Artificial intelligence (AI) and machine learning (ML) are powerful tools that companies can leverage to create better customer experiences. However, they traditionally require millions of dollars’ worth of investment. Alex Muller believes that every product team, regardless of size, should benefit from the same machine learning capabilities used at the most successful companies in the world.
Alex, the founder of SAVVI AI, is our guest this week on Innovation and the Digital Enterprise. Listen to the episode to learn more about democratizing machine learning.
- (01:47) - SAVVI AI
- (03:41) - Machine Learning Barriers
- (09:02) - AI and ML for Everyone
- (11:04) - Target Clients
- (14:02) - Little AI Decision-maker
- (17:40) - Who Knows the Customer Best?
- (19:21) - Family of Entrepreneurs
- (22:49) - Fintech Impact
- (26:44) - Military Mentors
- (31:29) - Chicago Talent
Pat: Welcome to the Innovation and the Digital Enterprise podcast, where we interview successful visionaries and leaders, giving you insight into how they drive and support innovation within their organizations. Today we’re welcoming Alex Muller to the show. Alex is a lifelong entrepreneur and his latest venture is SAVVI AI.
SAVVI is an end-to-end production-ready, artificial intelligence, machine-learning solution. Alex founded SAVVI to help democratize artificial intelligence and machine learning. Prior to SAVVI, Alex was the co-founder and CEO of GPShopper, a retail mobile application developer that helped major retailers navigate their digital transformation by helping them create mobile-first e-commerce solutions.
GPShopper was acquired by Synchrony Financial in 2017 and while at Synchrony, Alex was the SVP entrepreneur in residence and Chief Product Officer of FinTech / AI. Alex was named a “Top 10 CEO Disrupting the Retail Industry Through Technology” by Forbes. He received his bachelor of engineering from Tufts University and his Masters of Engineering and MBA from Carnegie Mellon University in Pittsburgh.
And apparently, being a native son of Argentina has some interest today, Alex?
Alex: Yes, I was going to add that fact. I was also born in Buenos Aires and wanted to say I’m so happy that Argentina won Copa América. Vamos Argentina and Leo Messi and my favorite player, Di Maria!
Shelli: Awesome, congrats. And welcome to the show, Alex.
Alex: Thank you.
Shelli: Can you please share with our listeners a little bit more about SAVVI AI?
Alex: Sure. About three years ago, we were working at Synchrony Financial and putting machine learning and artificial intelligence into different credit products. We realized it was really hard to do, and it shouldn’t be this hard.
Maya and I were at Synchrony for three-plus years - Maya is my wife and the co-founder of GPShopper. We were working together to develop AI-enabled products and we had worked really hard on developing a product that was focused on entertainment and travel.
We were about to launch at a festival called Coachella in April of 2020 when COVID happened. Around that same time last year, our contractual time with Synchrony was ending. We thought that it was time to make a change. We left Synchrony to take some time off, but that didn’t last too long. While hiking through the woods up here, we had time to reflect and we realized there were so many problems with deploying machine learning - absolutely a ton.
We thought, what if there’s a better way to do this? I know many of the people out there know that one of the easiest ways to start a company is to think about a problem that you currently have and ask, “if I solve this problem would other people be interested?”
The problem was that it was really hard to get machine learning into a product that actually makes smarter decisions. We set out to make that much easier to do. That’s the genesis of SAVVI AI.
Machine Learning Barriers (03:41)
Pat: That’s awesome. You mentioned some of those barriers when you were with GPShopper and Synchrony. What are some of those barriers from your experience that you’re really trying to solve?
Alex: Well, getting machine learning into production software is difficult. One of the biggest reasons it’s difficult is that you have to go out and collect data.
At Synchrony, we had access to a petabyte of data and a data lake. You would have to go through the data and say, “does this data help me make decisions?” No matter how much data you have, unless it’s structured in a very particular way, it’s not going to help you make good decisions.
