Podcast Episode 6: Kount VP discusses Machine Learning, AI, and how the technologies help the mobile payments industry
Today's podcast is sponsored by...
Kount is an ecommerce fraud prevention solution. The enterprise solution uses advanced machine learning technologies combined with customized policy rules and dozens of other features to prevent fraud for online businesses.Kount customers experience as much as 98% fewer chargebacks, lower fraud losses than ever before, and minimizes the need for manual review of orders. Get back to business and let Kount fight fraud for you. Learn more at www.kount.com.
Listen to the Podcast:
BuzzSprout - Click to Subscribe
YouTube - Click to Subscribe
Stitcher - Click to Subscribe
TuneIn - Click to Subscribe
Spotify (Coming Soon)
Intro & News
Welcome to episode 6 of the Mobile Payments Today podcast. Shake shack is moving forward with the testing of self service kiosks, but they're also going to be doing so in conjunction with old fashioned cashiers. The company CEO told analysts during a recent earnings call that five locations will test this hybrid system.
He also said that the company still has a lot to learn about cashless kiosks, but added that the hybrid format might be deployed in such areas as Seattle and the bay area where labor costs are high. So why am I bringing this up again? Well, if you remember my chat with ATM Marketplace editor, Suzanne Cluckey from episode two. We discussed this a bit and kind of where the cashless trend is going and Shake Shack's hybrid model was probably the result of the pushback they got from consumers earlier.
At one location in Manhattan cashless kiosks were the only way to interact with the brand. So what happened that led to some bad reviews on Yelp. Shake Shack decided to bring back cashiers to this particular location to go along with the kiosks. And I think what this approach shows that this hybrid model was that brands really need to pay attention to what their customers are telling them.
People want a choice, they do want choice, and now Shake Shack is making sure to give that to their customers and while a cashless society might be on the way, we're not there yet, so why not give people the choice to pay with cash?
Will: It seems as though machine learning, it's something that's getting thrown a lot around these days in regards to different industries and for those who are unaware, what's the connection right now between machine learning and the payments industry?
Don: Machine learning has gotten a lot of press over the last couple, three years. It has started to do some things, especially in the payment industry that took humans and other calculations a long time. The beauty of machine learning; it's taking models, algorithms and feeding immense amounts of data, looking for anomalies, looking for trends, groupings or sometimes people call them clusters of information.
Think about the data that he is reviewed for every transaction you go to buy a pair of shoes. They get your name, your address, your phone number, your email address, your payment information. They collect information about your device. There's literally hundreds of pieces of information about every transaction.
Now multiply that by millions and you can see why if I've got a well trained machine, it can do things in the blink of an eye that would take a team of people hours or days to do. The intricacies and complexities of the payment industry is a perfect area for machine learning to really speed up, provide better accuracy of the information that we get to review. I think that's why payment industry started with machine learning. It's really started to blossom just the last three, four years. And it's started to change things.
Will: You see machine learning and AI bundled together. We're started to hear more with these two, these buzzworth terms...
Don: Yes, and they are buzzworthy. Machine learning is actually a branch of artificial intelligence. And so when you say AI and machine learning, it's a bit redundant. We use them somewhat interchangeably, there's a couple of brands that they're supervised and unsupervised machine learning. Unsupervised is when you kind of set the path for what the machines do and provide you certain amount, certain types of data. Supervised machine learning is when you build models and you run the data through those models and it gives you a calculation or an algorithm based on what you're looking for. So it can get quite complicated.
Will: Sure. I can see how all this plays into fraud detection. Mckinsey & Co wrote a paper a couple of years ago about how Braintree was using its authentication tools in conjunction with Kount's fraud detection capabilities to basically authorize high volumes of transactions and verifications in realtime. Mckinsey cited that partnership as a great example of machine learning to combat fraud. So what can you tell me about that partnership?
Don: Sure. Braintree has been a great partner for us for many, many years. They were one of the leaders in saying "hey, we need to take the complexity out of payments". How do we make payment easy for merchants instead of this complex, fill out an application, wait 10 days and all this other stuff? They made payments easy and it really helped ecommerce and merchants move into ecommerce much more quickly and efficiently. And what they did with payments we did with fraud. Something that was easy to understand and easy to use. And so the merger of the two was really kind of a nice little match made in heaven. Now, when you talk about authorizations and fraud any ecommerce merchant, a company, what you want to do is authorize the maximum number of transactions that are good.
You and I go to make a purchase. I want to be authorized 95% plus of the tim. If I can provide information to a company like Braintree that says, hey, this, this information from from Will is valid, so authorize it. Then one, they get more authorizations because they have better information about the transaction itself too. They avoid the pitfalls of authorizing, have a fraudulent transaction so they don't get loss of product, loss of revenue, loss of chargeback, fines and fees. So if we can give a proper fraud screening prior to authorization, we can both avoid the cost of fraud and increase the approval of good transactions. One of my data scientists explains it to me this way. He said, Don, in 10,000 transactions, maybe 600 of them would be considered suspicious, possibly fraud.
He said, but what a lot of folks do is they lock up 10,000 folks to catch the 800. Now you've caught the 800, but you've locked up 9,200 innocent people and when you think about that for authorizations, yeah, I could just decline the transactions that looks suspicious and I saved myself from all those bad fraud things. But I might have just sent away a good customer and so having both of those things together allows me to maximize my authentication, which means maximizing my orders that I accept and minimizing the fraud that I I have to reject.
Will: What are some of the other key things for consumers that are made that might be unseen benefits with with machine learning?
