How One Guiding Principle Got Us To £1M In Revenue

 
emocash.png
 

By Matt Celuszak

The Guiding Principle

When I first set up CrowdEmotion, I had no idea whether emotion data collected through a machine would: 1) be accurate; 2) work in the wild and; 3) be valuable to anyone. In theory it was possible, but the hypothesis was still being tested at an academic level, much less in the real world. There was also a lot of bogus technology out there and I did not want to waste either our time or investors money. So, I needed to be super pragmatic.

I needed to demonstrate true human value.

This was 4 years ago when Twitter, Snapchat and many consumer sensations were really struggling to convert usage into profits without large amounts of investor funding.  I did not want to spend 10 years of my life building something people didn't truly value. My co-founder agreed.

So we took a pretty extreme approach:

We will not take institutional investment until we are categorically convinced we can make Emotion AI a profitable business.

This forced ongoing market validation as we built the technology.

You may think it was nuts starting an Emotion AI company with no institutional funding. In retrospect, it was. I have a special group of family and friends who helped me get this started and reminded me why we were there during those tough times.  Because of this, our people are diverse, and our products are unique. More importantly, our approach solves a harder problem in Emotion AI.

The harder problem

You see, there are many companies in Emotion AI that have developed technology which can recognise gestures, eye-movement, facial expressions, verbal tones, and language cues. In fact, over $400 million has been invested in over 38 companies in this space, yet we all face the same bigger problem - recognition only generates data, how do we interpret it?

The emotion recognition market is estimated to be $42 billion by 2022. An estimate which keeps going up every year Markets and Markets produces the report. So you can generate data on rapid eye-movement in a passenger of a vehicle, then what?  So you can track the number of smiles of somebody watching a video, so what?

That's where our guiding principle forced the hard question:

I have emotional data, now what?

Or reframed - we can generate emotional data, how do we make that data usable?

Ignorance helps

The founding team in CrowdEmotion comes from all sorts of backgrounds which has created an inquisitive environment in which we question everything. With a small runway, we took Emotion AI from academia into the wild and to see if we could drive value.  So, we partnered with the University of Nottingham's Dr. Michel Valstar and his team, a very talented innovation group in human computer vision.  This involved a crash course into the world and history of Emotion Artificial Intelligence.  As newbies, we questioned everything.

The lab based tech was impressive. But we needed to get this onto everyday devices which created problems in the wild like moving light scenarios and shifting heads, cameras, and processing power to name a few.  These create huge variability in accuracy because nothing is staying still. The University of Nottingham team jumped straight in and have since created arguably the world's leading face detection system - a strong base layer for facial expression analysis.

By questioning everything, we also learned that psychological models need to be retested. You see, most models are based on human measured data. And humans are notoriously poor at understanding their emotions.  This creates a lot of human bias in our current psychological models.  Stereotyping with bias?  Possibly the root of racism.  Another discussion perhaps.

Where Psychology meets Technology

With Emotion AI, we can quantify at scale. I mean by a factor of thousands. The dataset is so rich it feels magical.  It's like having superhuman listening.  Voltaire was right and our Empath Jing likes to reiterate - With great power, comes great responsibility (though she references Spiderman).  It makes us acutely aware of TRUST and the balance between privacy and value.  It also brings a whole new element to how we model people psychologically.  Or in other words, how we interpret emotions

So we started to revisit behavioural psychology models. Again, by following our guiding principle to achieve true human value, it forced us to seek out and frame use cases where emotions and empathy bring value in a measurable way.  Naturally, we not only have psychologists and anthropologists in house, but our network of psychologists help us reframe and redefine old psych models with an exponentially rich data approach.

Getting to Breakeven

In the process of testing the psych models, we learned that people value their privacy when others are listening intently.  So, we needed a place to generate data around the globe, test the psych models, produce enough value exchange for emotional listening while make money.

That's how we started with media testing at BBC Worldwide Media Labs.  Media testing is relatively harmless for users unlike other fields like mental health. If you fail in media, nobody dies. 

The first project was a disaster. The algorithms didn't work. As we had the face videos already, we quickly rebuilt with the University of Nottingham and retrained the model to produce much more accurate results.

BBC used this method again to help differentiate between claimed like/dislike of a show trailer and the emotional response.  We found that the guilty pleasure category really revealed the black swans in media.  BBC, Lightspeed GMI and CrowdEmotion went on to win an MRS Grand Prixe for this work shifting media measurement into a complementary model with claimed feedback.

While those projects continued, BBC Global News faced a problem all big publishers faced - quality of content v quantity of content. They didn't want to become a quantity advertising player, they wanted to embody quality engagement in everything, but had no way of measuring quality. 

That's where our ENGAGE product came from.  Our psych team worked closely with their progressive ad sales team to test, control and retest.

Last year, we helped clients across publishing, market research and media agencies all trying to do the same thing - put true value on audience time and engagement.  An exchange that every storyteller seeks with those most important to them.

With £330,000 invested, family and friends' love, and one guiding principle, I am pleased to say we are now at break even, having generated £1 million and facing hyper growth as emotions come online.

Final Lesson

I love working at the intersection of people and technology.  One thing that working with Emotion Artificial Intelligence has taught me is that the fear around AI taking over is simply a good driver to helping us to unlock human potential. People are really good at gut feel and intuition. They are also great at abstraction and creativity. On every project we do, I see everyday people transform into really empathetic, intuitive champions because of Emotion AI. 

A conversation for another time perhaps :)


To have a chat with Matt or see how Emotion AI might help transform you or your business, give us a call!

Image credits: Dev Patel via The Noun Project & monkik via Flaticon