Humans like to see data in what we call “aggregates”, which means summarized either as totals, averages, or metrics that we all can read. Computers don’t want that. Computers want every single data record, but more importantly, they want the causality. They want the before and after effect of everything.
Imagine trying to read a billion different records of a before scenario and an after scenario. That would be nearly impossible for us humans, but it’s exactly what computers want for machine learning. We’ve realized that even if data exists, it’s very rarely in an optimal format for machine learning.
There are also millions of articles right now about how using historical data has all sorts of problems from bias and preconceived biases, so that’s the first stumbling block.
The next stumbling block hasn’t been solved yet and it’s around making predictive models. Now everybody has predictive models, but there’s a problem. If I make a prediction, I have not made a decision. For instance, if I’m trying to decide what the right credit offer is to show Patrick or Shelli, I may want to make a prediction. For Patrick’s credit offer, will he repay it? Will it be profitable? Or will he like it? And I have to make those same predictions for Shelli. Now I have multiple predictions in each case and will need to choose between them.
People get lost and they think that all we need is a prediction. And then they need another piece of software to compile all the predictions and provide a decision. It’s funny - with biological intelligence, our brain is doing that all the time and we don’t realize that we are making multiple predictions.
There’s a great example of Tom Brady in football where he’s asked, “how do you know what receiver to go to?” He goes, “I have no idea.” He does not actively make a decision in those three seconds. His brain is trained at a subconscious level to look at three different receivers and behind the scenes, make predictions. Subconsciously, he’s predicting: Will that receiver catch the ball? Will he catch up to him? Can he make that throw?
These are three different predictions his brain is making to decide which receiver to go to. That’s what SAVVI really does. It’s very difficult to replicate all that, piece by piece by piece, in the machine learning world. Big companies like Google, Tesla, and Amazon can do it. But what about everybody else?
Shelli: That’s a really good analogy.
Pat: There are nuances that he’s trained his brain to figure out as a snap decision.
Alex: Exactly. It’s about multiple predictions happening to drive a decision and being weighed against each other. To be honest, that is how we learn and the way we’ve done it. Machine learning today has not exactly been that way. It’s been about making one prediction with a machine, and then a human makes a decision. That’s good too, but for a lot of things, it can be automated and made faster.
AI and ML for Everyone (09:42)
Pat: You touched on some of the other big companies that are doing machine learning. In previous conversations, you’ve mentioned part of your goal with SAVVI is democratizing AI and ML to make it available for everyone. Why is that important to you?
Alex: I’ve been working on leveraging AI at a variety of companies, both GPShopper and then with a small team at Synchrony. We had tons of resources at Synchrony because we’re a bank and we could put millions of dollars into it, but not billions. Whether you’re a small team at a big company or an independent startup, if the big guys have these big guns and nobody else does, that goes against the ethos that a lot of us have of trying to make it work for everybody.
I don’t want to live in a world where the only people who have access to the power of machine learning and decisioning are companies who can invest hundreds of millions of dollars in that direction. I would like to see us all have a little bit of that power. It’s also about solving a problem I’ve had many times over that we’ve wanted to solve at GPShopper and Synchrony.
We used to have these things called push notifications, which I’m sure either you like or are annoyed by, depending on what application you use all the time. Knowing when to send a push notification to reduce the annoyance factor is a great machine learning problem.
Machines can learn that you might like two push notifications a week, as long as they’re between the hours of four and five and not after seven. However, the solution to that problem with current technology may cost $100,000 to $300,000. And that solution to that problem may only have a value of $20,000 a year.
How do you solve that in a cost-effective way for somebody who’s only going to get $20,000 worth of value from that a year? You need to make that solution cheaper than a thousand dollars a year.
Pat: That’s a great point.
Target Clients (11:04)
Shelli: I love your thoughts around trying to democratize AI. You are trying to get this in the hands of the masses versus the few. That being said, who are your target clients?
