies get the most out of YouTube from planning, buying, reporting, channel management, all things that help, you know, get outcomes out of YouTube. Really? Yeah, we work with brands direct as well as their agencies.
- [02:00]
A
YouTube's kind of a big deal and people forget it.
- [02:03]
B
Exactly, Yeah. I mean we, we think of it sometimes it sounds funny to say that we concentrate just on, on YouTube, but really YouTube is just like saying we work on the Internet. You know, it's like it's just as big as the rest of the Internet. It's like. So, you know, it's certainly enough business for us to focus on and certainly enough challenges to, for us, you know, on the product team to fix.
- [02:25]
A
So what are the challenges like for someone who doesn't buy a lot of YouTube myself? Like, what if you were to talk to a media buyer planner, what are their complaints and challenges with YouTube?
- [02:35]
B
Sure, yeah. I mean they span. I mean one thing that's interesting about YouTube is it's really is full funnel. So some people have challenges of, you know, growing, growing their Brand awareness. And some people have challenges of like, how do I improve my return on ad spend. Right. So the challenges really span and then you know, the tools that are given to solve those challenges are not fully exhaustive. So that's why third parties like pixability exist, is to help solve those and fill those gaps that are difficult for Google themselves to fill. You know, there's really two ways to buy YouTube ads and that's through Google Ads and DV360. And what we've found is Google Ads is really purpose built for search DB360 really purpose built for display. I just finished your book actually Ari. So I'm like very deep ingrained in all of the background of all these things. So that was actually a very fun, very fun read. But yeah, so neither really purpose for YouTube and that's kind of how we developed our product over the years is if we were going to build, you know, buying platform planning platform for YouTube, how would we do it? And so we're kind of a layer on top of those two buying systems.
- [03:49]
A
Yeah, YouTube is the majority of the spend on DV360 yet it's sort of an afterthought. It's funny that way.
- [03:56]
B
Yeah, exactly.
- [03:57]
A
Right. So I guess so. Okay, so agentic is a hot topic. Is it just applying AI to the problem or is there an actual agent involved at doing something?
- [04:08]
B
No, no, we actually have an agent. Yeah, Pixie is an agent. So we've been applying AI for over a decade. So the idea of AI is not, is not new to us or new to the market. And we've been applying gen AI for over a year. So what's new and what the core difference between gen AI and an agent is? An agent makes decisions and it actually does things for you, right? It actually takes the actions.
- [04:30]
A
Yeah.
- [04:31]
B
And that's what Pixie does, is so we have 1200 different ways that you can curate inventory on YouTube and that's difficult for humans to remember. 1200 different ways to do. Right. So we have 1200 different ways to curate across over 30 million channels. So that results in trillions of data points. And there are things that humans are good at and things that computers are good at and computers are good at searching and understanding trillions of data points. And so that's where the agent really shine. And so with human input. So a prompt based approach and we even take like a media brief directly from a brand or an agency, upload that and then Pixie decisions. Agents are made up of different tools. So we've Integrated different tools and trained them with our data from pixability and that, you know, could be searching the web to understand trends about your audience. You know, might want, you know, moms who are going to buy meat on Sundays. Again, agents and web search tools are very good at understanding what other things those you know, women watch right now. It's not just food review videos and then we have a lot of proprietary data flexibility that then again I can search and understand and fill in those 1200 thanks with again input from.
- [05:54]
A
So maybe be helpful. Can you give me a side by side like what would, what would you do manually with these 1200 variables to reach the female know grocery shopper and what would you do with the AI?
- [06:05]
B
Yeah, yeah. So the way we used to do things, the, the manual approach is doing some research, understanding the brief and obviously that comes with some, some human bias which is good and bad. Right. There's, there's some, I understand what moms do. I have friends who are, you know, fit this demographic. And I'm going to kind of think about what they would do. I'm going to go search the web, I'm going to go to the website of the brand, I'm going to understand who their competitors are, going to go to YouTube and look at their YouTube channel. So we have, you know, we do still have and we will always have a sales strategy team that is that human input that's doing that. And so that's what that team needs to do. And you know, that takes, you know, half a day to understand and kind of get in the mind of a new business pitch and understand, you know, their shoes, etc. And then the new way of things is take all that information and let an agent distill that information for you and take those actions. So put in the prompt. It's not one and done. We could probably have a whole separate podcast about this. But I don't think the role of agents is like to replace 100%. And you know, as a product leader I actually have realized that trying to get to 100% hurts product adoption. So again, probably a totally different topic that we could dive into.
