AiThority Interview With Stefan Nandzik, VP, Corporate Communications, Signifyd
How do you interact with AI and other intelligent technologies, in your daily life?
When AI is working right, you don’t typically notice that you are using AI. It’s a subtle presence in your everyday life. Think about watching Netflix. You go to your account and there are videos and programming recommended for you to watch. That’s AI.
Of course, I work in marketing with a team, so ad targeting is something that we frequently work with. The proper targeting is heavily impacted by predictive engines, and for that matter, machine learning and AI technology.
I’m cheering for the AI to work. I want it to work because it could eliminate so many time-consuming tasks from day-to-day life. Say I’m heading to a meeting in my car and the meeting organizer changes the location of the meeting. It’s not going to be in the office. It’s going to be at a Starbucks.
I’d want my robot assistant to read me the message about the change. Then I could tell the robot to rearrange my route so that I’d end up in the right place at the right time.
Or better yet, my smart machine could receive and process that information about the meeting and recalibrate my GPS on its own, giving me a heads up that my plans have changed. And while it’s at it, it could order my latte decaf if it’s after four in the afternoon.
As a user, I don’t have to look at screens or take my hands off the wheel to push buttons or swipe right or left while I’m driving. In fact, my daily commute is a great AI use case.
I spend at least two hours a day in the car, just getting to and from work. I’m a big fan of Spotify and of listening to music on my commute.
Spotify delivers playlists to me based on what I’ve listened to in the past. It’s learned what I like and what I don’t to the point that it builds really good playlists of artists and songs that I’ve never heard of, but they are really, really good.
And while that makes my commute much more bearable, I don’t mind telling you, what I really can’t wait for is the day that the car can drive me to work without my help.
How did you start in this space? What made you join Signifyd?
I joined Signifyd at a very early stage. I was employee No. 17. I was drawn to the opportunity to help a startup go to market with a deeply technical product. I loved the idea of building everything from scratch. I wanted to create a modern approach to marketing, to use technology to construct a marketing engine for the 21st century.
Modern marketing, to me, is driven by the buyer. You have to find out how to reach them, figure out how you can get in front of them and find those stories that resonate with the potential customer that actually has a pain point you can help solve.
How Signifyd is trying to solve the problems of fraud, for retailers?
When you look at gross merchandise value or really the value of what online retailers sell fraud equals about one percent of that. But that’s really only a small part of the fraud problem and only a small part of what Signifyd solves for retailers.
The bigger part of the problem is fear of fraud and how that fear causes retailers to become overly conservative, to mistakenly decline good orders because they are afraid that those orders are fraudulent. Those false declines, as we call them, equal another two to three percent.
Retailers miss out on a huge amount of their revenue potential because of that. And it doesn’t end with the missed opportunity to sell to a customer. It also does tremendous damage to the customer experience.
How does Signifyd’s platform work and what role does AI play, in powering the platform?
We screen all the transactions of merchants, leveraging the scale of the Signifyd Commerce Network, which consists of 10,000 merchants selling in more than 100 countries. So, we recognize good customers across merchants, across geographies.
Our AI makes decisions on thousands of signals from a flurry of data points, including transaction and third-party sources. The AI makes a ship-or-don’t-ship decision and we package that with a financial guarantee, to put our money where our mouth is.
We make merchants whole on any order we approve that turns out to be fraudulent. That allows merchants to fully automate against our decision and to trust that they are not shipping bad orders or declining good ones.
Which other merchant systems does the platform integrate with?
We have simple plugins for mid-market merchants on platforms like BigCommerce, Magento, Salesforce Commerce Cloud, and Shopify, meaning our solution is available in those platforms’ app stores. In the enterprise space, we offer a direct REST API integration, which allows enterprises to fully customize their integration setup.
How do you measure success for your clients?
Our solution is designed to align our interests with our customers’ interests. We’re only paid for orders we approve, so, like our merchants, we’re focused on making sure they are able to ship every legitimate order.
But we don’t measure success solely by the number or percentage of orders shipped. We provide value in four ways. First, we optimize revenue by ferreting out fraudulent orders. Second, we identify legitimate orders that might have been mistakenly declined when manually reviewed or evaluated with a score delivered by a static system. Third, we reduce operational costs by providing a highly automated fraud review and order management system. And finally, we are able to automate much of the chargeback management process with our Chargeback Recovery product, which pretty handily beats the industry rate for recovering non-fraud chargebacks and can address consumer abuse scenarios.