Alex: Our target clients are anyone who has a product team. We’re trying to get our solution in the hands of the Product Managers or the technical Business Analysts. We care a little less about the industry. E-commerce, FinTech, and media are all industries that Maya and I have a lot of experience and context in, but we’ve been recently learning about companies in logistics.
This is one of my favorite examples. One of our pre-seed investors lives in Montana, about 45 minutes outside of Big Sky. Someone he knows has a bovine herd software management company, meaning he has software to help you manage heads of cattle.
I didn’t even know that was a thing, but it turns out that the cattle ranches have iPhones and Androids like everybody else. They want to understand things like, “what’s the best time for this cow to see a vet?” Seeing a vet is expensive and they really only want to see the vet if it’s going to be productive for the head of cattle.
They’ve built a stem using our technology and it lets the cattle ranchers themselves embed their knowledge into an AI tool. That tool can recommend what the best time is for this cattle to see a vet and what the best time is for this cattle to do the next step across their lifecycle.
It could really be anybody. It could be people who have a logistics problem that they’re trying to solve. For example, figuring out which of your 10 facilities should take on an order. Or it could be financial institutions that may have whole quadrants of data scientists, but they’re all working on fraud, and nobody has time to work on offer optimization.
When the offer optimization unit goes up to the 30 data scientists at that big financial institution and say, “Hey, can you help me on this email offer optimization? I want to make sure I send the right email to the right person at the right time.” They go, “Absolutely. As soon as we finish this fraud, anti-money laundering, and computational investment work, we’ll get to you.” That’s four years from now. So it could be any of these types of people.
Little AI Decision-Maker (14:02)
Pat: Could you expound on what a stem is and how it relates to your solution?
Alex: Absolutely. Thank you for calling me out on that. As any founder who lives and breathes their product every day, sometimes you get a little too comfortable with your own language.
A stem is a little AI decision-maker. If you use SAVVI, the first thing we ask you to do is build a stem. You build a stem by telling it what you want it to decide.
In a FinTech example, you’re deciding whether to approve, review, or reject a transaction. Much like an employee, you give your stem a goal. In the case of the financial example, you’d want the goal to be a successful transaction. You want to increase that goal while also reducing the goal of a return or what’s called a “non-sufficient funds event”, here somebody says they have the money, but they don’t actually pay it back.
Ultimately, they want to increase success and decrease failure. Easy enough. Given your goal is to accept or reject transactions and you want to increase successful transactions, what do you think would influence that? Maybe the amount of money they’re looking to spend, how much money they currently have in the bank, or how much money they may already owe somebody. Basically, a stem works just like a person would work to evaluate a decision on a set of criteria.
Sometimes I like to say I pride myself on being a data-driven decision-maker, but a stem truly is a data-driven decision-maker. The best part about it is that it puts you, as the person, in control of the system. We’re not replacing a person with a stem.
It’s more about saying the person has to be strategic. Computers are really good at microtactics. The real thing where humans are way better than machines is intuition. You’ve got to separate those two. For instance, one of my favorite data scientists always used to say that there is an art and a science to data science. Data science is the field of science of machine learning and artificial intelligence. The science is those algorithms. Those are the stats that all of us forget, that you’ve learned to objectively analyze and come up with a prediction. The art is what you should include in the math.
That art is often done better by somebody who knows their product. Somebody who might be in marketing. Somebody who might be the business leader. Take an e-commerce product, such as a shopping product. It’s generally not the data scientist who knows the customer’s pain points. It’s often the product designer or the user experience designer. Those people actually have a better sense of what should be considered. That’s the balance. We want humans to be strategic, so the machines can figure out the microtactics.
Who Knows the Customer Best? (17:40)
Pat: As you were talking about the idea of who would know the customer best, I was thinking that it’s probably the people in the call room. They deal with customer pains every day. How do you engage that brain? How do you get those people, their experiences, and their common requests into that hightower of IT or technology?
