Fighting fraud and reading the political tea leaves have a lot in common. It’s all about what data is collected, how it is interpreted and understanding how that data reflects reality — if it even does.
In this political season, it’s impossible to escape the slew of polls the media flashes across the 24-hour news screens. But, as we’ve seen this election season, poll results can be wildly off.
But what does that have to do with fraud and payments?
More than you may think, which was demonstrated by a conversation that MPD CEO Karen Webster had with Forter CEO Michael Reitblat on the lead-up to the next big primary in New York tomorrow (April 19).
As the founder of a company that helps merchants stop fraudsters from bilking them out of their hard-earned money, Reitblat brings his years of work in fraud (and the Israeli Army) to understand how people work by observing behavioral data. Webster thought there might be some lessons to be learned when observing the wild swings that we’ve observed over this very unusual political season.
“The similarities here are that some people lie, and you need to understand which ones are the ones who do and you need to look out for — which is the same as in payments fraud,” Reitblat said. “The second thing is: People change all the time, and part of the whole poll model today is talking to a small group of people that you think represent the whole population that you are measuring — and then realize that’s not the case.”
In fraud prevention, Reitblat says, you realize pretty quickly that every person and every transaction is different, so it’s impossible to generalize. Pollsters generalize. They take a sample from a small group of people (1,000 people) and try to guess what it says about a group of 10 million people, Reitblat points out.
The obvious flaw?
“Interests in those groups change all the time,” he says. “Anyone looking to understand and predict behavior needs to stay up to date — to change those groups. The second piece is how to assemble a group with the right people.”
Which is why some pollsters have begun to turn to machine learning in order to track broader groups of people and gain a more accurate estimate of behavior. Machine learning, of course, is a common method used for fraud analysis.
But there are gaps in that system, too, Reitblat says. And while the pollsters are under high pressure to deliver specific results, they often skip over the one big factor needed to truly deliver results that may be rooted in reality.
“The problem with machine learning is that you don’t actually understand a person individually,” he says. “[Pollsters] don’t actually learn about a specific type of person. To be accurate, you need to understand the people and what they won’t actually lie about, which will give you an insight into what they are really thinking.”
Margin Of Error
The other problem pollsters face is that they often use outdated methods of data collection (phone calling), which can generate an unrepresented sample that only reflects one type of voter — people with landline phones who answer them. If that same technique was applied to studying how fraudsters commit fraud, fraudsters would win more often than not.
“There are limitations by just observing people who are talking to pollsters since a lot of people don’t want to,” Reitblat explains. “Pollsters might be missing whole types of the population.”
This is where how pollsters study people and how Forter and Reitblat study fraudsters differ. Or, as Webster points out: You don’t necessarily ask fraudsters what they are doing. You observe their behavior.
Seeing what they are doing is the best way to understand what they might do in the future.
This isn’t quite how pollsters work.
The problem with asking voters how they will vote or how they feel about a particular issue is that people don’t always know what they want. And, for people who think they do, that opinion could change not long after they are asked the question.
And that point of view has the potential to change multiple times.
Manipulation Of Stats
Without pollsters actually recording what people say and seeing how it stacks up to how they actually act, it’s hard to present accurate data.
In that regard, Reitblat said, tracking fraudsters is almost easier than pollsters tracking people.
“Our job is almost easier because we can track a behavior of a person across a long period of time. We can say this is how those bad guys usually behave, this is what changed. We see a lot of different transactions — whether with a website or actual payment. Where a pollster, in most cases, is looking once every few years. People can change a lot in four years, so whatever you knew about them then doesn’t matter anymore.”
Which is why, in so many cases, a lot of poll results are off by a huge margin.
And because people are human and humans are prone to change their minds a lot, humans can be swayed by many, seemingly arbitrary things — something as small as an off-the-cuff remark a candidate makes during a debate. Or a political ad. Or a poll, for that matter.
The influence of which can’t be underestimated. Sometimes, polls become self-fulfilling prophecies as people will move toward the candidate they see as the winner — which sways public opinion, which then sways the polls.
Which isn’t always that different than fraudsters. As Webster points out: At their very core, fraudsters are human beings. But one thing pollsters are still struggling to figure out is how to understand human beings and how to calculate for shifting mindsets.
“Fraudsters are people … Because we see what they are doing, and have done, we can actually predict their behavior pretty well,” Reitblat says.
But pollsters? That’s still a work in progress.