histories, building sophisticated algorithms to determine trustworthiness? You wouldn’t walk into NASA and say, “Why build all those fancy spaceships when all you need is a slingshot?”
Still, out of respect for Benchmark, Thompson thought he’d indulge Shaked for a few more minutes. “So where did you learn how to do this?” he asked.
“Hunting down terrorists,” Shaked said matter-of-factly. His unit in the army had been tasked with helping to catch terrorists by tracking their online activities. Terrorists move money through the Web with fictitious identities. Shvat’s job was to find them online.
Thompson had heard enough from this “terrorist hunter,” too much even, but he had a simple way out. “Have you tried this at all?” he asked.
“Yes,” Shaked said with quiet self-assurance. “We’ve tried it on thousands of transactions, and we were right about all of them but four.”
Shaked said his company had analyzed forty thousand transactions over five years, since its founding.
“Okay, so here’s what we’re going to do,” Thompson said, and he proposed that he give Fraud Sciences one hundred thousand PayPal transactions to analyze. These were consumer transactions PayPal had already processed. PayPal would have to scrub some of the personal data for legal privacy reasons, which would make Shvat’s job more difficult. “But see what you can do,” Thompson offered, “and get back to us. We’ll compare your results with ours.”
Since it had taken Shvat’s start-up five years to go through their first forty thousand transactions, Thompson figured he wouldn’t be seeing the kid again anytime soon. But he wasn’t asking anything unfair. This was the sort of scaling necessary to determine whether his bizarre-sounding system was worth anything in the real world.
The forty thousand transactions Fraud Sciences had previously processed had been done manually. Shaked knew that to meet PayPal’s challenge he would have to automate his system in order to handle the volume, do so without compromising reliability, and crunch the transactions in record time. This would mean taking the system he’d tested over five years and turning it upside down, quickly.
Thompson gave the transaction data to Shvat on a Thursday. “I figured I was off the hook with Benchmark,” he recalled. “We’d never hear from Shvat again. Or at least not for months.” So he was surprised when he received an e-mail from Israel on Sunday. It said, “We’re done.”
Thompson didn’t believe it. First thing Monday morning, he handed Fraud Sciences’ work over to his team of PhDs for analysis; it took them a week to match the results up against PayPal’s. But by Wednesday, Thompson’s engineers were amazed at what they had seen so far. Shaked and his small team produced more accurate results than PayPal had, in a shorter amount of time, and with incomplete data. The difference was particularly pronounced on the transactions that had given PayPal the most trouble—on these, Fraud Sciences had performed 17 percent better. This was the category of customer applicants, Thompson told us, that PayPal initially rejected. But in light of what PayPal now knows from monitoring the rejected customers’ more recent credit reports, Thompson said, those rejections were a mistake: “They are good customers. We should never have rejected them. They slipped through our system. But how did they
Thompson realized that he was looking at a truly original tool against fraud. With even less data than PayPal had, Fraud Sciences was able to more accurately predict who would turn out to be a good customer and who would not. “I was sitting here, dumbfounded,” Thompson recalled. “I didn’t get it. We’re the best in the business at risk management. How is it that this fifty-five-person company from Israel, with a crackpot theory about ‘good guys’ and ‘bad guys,’ managed to beat us?” Thompson estimated that Fraud Sciences was five years ahead of PayPal in the effectiveness of its system. His previous company, Visa, would never have been able to come up with such thinking, even if given ten or fifteen years to work on it.
Thompson knew what he had to tell Benchmark: PayPal could not afford to risk letting its competitors get hold of Fraud Sci-ences’ breakthrough technology. This was not a company Benchmark should invest in; PayPal needed to acquire the company. Immediately.
Thompson went to eBay’s CEO, Meg Whitman, to bring her into the loop. “I told Scott that it was impossible,” Whitman related. “We’re the market leader. Where on earth did this tiny little company come from?” Thompson and his team of PhDs walked her through the results. She was astounded.
Now Thompson and Whitman had a truly unexpected problem on their hands. What could they tell Shvat? If Thompson told this start-up’s CEO that he had handily beaten the industry leader, the start-up’s team would realize they were sitting on something invaluable. Thompson knew that PayPal had to buy Fraud Sciences, but how could he tell Shvat the test results without