Alex: It’s funny. Even though I’ve been the CEO and our co-founder, I’ve always been a product manager at heart. I love that the role of product management has become an ascended role. Now there are chief product officers who report to CEOs, and they’re very important roles.
I’ve had over 50 product managers that have worked for me over the years. They all hate me because it’s the one role I know best and I’m always hard on that role. The point here is a good product manager has a strategy for ingesting the ideas and insights that come from the call center, from the clients directly, and from the salespeople.
Oftentimes, when you’re looking at a company pinwheel, the product managers are in the center. I would like the product managers to take the time to talk to the client support teams and make sure that their feedback is in the loop. That’s why I liked that role as a role that builds the stems. They’re the ones who have that overall dominion to talk to the various constituents.
Family of Entrepreneurs (19:21)
Shelli: Your wife is a co-founder. Is she also a product manager? Does she come from that same background?
Alex: No, our joke is that sometimes I build it and she sells it. But sometimes, she sells it and I build it.
Pat: True entrepreneurialism right there.
Shelli: The art and the science.
Alex: Exactly, and it worked out very well for us with GPShopper. We do overlap on the product at times. She is a very competent technologist and could be a great product manager if she wanted to. The reality is that her focus is more on communication and business strategy. I’m working with clients and my focus is more on product development and working with our team to build and solve the next generation’s problems.
Shelli: How do you turn it off? How do you separate work life and personal life when you’re both very passionate about this business?
Alex: I could pretend for five minutes that there’s a magical way to turn it off and there isn’t. The problem with founders is that we’re wired differently.
Maya and I like to be unbounded and also like to be held responsible. It’s a weird duality. I don’t mind the failure, but I want the credit too. Maya is the same. We want to put ourselves out there. We don’t want to fail, we obviously want to succeed, which means we have to try it. And I think when you’re wired that way, it’s hard to turn it off. When you’re forced to turn it off, you end up just thinking of the next thing to do.
To answer your question, it’s a blessing and a curse. Both my parents and Maya’s parents were both immigrants and entrepreneurs so maybe it’s in our DNA.
My brother and sister are also entrepreneurs. Oh, and talking about my brother, he’s an entrepreneur in Chicago, actually. It’s a further reinforcement of our love for that city, especially from a technological perspective.
Pat: It is very common when I see entrepreneurs and founders - they can’t turn it off. My wife has asked me a number of times, "Why don’t you get a hobby?" And I said, "Well, I’ve got one. Mine just makes money. Why is that a problem?"
Alex: At least we hope it does, but sometimes it doesn’t. We have to live through that too.
Pat: Right. It’s about building something. The accountability, the ownership, the pain, the pleasure. It’s all there for you if you want it.
We had a conversation about FinTech, the investments going on in that space, and some of the challenges with AI. I really wanted to switch over and get your thoughts on how you see AI impacting that environment.
Fintech Impact (22:49)
Alex: AI is in fertile ground in the FinTech community. AI and ML are about learning to do things better. When it comes to FinTech, you’re often dealing with a lot of statistical data, bank accounts, large amounts of money, and transactions. You’re also dealing with fraud, which generally leaves a pattern.
One thing machine learning is awesome at doing is pattern recognition. When you think about the different use cases in financial institutions - whether it’s credit approval, transaction approval, fraud detection, anti-money laundering detection, predictions on profitability - you have great use cases.
Now, I don’t think I’m going out on too big of a limb by saying that the current model is one that’s really based on credit history. Credit history serves a decent purpose, but there are a lot of problems with credit history. Sometimes, immigrants have great jobs in the United States, but they have no credit history for example.
They come to America, they’re on an H-1B, they’re making $130,000 a year, and they can’t get a credit card to save their lives. That’s a lot of income for somebody who can’t get a credit card. The reason why is because credit history requires that second word: history.