- [07:20]
A
But that's interesting.
- [07:22]
B
The goal I think of agents broadly should just be to get to 80%.
- [07:26]
A
And let humans customer adoption or are there some customers who just don't want it? So it's 80% of all customers will be using AI or is it everyone wants it but they only want to use it for 80% of the job.
- [07:36]
B
Yeah, 80% of the job. Okay, 80% of the job should be the goal because you still want, like to have that human input. So Pixie chats back and forth. It doesn't just. You never ship, you know, think of it as if you're writing an email. You're probably going to go to ChatGPT and ask it for some ideas, but you're going to put your own flavor on it and you're going to read it, hopefully before you send it.
- [07:58]
A
So does Pixie have like a personality or a gender?
- [08:03]
B
We do refer to her as a she. We have sort of gendered her as we joke around about what personality we should train her to have. But we haven't gone too deep. We. We're also launching Pixie for Reporting. And so you can imagine Pixie for Reporting could be a little bit sassy. Yeah, Really?
- [08:24]
A
I was thinking Pixie reporting would be more down to earth or accounting.
- [08:28]
B
Like, he was like, well, this isn't working well. Like, this is really. You got to put more money here, you know? Yeah. But right now she's pretty neutral and she is basically an accountant. Yeah. All right, that's a good personality description.
- [08:43]
A
All right, so let's step back a little bit because you're sort of an expert in video and what's going on out there. So AI seems to be coming at video from all different angles. You have Sora, you have Facebook vibes, you have, you know, what's the Google one is a funny name. The three Banana or whatever it's called. So what's going on on the creator side, distribution side, consumer side?
- [09:06]
B
Yeah, I mean, I think this is completely changed in the last two weeks. I mean, if we had this interview two weeks ago, I feel like it would be very different. But again, a lot of these things launched, you know, in September, October. So it's changing very quickly. And I'm as a human a bit concerned about how easy it's going to be to make content using AI with very simple prompts. If you've played around with Sora, it is alarming. It's very easy to put your friends in compromising positions and it looks very real.
- [09:45]
A
Martin Luther King doing a DJ set the other day.
- [09:47]
B
Yeah, yeah. I mean, changing history is a whole nother thing that's very also concerning. But, you know, from a pixability point of view or from an advertising point of view, I see that as the new brand safety challenge. Brands probably do not want to support content that is malicious and AI created. There's a lot of bad content. But when you can put people into situations that look Very real and it's very difficult to understand. I mean I have a hard time now I think again a month ago, very easy to tell. They're really easy. This is AI created tells now much more difficult.
- [10:28]
A
So on that subject, do you think brands are going to think of it as a category like I don't want to be on AI videos or are they going to evaluate it on the same criteria that they would human videos?
- [10:41]
B
Yeah, it's a really great question. We've been having the same exact conversation internally and I think it's probably the latter. I think there's a lot of really positive and creative fun AI created content. You know, these are tools that make it easier for a lot of people to make content, which is great. You know it, they're more accessible. You don't need in depth training or expensive software now. So I think brands will see that as a positive. And there's some fun stuff out there of giraffes riding skateboards and like that's fine. But yeah, I think it will fit into, you know, there will be nefarious AI generated content that, that people want to stay away from.
- [11:25]
A
Yeah. Do you, what do you think brands will feel about the licensing? Will, will they be cautious about being next to content that is not clearly licensed?
- [11:36]
B
Yeah, I think so. I think they don't know what to think yet. But I think they're the, the public on the publisher side, on the I created these characters side. That's certainly going to be a whole new, you know, copyright issue. I'm seeing a lot of that on Sora right now of characters, cartoon characters, very good at creating your own scenes and there doesn't seem to be a lot of copyright restrictions on that yet. I'm sure that will change. And same thing, I put myself in Hogwarts with my dog walking up and that's fine. So maybe that kind of content is again, I'm supporting, you know, types of movies and TV and, and, or and there's things that are, are negative. So I think brands will have a hard time in the next few months navigating what is good and what is bad. And I think, you know, hopefully people like us can help.
- [12:30]
A
Yeah, I, I think it's analogous to YouTube. In the early days of YouTube they, they got a massive lawsuit by Viacom and I think that was kind of the, the moment that a lot of people waited and see was they were, they took a wait and see attitude towards like, well let's see what happens with the Viacom lawsuit before we commit.
- [12:49]
B
Right Yeah, I agree. I mean, I think there is going to have to be legislation around this and that we'll certainly see a lot of copyright lawsuits coming pretty quickly, I would imagine.
