A conversation with Andy Rossmeissl, CEO of Farraday
In the age of “cookie-less” data, how does machine learning adjust to predict customer buying habits? Andy Rossmeissl, the CEO of Farraday, a prediction platform in the consumer brand space, says it starts with identifying customers once they identified themselves. Sourcing that feature data can be the key to optimizing machine learning to create a better customer experience.
Full transcript of the conversation below:
As we know, we’re in the cookie-less data realm. So how do you leverage that barrier that was such a key tool for a lot of marketers in the past and using AI to create a better personalization when it comes to brands and their marketing?
Yeah. It’s a great question. I mean, the demise of cookies has been often foretold and never seems to fully happen but it seems to be like, well, we’re on that track now. I actually, believe it’s the right thing to happen for a lot of reasons, moral, ethical but also practical and it’s something that we actually, really founded Faraday around 10 years ago, where we made a decision within the first weeks of starting our company, that we would not be using cookies of any kind to track the customers of the brands we work with. It’s an odd thing, I think, to identify somebody in a way where they never know what their identifier is. Rafer, you probably, can’t repeat your tracking cookies off the top of your head and I certainly can’t.
At Faraday, we chose 10 years ago to only identify people by the identifiers they know they have, the address on the outside of your house that you’re very comfortable, everybody’s seeing when they pass by your mailbox or apartment building, the email address that you freely share with companies of all kinds, a phone number that’s in the phone book. These are all identifiers that humans have been very accustomed to having and sharing. And human beings don’t move as quickly as technology does and we really feel like there needs to be more of an anthropological underpinning to a lot of the ways that brands engage the market digitally.
And so how do you integrate AI into something like that? And I know I’m getting a little bit under the hood and I’m sure it’s not a simple answer but that seems… My guess is that cookies were probably, a methodology that saved a lot of time, it was easy to automate. Are you at liberty to share a little bit into your processes?
Yeah. We’re wide open here at Farraday, it’s something we love to talk about. And I think I can talk in general terms so that maybe other folks listening could start to think about some of the same principles. But really, what you have to do is start to make a commitment really throughout your infrastructure that you will not identify people until they’ve identified themselves to you. And when people identify themselves to you, this is actually very common. Anytime if somebody orders something or signs up for a newsletter, they’re identifying themselves to you.
So, it could be a name, address, email, phone number, these identifiers that we feel comfortable sharing and we know are in fact, the identifiers that people use when they’re voluntarily identifying themselves, it’s no coincidence. And when it comes to machine learning, the way to think about data is there’s really two kinds of data that you need. So, you need some label data, we call it in the field, which is indications of where a certain outcome has happened in the past. So somebody bought something, somebody signed up for an email address, somebody churned, somebody bought, again, these are all pieces of data that can be collected that are indications of success.
Meanwhile, you need a totally separate kind of data which people in the practice call features. So, these are characteristics of me, you, everybody that don’t necessarily have anything directly to do with the labels, with the actions that we are taking but within which a machine learning algorithm may be able to discover patterns that are in fact correlated with those actions. So, when it comes to label data, it’s a fairly straightforward process, brands are looking for things like purchases. And when a purchase is made, you clearly have an identity which is provided to you whether it’s a digital purchase with an email or let’s say it’s a physical purchase and you have a mailing address. So, that is usually straightforward for brands to wrap their head around, but the feature data can be much harder.
I think, a very common approach for brands these days is to try to capture things like clicks and website activity, search history, things that maybe users aren’t fully aware that they’re sharing and that are attached to their identity. We choose typically not to start there although brands can certainly choose to layer that on. But at Farraday we decided a decade ago to take a different approach which was to work with long established, certified data vendors whose job really emerged back in the cataloging era when that was the way that many things got sold and as part of those engagements with direct people, data was captured about purchasing behavior. These companies like Epsilon, is a key partner of ours, has done an incredible job of securing consumer permission, making sure that these are opt-in relationships, that the data is mostly focused around people’s purchase and transactional history. The commercial behavior that over thousands of years, humans have become very accustomed to being a part of these open markets. So, if you can source your feature data in that kind of a way whether it’s working directly with Farraday or choosing to pay large amounts of money to license the data yourself, I think you can do a really nice, strong job of completing that machine learning puzzle without having to resort to tracking cookies.
Fascinating stuff. So, I know you can’t give specific examples but you work with a lot of larger companies, mid-size companies, Liteboxer, Etsy, financial services firms, Delta Dental, Procter & Gamble. Are you able to give any case studies just in general terms, in terms of how this approach has been able to be successful?
We work primarily with three different kinds of predictions here at Farraday. It’s an infrastructure platform for consumer behavior predictions. So, imagine you’re a brand, you have a bunch of data and you want to predict something about your customers, that could be their likelihood to do something, buy, convert, buy again, any kind of a binary yes or no… Rafer is likely to do this or not. The second thing we do is persona predictions. So, we will take your customer base, we’ll use machine learning to divide it into a small number of very thematic personas that you can use to adjust your outreach to really speak to people and increase conversion rates. And then finally we forecast LTV, lifetime value, things like the lifetime length, those kinds of things. So when you’re a brand, we think of those three things as ingredients. And it’s really up to the brand to determine what recipe they’re trying to make. Some brands are really focused on improving conversion rates off of leads.
Some brands are really focused on working their way through unrealized LTV with customers that have a lot to offer but maybe aren’t spending at the rate, the brand expects, maybe it’s finding totally new customers out there that otherwise would not have heard of the brand. And what the brand does is essentially, push data about their customers to Farfaday. We enhance it with all of these consumer profiles we have and then the system automatically builds the correct machine learning models to predict those things that the brand wants. So it could be anything from, you’ve got Salesforce where you’re capturing a lot of leads, you’ve got a limited number of reps whose job it is to go out and try to convert those leads, maybe this is a bigger ticket purchase, like a rooftop solar panel. And in that case, you really need to know which of these leads are most likely to convert. That’s one of these things that is a very straightforward, powerful, practical application of AI that for big companies has been the normal course for many years. But companies in the middle and smaller companies have I think, really struggled to find the right people, spend the fairly enormous amounts of money to collect the right amount of data, put together the right infrastructure and Farraday’s here to change all that.
The post Optimizing Machine Learning to Create a Better Customer Experience appeared first on TheCustomer.