Don: One of the unseen benefits; let's say there's a data breach and financial information is compromised, goes out on the market, it gets sold to a dozen criminals and now they try and monetize that by going to websites and use it. Websites using Kount would find that information and go, nope, this is a fraudster trying to use information that they purchased online and so we protect that consumer in that. In that respect, they don't lose that money or they don't go through the hassle of having to call their bank and say, that wasn't me and throwing out an affidavit and all that kind of stuff. So most consumers don't see that, but it's an unseen benefit. The other side of that is the amount of data. When you think about the payment, think about the data that was looked at and what a merchant could do with that data to better their customers.
So maybe they find out that, gosh, every time somebody buys a shirt, they buy a tie with it, so why don't I just put a promotion together for when you buy a shirt, you get a free tie. Or I know that Will is a Dunkin's coffee drinker and he comes in five days a week to pick up his coffee that is preordered using a Dunkin' app. I'm going to say, hey, Will, if you do that 10 times this month, guess what? The weekends on us. So that type of information, while it's not considered fraud mitigation, using the information that has reviewed during a fraud review can be used to help you market to your customers, help you service them better, help you provide better products, help you give a opportunity for promotions and incentives, and treat your best customers.
It can even feed your loyalty program. You could even use some of that information to reach out to your folks that have signed up for your loyalty program and say, guess what? You're a loyal customer. We know that you love coming in and getting your Dunkin' coffee. We're going to give you this for free just because we like you, so I think there's lots of information and it all comes down to that data and it all comes down to that machine learning saying, these are good folks, these are bad folks using that, you know, the billions of data points that we get to review.
Will: You mentioned earlier about the connection between machine learning and what it can do in the event of a data breach. Is that something you anticipate people looking into more investing more as, as we go forward?
Don: It's a good question and it's already happening. I'll give you an example. Name any of the large data breaches over the past couple of years. You know, millions of records, hundreds of millions of records are exposed. They go out on the marketplace. At Kount, we couldn't tell you where the breach came from before somebody announces it, but we know that there's been a breach. And the reason we know is we see the activity and the increase in data that's flowing across the channels of ecommerce.
I use this analogy, you walk into your kitchen, you get home, you go into your kitchen, there's two inches of water on the floor. You know there's a leak, you just don't know where it is, you still have to deal with the leak. You've also got two inches of water to deal with. Using the techniques that we've used when there's a data breach, we may know long before our merchants know, we may know even sometimes before banks.
Now because of the way our machine learning looks at that data compiled that runs it through models and goes, you know that's wrong. That's bad, that's not Will. The other thing is with the sophistication of fraudsters today, there is something that's been building over the past year or two and it's called synthetic. This is where whether through a data breach or other other ways of stealing information, they may be able to get Will's social security number, credit card number, Jim's address and email and put all this information together. And so individually, all the pieces look right, but it's sort of a Frankenstein of stolen data. That's been assembled, the person that they might submit to make this order doesn't exist, but all the parts and pieces are accurate and unless you've got something like machine learning in place that can put those parts and pieces in the right places, it's really hard to detect.
Will: One more question, what does the future hold in this area in terms of the marriage between machine learning and payments? As the technology improves?
Don: I'm very bullish on it. I think we're just touching the surface here. While there are some really sophisticated machine learning algorithms and models and processes out there, it's only going to get better.
What happens? The more data that you can put into the system and the more relevant that you can add to that, the more intelligent artificial intelligence becomes. So it will only get better. We'll get better. Models will get better modeling as the data increases in what we're allowed to see and look at and compare as industry start to open up.
When you think about government, healthcare and ecommerce, that data is just going to continue to roll up and it's only going to be through artificial intelligent technology like machine learning that we're going to be able to look at that data, make a determination of whether we think that's valid or invalid and then act on it.
I'm pretty bullish. I think it's only going to get better. We've seen marked improvements here at Kount even in the last couple of years as we continue to work on our models and they get better and better and better. So I'm pretty optimistic about what this technology is going to bring to the future.
Will: Well, I definitely learned a lot here today. We will talk to you in a couple of weeks for our last show here. I appreciate the time.
Don: Thanks Will, I appreciate it.
Will: Thank you for coming on. First, I want to discuss this link we're seeing between the fast casual space, QSR and mobile payments. What do you make of this link?
Cherryh: It's turned into kind of like table stakes almost. You know, mobile payments was pretty new a few years ago and only a few restaurants were experimenting with it. But now it seems like almost every restaurant is, if they don't already have it, they're looking at it because they're just trying to meet their customers' expectations and their customers want to use their phones. They want to go quick and they want to have that technology. So anyone not looking into it, is left behind.
Will: A couple of weeks ago we saw that GrubHub made a big acquisition, buying LevelUp for $390 million dollars. So what does this mean for GrubHub going forward in terms of this acquisition for them?
Cherryh: I think it was a really smart move and I think that, you know, LevelUp has a lot of the analytics GrubHub can use. So they're going to be able to kind of integrate and offer their customers a lot more than just delivery now. GrubHub is competing with UberEats obviously, and so anything they can do to get an edge is going to be great for them.
Will Hernandez has 14 years of experience ranging from newspapers to wire services and trade publications. Before becoming Editor of MobilePaymentsToday.com, he spent two years as the content manager for PaymentsJournal.com, a leading payments industry news aggregator and information hub published by Mercator Advisory Group. Will spent four years covering the payments industry as an associate editor for multiple publications in SourceMedia's Payments Group based in Chicago.