Realeyes VP Shares Insights on the Aakash Gupta Product Growth Podcast
Keith O'Brien
Our VP of product Scott Jones recently appeared on Aakash Gupta’s Product Growth Podcast, where he discussed several important topics like the genesis for Realeyes Verify, scaling AI in facial recognition, and what’s coming next.
How Realeyes Verify came to be
We already had a pretty deep relationship with a consumer app company related to work we do in several fields, including AI data collection. We built a competency for collecting data for models that could train computer vision systems, build avatars, etc.
We had a deep relationship with a company doing that and other things. And we have very powerful intelligence gathering from our co-founders where I joined in the Fall of 2022 and my focus area for the first several months was finding scrappy startups that wanted to innovate with attention and emotion. Games, for example, can see the attention and emotional states. Wellness, online education apps, as well.
We quickly saw startups were mostly trying to survive. We learned the consumer app company had a need for a face verification model - but couldn’t build their own for PR reasons. We were able to capitalize on that information and get ourselves in the running for this RFI. We went from 0 to 1 on a face verification model in three months.
We won the RFI, partly due to the provenance of our data, which no one else has. We then looked at what other problems we could solve with this new competency. Because of our core business - ad testing - we could participate in online market research.
Developing and Testing MVPs
You build up this vision of this beautiful solution we could build. But you may not be able to afford to build that right now. It’s the classic thing - I [want] to build the Ferrari, but I can’t justify the Ferrari yet. What you need is a car that you can get out into production. I can build the Ferrari once I get enough money flowing; so you think what's the least I can do to build a car that is fast enough. It just has to be scalable enough to solve this problem and make customers’ eyes light up.
|
...fraud had gotten really bad; that 30% to 40% of what you're going to collect is likely at minimum going to be attributable to bots, click farms, or people lying about who they are. |
Tackling Online Fraud and Bot Detection
The stakes are high in market research because you get bad data and the insights and outcomes you get will be wrong and misleading. If you’re supposed to have a very particular audience segment giving feedback or um giving inputs, but it's the wrong audience, that's a problem.
It was a problem we were experiencing firsthand and we knew the entire online market research industry was also suffering from this. No one could solve it from the conventional tools and techniques so we put together a hypothesis, a demo, and a point of view on it.
We have a healthy Rolodex from our work in this market research space, so we were able to get in a room with the C-suite of the world's largest panels so the supply side of online research and buyers.
They all agreed that fraud had gotten really bad; that 30% to 40% of what you're going to collect is likely at minimum going to be attributable to bots, click farms, or people lying about who they are.
|
I have not seen anyone in our swim lane of attention, emotion, and, now, the identity piece I’m working on... |
On What’s Next
We’re in a unique position with lightweight models that can run on the client side - that hadn’t been possible before with real-time telemetry. There’s a whole new world opening up where you can have hardware and software see users in real-time and make interesting decisions off of that. That’s where I see all of this going.
Can you run it as close to the user on the client side as possible? I have not seen anyone in our swim lane of attention, emotion, and, now, the identity piece I’m working on that has come close to jumping into where we’re at.
Listen to Podcast now on Apple, Spotify, and YouTube...
