Bringing a Generative AI Product to Market: RJ Talyor, Founder CEO, Backstroke and Pattern89 (Part 2)

Sramana Mitra: So let’s go to the point where you’re starting the first company. Tell me what was going on in the market, what product angle did you take, and why?

RJ Talyor: At that time, there was a huge proliferation of social channels. We had Facebook and then Instagram, Twitter, LinkedIn, Pinterest, and Reddit. Then there were all these other social channels that were coming out, and marketers were trying to experiment with their dollars. At the time, there weren’t any rapid experimentation tools on the market to help marketers figure out which ones worked and what messaging worked.

So, we built a solution to help them rapidly experiment with thousands of versions of their ad spend across six social platforms. Then we built up a data co-op during that time because we asked each of the marketers who used our platform to share their metadata anonymously.

Over time, we ended up with seven thousand brands who had connected into our data co-op, so we didn’t actually have to run the experiments. In fact, our AI could predict what campaign would work best, specifically what image or what video or what text would work best across platforms like Facebook, Instagram, and Google. So, we pivoted away from running the actual experiments.

Instead, we’d ask the marketer for the parameters of the experiment, and then we would predict what went, so would it be a man with a beard and glasses in front of a bookshelf, or would it be a woman in a hoodie or in a sweater in front of a window? These are the types of things that marketers are trying to figure out, and they run experiments to do it, but with our platform, which was called Pattern89, they didn’t have to run the experiment. They could just put the parameters in and then predict which one would work, and we were over 95% accurate.

Sramana Mitra: Can you double click down on the technology, the AI technology that was doing this kind of prediction? Elaborate on what signals was it drawing? Since the company is sold, can you open the Kimono a little bit and educate us on how you did what you did?

RJ Talyor: Every marketer has to decide what image to use or what video to use or what text to describe a product. So let’s say we’re selling this gray sweater. Marketers are just trying to figure out the question of should the gray sweater be kind of a lay-down, just a picture of the sweater, or should it be on a man, or should it be on a woman, or should it be on a man with a beard or glasses.

So, we used computer vision to understand what are the attributes of a picture- man, beard, glasses, brown hair, smiling, not smiling, and other such types of dimensions. Then we also used all the natural language processing (NLP) capabilities to tag all the headlines as well as any text that was appearing in the ad overlay or in the text headline or the body copy.

Then we would understand things, like was it a certain number of nouns or adjectives, were there certain keywords, was it a dollar sign or a percentage of, exclamation points, all caps, all these different dimensions go into either performing or not performing as it’s measured by a click, a conversion, a purchase, or a return on ad spend.

So marketers would say, “I want to get X return on ad spend, what image should I use? Or I want to get X cost per click. Which of these images within my recent photo shoot would work?” So those are the questions that we could answer using a combination of computer vision, NLP, and the metric that the advertiser cared about.

This segment is part 2 in the series : Bringing a Generative AI Product to Market: RJ Talyor, Founder CEO, Backstroke and Pattern89
1 2 3 4 5 6 7