They’ve just moved from South America or Europe, and they don’t have any history in the U.S. We can use AI as an augment, a supplement, or a replacement for traditional process like credit history. You could actually look at things like bank cash flow and say, “Hey, I think I can make a better prediction over the person’s creditworthiness.”
That’s actually better for both the economy and for the financial institution. And it’s better for that person who recently moved here. There’s a lot of history about credit history that implies socio-economic upbringing. You’re weighted down based on where you come from.
But I think there are different techniques that can replace the history. Most credit providers have this thing called the decision tree, which is basically a whole bunch of if-then statements. For example, if the credit score is less than 600, don’t provide this interest rate. Or, if the user doesn’t have this job or is unemployed, don’t do PDQ. Those if-then statements, sometimes they don’t learn. And sometimes looking at somebody who is unemployed may not reflect the fact that their employment is highly probable and that may not be effective when looking at somebody who’s had very careful cash flow management all their life. That’s what you need to look at because maybe they have cash flow from a different source that’s not their job. Like, they have a grandmother who keeps giving them $2,000 a month and that’s not embedded in any job information, but that is part of their cash flow and then they’d be credit worthy in that context.
Military Mentors (26:44)
Pat: One of the other topics I want to talk about is your mentors. We want to talk about people who have had an influence on you and your career.
Alex: Shelli, you mentioned that you did something with soldiers, right?
Shelli: Both Patrick and I belong to an organization called Project Relo. We take corporate executives out and pair them up with military executives over the course of three days. By the end of the three days, you can imagine the stories that are told and the information that’s shared. That’s actually how Patrick and I initially met.
Alex: Awesome. That actually leads me to one of my first mentors. When I left college with a Bachelor’s in Engineering, I decided to move to Israel and live with my grandmother and some of my cousins out there.
I was looking for a job and I finally got one. I was a glorified repairman, or service engineer. I used to fly around the world - Venezuela, Moscow, all over Europe, even to places like Tomball, Texas. That was probably one of the most exotic places I’ve ever been to.
I fixed digital printers. I once had this issue in Venezuela where the person brought a broken machine in. These machines cost half a million bucks. So anything they could get in terms of spare parts or other things was really useful.
I flew back to Israel. I did the successful print. Everything was successful. A week later, the guy decided that he wanted free extra equipment, free parts, spare parts, and warranty. He said I was drunk on the job and he said that I hadn’t done it correctly. That wasn’t true and this gets to my mentor, Leor, the big burly guy who was a captain in the Israeli army. He was my boss. Instead of being like, “the client’s always right”, he said “no, my guy did not do that. I’m sorry. No shot.”
And Leor was like, I do not care. He’s my guy, he’s on my team. Granted, when I was in Israel working for him, he wrecked me. He made me work 14 hours a day. In order to learn how to fix these printers, we had to build them on the factory floor and I used to smoke cigarettes.
And the ink is this material called methyl ethyl ketone, highly explosive. He looks at me smoking a cigarette and says, "Alex, come here." I stand right next to him and he puts a cigarette out in the methyl ethyl ketone. I did not know it wouldn’t explode. I jumped like the scared American I was, and they all laughed at me, being the American and the guy who wasn’t in the military.
When push came to shove and it was them or us, he was behind me 100%. I take that with me everywhere I go. As hard as he was on me and as hard as he was on making sure I did things right, he had my back in a way that I would always remember with an “on the ground we’re one unit against the others” mentality.
And to me, that was a huge mentor. Shout out to any of the soldiers and field leaders of the military because they learned something that I think the rest of us have to learn in a different way. I would say they learned it the harder way, but they learned it the faster way.
Chicago Talent (31:29)
Shelli: Agreed, agreed. I think we talked earlier, Alex, about why Chicago is one of the most underrated sources of talent. You mentioned that your brother is an entrepreneur here. I’m curious, what do you think about that topic?
