An Image is Really Hundreds of Data Points That Tell Us Who We Are
This is transcript of the keynote by Anastasia Leng, CEO of Picasso Labs, from our 2016 LDV Vision Summit.
Thank you. Hi, everyone. My name is Anastasia Leng and I am a former Googler using technology to measure creativity. Now, before you decide if that statement alone makes you love me or hate me, let me tell you a little about how we're putting science into something that has traditionally been an art.
As many of you know, human brains process visual information sixty thousand times faster than we process text. But the result of that speed is that we walk away with a very subjective understanding of an image. I like this image, or I don't, this is a good image or a bad image. We've all probably sat in a meeting discussing a photo with someone who is more senior than us who says, I don't know why, but I just don't like this image, and there's nothing we can do about it.
Technology is very different; technology looks at an image objectively. It gives you the most comprehensive set of metadata contained within an image. With that data, thanks to image recognition technology, we've done things like build visual search, we've done things like build content recommendation systems.
What if we could use this data to understand how your user's reactions change based on the content of your image? What if we could use this data to better understand how their perception and their interaction with your brand changes, and the why behind image performance?
What if we could use this data to measure and optimize your visual voice at scale. Now, this in itself is not really a new concept. Psychologists, have been toying around with this for years. They've been looking at how visual stimuli change users behavior, perceptions, and ultimately their reactions to a brand or a different environment.
For example, in the late 1990's, the government of Scotland decided that they wanted to reduce crime rate, especially at night. At the same time, the government of Japan wanted to reduce suicide rates in train stations where people were jumping in front of the tracks. They installed blue lighting, which is meant to have a calming effect. They saw a 9% reduction in crime.
Pharmaceutical companies, big pharma, has been accused of using color psychology to influence their trials thereby heightening placebo effect. Consumers, or their trial participants, rate red pills as having more of a stimulant effect and blue pills as having more of a depressant effect. Now this varies by gender and this varies by country, it is very culture specific, but there is some reaction that's real, we react to visual stimuli very differently. As a fun fact, one of the common complaints about Viagra in the U.S. is that the pill is blue and it doesn't match up with the reaction that consumers expect it to have.
Now brands have known this and they've used this anecdotally one-off. If you open your phone now and look at any of the food apps on your phone chances are they will be red, or orange; that is not an accident; porn's the same way, it's not an accident - it is because they want you to be impulsive. If you look on the left side you'll see a bunch of brands that are blue. Those brands want you to give them your money or your data and what they're saying is, we are safe, we are trustworthy. If you're going for global domination, this rainbow effect here in the bottom seems to be the lucky ticker.
But it really is about so much more than just color and to illustrate this I want to tell you guys about a personal anecdote. So, I'm a freak about EV testing, which has bled a bit into my personal life, and when I was fundraising money last August and pitching my first clients, I started testing it out, experimenting with how the way I looked impacted my conversion rate at a meeting.
I started to look at whether investors' or clients' reactions changed based on whether I wore my glasses or not. This is nowhere near statistically significant, right? But the reality is, I now wear glasses to every investor meeting because I saw that my conversion rate was higher. While this is nowhere near statistically significant, what it does tell you and what we all fundamentally believe is the way we look, the things we wear, impacts the way people react to us.
If I was trying to optimize for conversion rate and dating, it'd be the other way around. The question is, and actually this is probably very context dependent, but if this is the reality, why wouldn't this be the same thing for brands?
Brands spend billions of dollars creating millions of images. Those images contain trillions of data points about consumer's actual revealed preferences about the visual content within those images. But brands have no idea how to harness this data.
Brands spend billions of dollars creating millions of images. Those images contain trillions of data points about consumer's actual revealed preferences about the visual content within those images. But brands have no idea how to harness this data. And this is what Picasso Labs does, we give you very specific performance insights to help you understand the “why” behind image performance and help you better understand who your audience is and how different parts of your audience respond to different visual content.
Now our technology is never going to tell you something like always use red on Instagram or you know, blondes are always better in your display campaigns; what we believe is audience react differently to different visuals based on who is the brand behind them so we do very personalized image recognition and machine learning to understand what is about a specific audience reaction to your brand that causes an impact in behavior.
As a result of the way we work, a lot of the insights we gather we can't really talk about because they are seen as competitive advantage and very proprietary to the brand but we have been working with a number of luxury fashion companies who've let us expose a little bit of the data. So a few months ago, we were working around fashion week with companies who wanted to understand what type of image style worked best. This is luxury fashion companies on Instagram. What they were measuring was increase in engagement. Now engagement on Instagram is likes, comments, etc. I'm gonna let you guys guess and I've started you off with an easy one, which image style, for luxury fashion brands right, I'm talking Chanel, Louis Vuitton, Prada, etc., gets the highest engagement on Instagram? Raise your hand if you think it's runway. Okay, a couple of hands. Raise your hand if you think it's editorial, right? What about street style? Yes, okay, told you guys this was an easy one, Street Style is absolutely right.
Now the fascinating thing here is we all knew that right, not all, most of us, most of us knew the answer here, maybe we were lucky, maybe it was an intuition. Around fashion week, most luxury fashion brands that were monitored, their engagement rates dropped. Part of that is the saturation problem, part of that we saw that even outside fashion week cycles, any runway content just seems to really drag your performance down and we've analyzed this by looking at millions of images across a bunch of fashion brands.
Now the second one, what type of model shot works best? So raise your hand if partial body face visible you think is going to be the winner here? Okay, what about partial body no face? Okay. Full body? Okay, so I think full body's got it but no one seems quite sure, this one's a bit harder and the results were really surprising. No face wins, in fact if there is anything, any takeaway that we've seen across most luxury fashion brands is cut off the face right? Which is crazy if you think about the amount of money people spend hiring models strictly on the attractiveness of their face and actually you consumers don't want to see it.
So, that's Picasso Labs, our mission is to foster creativity through data, we really believe in giving you the kind of data that helps you make smarter creative decisions so that next time you're in a meeting with someone who says "I just don't like it" you could say "Well I don't care because I have data to show that our users do".
Thanks very much.