Improving merchant performance on all these things substantially improves the customer experience that our customers are able to provide their customers. Orders aren’t delayed for manual review. Good customers’ orders aren’t falsely declined. Merchants are able to better understand which customer complaints are legitimate and which are being levied by consumers looking to take advantage of a merchant.
But after all that, maybe the most straightforward measure of our merchants’ success is the ROI they realize with Signifyd. On that count, Forrester conducted a Total Economic Impact report for us, looking at our results with a large omnichannel retailer. In that case, Forrester concluded, Signifyd delivered an ROI of 3.8x over a three-year stretch.
How is Signifyd working to combat the issue of Malicious AI?
In fraud, or with any other cybersecurity field, you typically are operating on this edge case of Machine Learning, confronting a malicious opponent, which already is more challenging than other AI scenarios.
When you think of other uses of AI, even in e-commerce, you might think of a personalization and sophisticated engines that aspire to provide one-to-one personalization based on a shopper’s past behavior and transactions.
In those cases, the shopper is not out to trick the machine. They want the machine to help them. Like when I talked about Spotify. I’m not going to trick the machine so that it builds playlists that I don’t like. I want the machine to learn, so I help it as much as I can. But a malicious opponent knows a machine is at work. A malicious opponent wants to beat the machine.
My colleague Swami Vaithianathasamy wrote a fairly detailed piece on this very subject for AiThority. He talked about how fraudsters are now engaged in model extraction a process by which they all but hijack Machine Learning models in the cloud by reverse engineering their methods for identifying fraud.
There are ways to stay ahead of malicious opponents. At Signifyd we make sure to have domain experts working side-by-side with the machines. We have experienced people that can anticipate the movements of fraudsters and counter new attacks in real time, leveraging Machine Learning as the tool in their toolboxes.
What’s your view about AI working in conjunction with humans?
The role of AI is really to augment what people do. Taking the grunt work off their plates, and the repetitive and dangerous work that can be done by AI and bots for that matter. Someday we’ll commonly see AI running bots to execute physical work.
AI is a liberator. Humans can use their time and energy assessing challenges and tapping into their creative potential to overcome challenges versus following structured rules and following checklists. Those are the machine’s job.
But and maybe it sounds funny to say we’re not going to leave the machines with just grunt work either. Machines make valuable research assistants. One of the things they are good at is processing huge amounts of data. They can turn up gems that form the basis for human researchers to explore bigger ideas.
Think of all the other fields where AI can serve as helpful assistants to researchers. Financial analysts and advisors can turn to machines to comb through thousands or millions of financial reports and stock performance data to steer human advisors’ course on investment planning.
Machines can plow through property records, court records, government spending reports and city council minutes to provide investigative journalists with the grist for in-depth stories that help citizens better understand the world around them.
Physicians and medical researchers turn to big data and AI to understand the origin and course of diseases. That understanding allows them to unleash their expertise to design treatment and cures for some of the world’s most stubborn ailments.
I believe the majority of jobs will be affected by AI. The way we work and what we work on will continue to change. So, the main topic we should be thinking about as a society is, how do we enable our workforce to embrace AI. How do we make sure workers have the skills and ability to do the new kind of work?
Where do you see AI/Machine Learning and other smart technologies heading beyond 2025?
When I think about AI today, most of it is where AI in fraud was until a few years ago. Fraud-detection systems would use Machine Learning to deliver a score, a number that would indicate where on the spectrum of almost-certainly-fraudulent to not-fraudulent an order would lie.
How has AI automated machine-based decision making?
Today, a lot of what we see in AI is machines making a recommendation and humans making decisions.
The next stage is for the AI to make a decision for a person. The person monitors that decision and governs it, which is basically what Signifyd does in the fraud space. The machine makes the decision on its own and merchants can automatically feed that decision into their order management system so good orders go out without human intervention – and delays. But all the while domain experts are monitoring the decisions and the results of those decisions.
Think about self-driving cars. They don’t simply tell drivers where to go, how fast to go when to stop, swerve, slow down. They do it. They drive. But for now, when we see those distinctive Waymo crossovers breeze by, there is always a human behind the wheel and we expect them to be responsible for what happens with that vehicle.