- [13:01]
A
So is, what are you seeing in terms of, as a YouTube expert, are you seeing a lot more AI created videos on YouTube as content? Are they becoming popular?
- [13:11]
B
Yeah, I think they are becoming popular. And this isn't new, right? I mean, we've seen a lot of AI generated content over the last couple of years. I think the difference is how good it's becoming and people being able to discern if it's real or it's not real. But Google's been giving creators tools to make AI generated content or AI supplemented content with things like AI generated backgrounds and those kinds of things to create shorts content really easily. I mean, we've seen shorts as a video type accelerate tremendously over the last couple of years. And a lot of that is, you know, by easy tools to either create content using AI or change content from, you know, different horizontal to vertical. Those kinds of things, those kinds of tools that make it better user experience.
- [14:04]
A
Do advertisers see shorts and regular videos as different media?
- [14:08]
B
They, I'd say it's going away a little bit. When shorts started, everyone was like, how do I, how can I get in the Shorts feed and what do I do? And I need to be there. And now I think it's just more integrated across YouTube's like ad products. And so they just see it as, it's part of YouTube now, but it's.
- [14:25]
A
Mostly vertical versus horizontal. So you need different creative and probably shorter creative.
- [14:29]
B
Exactly. Works better, right? Yeah. But again, like from an AI driven ad format point of view, Google sort of does a lot of that optimization for you where you're like, oh, if you have vertical ad and it's short, like we're going to put it in the shorts feed.
- [14:43]
A
Yeah, that's where the rubber hits the road on AI right now. Like doing something like that immediately benefits everyone involved.
- [14:50]
B
Yeah, exactly. Yeah. I mean, and obviously Google's leaned into a lot of that and they're leaning into a lot of their AI more driven ad formats, which we didn't talk about up front when you asked me about challenges, but that's certainly one as well is it certainly squeezes the most performance out of YouTube and they're great ad formats for performance based, but they come sometimes with a lack of transparency and the ability to understand where your ads are running or choose where your ads are running. But oftentimes advertisers don't necessarily care if I'm trying to get the best outcomes. So the advertisers that work with us tend to want a balance of those two things. And that tends to be. The challenge is like, I want the best performance, but I want to know where I'm running, and I want to know that I'm running within the guidelines that my brand has set.
- [15:38]
A
Gotcha. All right, let's do our quick lightning round. We'll ask you some brief questions so you can give me some relatively brief answers.
- [15:45]
B
Okay.
- [15:46]
A
What's your number one competitive advantage?
- [15:49]
B
Data. That was a little too brief. Okay. Too brief. Okay. All right, all right. Got it. Yeah. Like. Well, We've worked on YouTube for over a decade. We've been working with partnerships to enrich our data set. And so now as we're moving into this, you know, agentic world, AI world, the differentiator becomes data. How can we train these agents to be the smartest and get the best out of YouTube? And so data is really our core.
- [16:15]
A
And you don't have unique data about YouTube, do you? Or you just have the public, what Google's willing to give out and then just use it different ways.
- [16:23]
B
And we have partnerships outside, and then we've also run a lot of ads so we understand what works best and those kind of benchmarks that. That would be unique to us. And then the partnerships that we've built.
- [16:33]
A
Got it.
- [16:33]
B
Also become kind of a unique data set.
- [16:35]
A
And what's on the converse? What's your biggest market challenge?
- [16:38]
B
Keeping up with all the changes. I mean, it's sort of the most exciting part of my job as chief Product officer is understanding that this landscape is like, I mean, a year ago again, a month ago. Everything is so different. So I've been at Pixability for a really long time, but it feels like every day is a new challenge. So that's super exciting.
- [16:57]
A
Yeah. And last question. If Pixability was an animal, what animal would it be?
- [17:02]
B
Elephant. You know, we're lots of data. We're big, thoughtful, smart, and we've got a great team around us. I feel like elephants are, you know, herd animals.
- [17:17]
A
You heard it here. Pixie the elephant. You need, like, a plush toy to give out to your sales team. And the marketing's writing itself.
- [17:26]
B
Perfect. Yeah. You got to change the logo and.
- [17:28]
A
Yeah, absolutely.
- [17:29]
B
Pixie can be an elephant. Love it.
- [17:31]
A
All right, Jackie Paulino from Pixability, thank you so much for joining us.
- [17:35]
B
Thanks, Ari. Super fun.
- [17:44]
A
Sam.