Alex: Here’s the story. So we started GPShopper in my living room in New York City, mostly because garages in New York city are totally expensive. We started this company around 2010 and it was in our living room.
Then we met this great product manager who was living in Chicago. We had just opened the New York office at the time. Him and his wife came to New York City to look at apartments. They’re like, “no shot.” They were not going to live in New York City. He said, “give me three months and I bet you I can hire incredible people in Chicago for less than what you would have to hire them in New York City.”
I’m like, “sure, why not?” So he opened an office in Chicago and from then on, the Chicago office doubled in size almost every year while the New York office would increase by about 20%. We just kept getting great talent that was both hardworking and respectful.
Nothing against the east or west coast. I think Facebook’s very public about the fact that the average engineer stays at the company for about 18 months. I think that’s very different in Chicago. I think lasting bonds form and people are excited to work on things. If you’re a respectful employer and you provide a good work environment, people will stick around. Even at SAVVI, there are people we’ve worked with now for 5-10 years together. They were working at GPShopper too.
I’m a New Yorker. I love New York, but I will say that Chicago is a fantastic market for talent. I went to Carnegie Mellon, but the other school that I got into was Urbana-Champaign, Illinois. I ended up going to CMU, but I’ve probably hired two dozen people out of the University of Illinois. I love that school and they provide great tech talent, as well as the schools in Chicago. We love this city.
Shelli: That’s awesome. By the way, the founder of the company that I work for is Larry Gies. Gies College of Business at the University of Illinois is named after him so we’re obviously huge advocates for them.
Alex: To that point, I’m back here in August. So I’m spending the week out there. Chicago in the summer is always great.
Pat: I can speak for all Chicagoans that we really appreciate when New Yorkers tell us that we’re awesome. It really makes us feel great about ourselves and it’s really what we need.
Alex: Yeah. I don’t know if there’s some sarcasm in there, but…
Pat: It’s a little bit, not too much.
Alex: We were living in Chicago and now we live in California. Everybody who is east of Berkley is somehow 30 years behind in technology from here. So I’m getting a dose of my own medicine a little bit.
Pat: Well, that’s good. Hey Alex, thanks so much for being on the podcast. It’s really great stuff. It’s such a joy hearing about your success, the purpose of SAVVI, and all the things that you’re doing. You have such a bright future and we’re really hoping for great things for you ahead. Thanks for being on the show today.
Alex: Absolutely. Thank you for having me.
Shelli: Thanks, Alex.
Pat: We also want to thank our listeners. We really appreciate everyone taking the time to join us.
Shelli: If you’d like to receive new episodes as they’re published, you can subscribe by visiting our website at dragonspears.com/podcast. Or find us on iTunes, Spotify, or wherever you get your podcasts.
About Our Guest
Prior to founding SAVVI AI, Muller served as Senior Vice President, AI-Enabled Products & EIR Synchrony Ventures at Synchrony Financial. In this role, he led innovation and product development in several domains, including mobile and artificial intelligence.
Previously he was the CEO and co-founder of GPShopper, which had been acquired by Synchrony Financial in 2017. Forbes named Muller one of ten CEOs whose companies are disrupting the retail industry through technology. He is revered as an industry expert on the topics of mobile payments and security.
Inc. named Alex a person “at the top of his game” and a leader in his industry, and CIO Magazine cited him as a go-to expert in the retail and eCommerce space. Alex has served as an Adjunct Professor at NYU, lecturing on mobile application strategy and execution. Alex graduated from Tufts University with a BS in Engineering before earning his MS Engineering and his MBA (Tepper School), from Carnegie Mellon University.
Subscribe to Your Favorite Podcast
If you’d like to receive new episodes as they’re published, please subscribe to Innovation and the Digital Enterprise in Apple Podcasts, Google Podcasts, Spotify, or wherever you get your podcasts. If you enjoyed this episode, please consider leaving a review in Apple Podcasts. It really helps others find the show.
This podcast episode was produced by Dante32.