In fact, it’s worth noting that a lot of the accidents involving self-driving cars involve human error either the supervising human or the driver of a second car.
We’re moving relatively rapidly to a world where we’ll let the cars drive. It requires real-time decisions made by machines, not in the cloud but running on local hardware, or edge computing. Machines that are available constantly, no matter the conditions.
That is the point at which we have autonomous machines self-driving cars. I’m eager to see them on the street.
Which AI start-ups and labs are you keenly following?
I’m personally very interested and passionate about making sure young people, fresh out of college or high school, take up the opportunity to think of a career in AI and tech as an opportunity. Every job will be affected by AI. Some literacy in how to work with artificial intelligence, with agents and assistants, will come in very handy. I see a huge demand in the market for this generation of engineers that can actually build, develop and maintain these AI systems.
All that to explain that the labs I’m most interested in are the colleges in the U.S. and around the world. We’re starting to formulate ways to cooperate with universities to build curricula that synch AI with the cycle of a business and real-life applications.
Universities also give us our best chance of diversifying the field of AI. We need to do more to make sure that the next generation of AI engineers is very diverse, that it looks like the users and consumers it is serving.
Diverse thinking is important in developing any product or technology. AI can’t be male-dominated. Diverse points of view result in better products and help expose blind spots that a dominant demographic in a field might have.
In the early days of voice recognition, devices had trouble recognizing female voices a problem that persists to some extent. The systems were designed and trained by men.
Without diversity, it’s hard to correct AI biases when they appear. Consider a machine analyzing a hospital scene. The machine may identify men in scrubs as doctors and women in scrubs as nurses. If the domain experts monitoring the AI share the same biases they won’t engage in exercises designed to correct the bias by questioning the datasets.
What technologies within AI and computing are you interested in?
Graph databases are changing the game when it comes to storing knowledge. SQL databases rely on tables. But our reality is structured in nodes and pointers. Think of an org chart. It’s a network structure.
Representing that in traditional database systems turns up tons of complexities, particularly if those networks become huge datasets – as is typically the case for Machine Learning scenarios.
This is where graphs can complement AI. Graphs can store context, the story of how things fit together, how they belong to each other. If you start taking the connection of things into account, machines can figure out how one thing impacts another.
As AI will be able to read and store contextual data in these new types of graph databases, it will allow us to do things that aren’t even conceivable now.
What’s your smartest work related shortcut or productivity hack?
I’d have to say my adherence to David Allen’s “Getting Things Done.” It’s a book and a strategy, or philosophy really, about how to efficiently deal with the onslaught of information, requests, and tasks that confront us every day.
In simple terms, the method requires a significant organizational effort to start, but then it provides a system of priorities, independent from rigid deadlines, allowing flexible adjustments. Today we also call this agile.
I’ve adapted it somewhat to run our marketing team. Essentially we take a scrum approach, similar to what engineers do when working in an agile system. The system consists of constant planning and sprinting. Planning projects and running a sprint to get them done, without distraction.
I’ve found it to be very effective.
What would have been your alternate career choice?
Likely an artist. I’m not exactly sure of what kind. What interests me about artists is that they have the ability to tap into their creativity to identify things that people haven’t seen before and then they find a way to show it to them.
It could be in all sorts of forms: visual arts, music, film. I think I’d enjoy researching and finding compelling themes, topics, and trends and then iterating on these concepts, developing new angles and creating something different to share as my contribution to the community.
That desire is what’s been driving me in my actual career as well. And I think it would be driving me no matter the career I chose.
Thank you, Stefan! That was fun and hope to see you back on AiThority soon.
Stefan Nandzik is Signifyd’s Vice President of Corporate Communications. His “what if” approach to business problem-solving constantly challenges conventional wisdom and means that he is never afraid to upend the status quo to lead change.
Signifyd, the world’s largest provider of guaranteed fraud protection, enables online retailers to provide a friction-free buying experience for their customers. Signifyd leverages big data, machine learning and domain expertise to provide a 100 percent financial guarantee against fraud on approved orders that later turn out to be fraudulent. This effectively shifts the liability for fraud away from retailers, allowing them to increase sales and open new markets while reducing risk. Signifyd counts among its customers a number of companies on the Fortune 1000 and Internet Retailer Top 500 lists. Signifyd is headquartered in San Jose, CA.