Jason Tan Built a Machine to Spot Risk. Then He Turned the Lens on Leadership.

The useful thing about rejection is that it rarely consults your five-year plan. As a University of Washington senior, Jason Tan had narrowed the future to three respectable doors: Google, Amazon and Microsoft. He interviewed at all three. None opened. He was young, technically able and, by his own account, miserable at the whiteboard theater of someone watching him solve a problem. The dream evaporated one nervous interview at a time.

Then Zillow called. A friend had interned there and passed along his name. Tan knew little about startups and less about the new real-estate site, but failure has a marvelous way of improving one's curiosity. He joined as an engineer in 2006 and discovered the pleasure of a small team: responsibility arrived early, contribution was visible, and good work could outrun a formal title. Fred Sadaghiani, his first manager there, would later become Sift's founding chief technology officer. Zillow co-founder Rich Barton would invest. What felt like a detour was quietly assembling a network.

This is not quite the fable Silicon Valley usually tells about destiny. Tan did not emerge from childhood announcing a company into a hairbrush. Born in Taipei, he moved through Japan and Singapore before his family settled in Seattle in 1997, when he was 12. He was good at mathematics, enjoyed tinkering with computers and judged software engineering to be both interesting and likely to pay the bills. At 16, through the University of Washington's Academy for Young Scholars, he skipped the last two years of high school and started college. Ambition, in his version, came with a spreadsheet's practicality.

Five ideas walk into an accelerator

After Zillow came engineering work at Optify and a turn as chief technology officer at BuzzLabs. In 2011, Tan moved to San Francisco with co-founder Brandon Ballinger for Y Combinator's summer batch. They arrived with faith in machine learning, a list of five possible businesses and no special knowledge of fraud. So they asked people they knew inside companies a blunt question: what difficult problem would you pay someone else to solve? Fraud kept appearing.

The existing tools were dominated by static rules. Block this country. Flag that dollar amount. Stop a transaction matching yesterday's bad pattern. Fraudsters, inconveniently, were not static. A rule built after an attack was a photograph of a getaway car. Tan and Ballinger wanted a moving picture: software that could learn from behavior across transactions and make a risk judgment in real time.

Sift Science launched a public product in 2013. Its first promise was prosaic and useful: an API that let online merchants use large-scale machine learning without building the machinery themselves. By 2018, Tan said the platform was classifying more than 35 billion user events a month across 16,000 sites. The vocabulary widened from payment fraud to account takeover, fake accounts, spam and content abuse. The company shortened its name to Sift; the larger idea became “digital trust and safety.”

Trust is a dangerous word in software because it sounds warm until the transaction fails. Tan's version was probabilistic. Most legitimate customers should move quickly; suspicious behavior should earn scrutiny. The aim was neither universal suspicion nor blind welcome. It was judgment at the right instant. His favorite analogy was airport security: the internet should feel more like a well-run pre-check lane than an interrogation every time someone logs in.

The product beneath the product

The machine-learning argument was technical. The company-building argument was personal. Tan has said that one reason he started a company was to choose the people with whom he would spend eight to twelve hours a day. Work occupied too much life to make the company around it an afterthought. He wanted colleagues who learned from one another, solved hard problems and had fun doing it.

There is evidence of the fun. Tan enjoys freestyle rapping and once nominated “Great Wall” as his future DJ name. His childhood AIM screen name was jtgameover. His sisters compared him to Po from Kung Fu Panda, which he translated as goofy, free-spirited, optimistic and resistant to adversity. Barton, writing about their Zillow years, remembered someone who organized soccer and music for happy hour as readily as he wrote code. Serious companies are sometimes built by people willing to look ridiculous before lunch.

But culture is not office entertainment with better catering. On Sift's fifteenth anniversary, Tan called it a Type 1 decision - difficult to reverse and painful to repair. Strategy, go-to-market plans and roadmaps can misfire and be changed. Culture teaches everyone how decisions are really made. Leave a defect there and the organization reproduces it.

That conviction acquired scale. Sift raised $53 million in a 2018 Series D, taking its announced funding total at the time to $107 million. Tan has since said that during his CEO tenure the company grew from an idea to roughly $100 million in annual recurring revenue and more than 400 employees. Numbers of that size are catnip to business profiles. They are also poor witnesses. They can tell you that a company expanded without explaining what the expansion asked of the people inside it.

A jazz musician leaves the podium

In January 2021, after nearly a decade as chief executive, Tan announced that he would move into the role of founder and executive chairman. His metaphor was musical: he was a jazz musician, while Sift at its new scale needed an orchestra conductor. The line carries more humility than the usual founder handoff. Improvisation had produced the company. Operating it now demanded a different rhythm.

He would remain involved in board alignment, long-term strategy, capital decisions and exploratory technology. The larger ambition was institutional: he wanted Sift to endure for decades, which meant designing a company able to outlive any one person's authority. A founder can claim to be building an institution. Succession is the moment the claim receives a test.

Titles, of course, never surrender as neatly as a baton in a press photograph. Founders fuse identity to the organization because for years the fusion is rewarded. Tan's recent essays are frank about the resulting distortions. He describes chasing milestones, mistaking motion for impact and allowing fear to power decisions that looked responsible in the short term. The vocabulary is striking beside the hard nouns of cybersecurity: love, service, vulnerability, enoughness. He is no longer merely explaining how an algorithm judges a customer. He is asking what fuels the person who judges everything else.

There is an elegant loop here. Sift began because static rules failed against people who adapted. Tan's latest leadership argument makes much the same claim about executives. Old reflexes can survive after their useful moment. A behavior that protected a ten-person startup can constrain a global company. The leader has to update, introduce context and accept that confidence is not the same as accuracy.

The fraud fighter's second model

At 40, fifteen years after Sift began, Tan published advice to his 25-year-old self. The notes were less concerned with fundraising or product-market fit than one might expect. Pay attention to the work that does not resemble work. Tell the truth about fear. Do not confuse relentless effort with useful effort. Build a life capable of holding ambition without being entirely defined by it.

This is not a repudiation of the company he built. It is an attempt to understand the conditions under which the building was done. Sift's founding insight was that better signals permit better choices: fewer false alarms, less needless friction, more room for good actors to move. Tan now applies that principle to leadership. Notice the signal. Examine the pattern. Add friction only where it helps. Update the model before the model hardens into fate.

The failed Big Tech interviews remain the perfect opening scene because they expose the limits of one judgment. Three famous companies looked at a nervous young engineer and passed. Zillow looked again. Years later, thousands of businesses would use his company to make their own split-second calls about strangers online. The irony is gentle and durable: every system for judging risk must leave room for the possibility that it is wrong.