The useful thing about failure, in Evan Reiser's telling, is that it leaves evidence. Long before he became the founder of a cybersecurity company valued at $5.1 billion, he was a college student whose appetite for late-night games had outrun his discipline. Reiser had grown up absorbed by the early internet: playing games, building them, and worrying at online puzzles until the hour became unreasonable. At Rensselaer Polytechnic Institute, unreasonable hours sent him into a wall. He failed his second semester.
The summer afterward was less cinematic than corrective. He took a job at an ice cream shop and wrote to the dean asking for another chance. The request worked. More important, so did the reset. He returned, made the dean's list, and earned a degree in computer systems engineering and computer science. “That was when I developed my work ethic,” he later said. Many founder biographies sand an early stumble into a decorative bit of grit. Reiser keeps the splinters in his version. They explain what came next.
In Manhattan, at his first software engineering job after college, he arrived by seven in the morning and stayed late. The former gamer had not misplaced his capacity for obsession. He had given it office hours. Still, industriousness was only one lesson. His twenties would supply several others, all carrying invoices.
The server bill always arrives
In 2007, Reiser started GamerNook, a social space where gamers could meet and talk. Users sent messages about how much the community meant to them. Meanwhile, its founder spent his savings on server bills, ate from Burger King's dollar menu, and slept on a friend's couch. He even played poker to help finance the attempt. GamerNook had emotional product-market fit and dreadful arithmetic.
His next company supplied a more sophisticated education. Bloomspot sold luxury experiences, roughly the tailored-suit cousin of Groupon. It made money and employed more than 100 people. Yet the cost of acquiring each new customer kept climbing. Scale, Reiser learned, can magnify a flaw as efficiently as it magnifies a virtue. JPMorgan Chase acquired the business in 2012, but the unit economics stayed with him.
“You have to build things that lots of people value.”Evan Reiser, on the lesson behind scale
That sentence sounds obvious until one notices how many young companies build for applause, acquisition, or the flattering geometry of a growth chart. Reiser had tried applause. He had studied growth. With AdStack, founded in 2011, he moved closer to durable value by using behavioral data science to personalize advertising and email. TellApart acquired AdStack; Twitter then acquired TellApart. The serial founder became an operator inside a large public platform.
From predicting attention to protecting trust
At Twitter, Reiser led product and machine-learning teams in the advertising business. He managed an organization of about 50 engineers tied to roughly $2 billion in annual revenue. The work turned patterns of behavior into predictions: which message might interest a person, which placement might perform, which auction might clear. He also learned a manager's less glamorous craft: put people in roles suited to their strengths, state mutual expectations plainly, and operate with enough discipline that clever systems remain useful.
By 2018, he was ready to run a company again. Working from Greylock's incubator, he spoke with enterprise technology leaders about persistent problems. Business email compromise kept appearing. The attack was oddly resistant to familiar security machinery because it often contained no malicious attachment and no suspicious link. A compromised account could ask a finance employee to change payment details in perfectly grammatical prose. The message looked harmless in isolation. Its danger lived in context.
The behavioral loop
Reiser and Sanjay Jeyakumar founded Abnormal around a reversal of perspective. Instead of beginning with a catalog of known bad signals, the system would learn the normal behavior of an organization and notice deviations. Who usually communicates? Which vendor receives payments? Does this executive normally make requests in this way, at this hour, from this account? The advertising years had taught Reiser how to model human patterns. Cybersecurity gave the method a different moral weight. The model was no longer trying to win a click. It was trying to prevent a confidence trick.
A company named for the thing it hunts
The founders initially called the company Abnormal AI. In 2018, the name produced the wrong kind of anomaly. Enterprise buyers heard artificial intelligence and imagined science fiction rather than applied software. So the founders chose Abnormal Security, a sensible name for a market still learning to separate machine learning from movie robots.
The product found its opening. It connected to Microsoft 365 or Google Workspace through APIs, modeled the communication environment, and made decisions without forcing every message through another gateway. By August 2024, Abnormal had crossed $200 million in annual recurring revenue. A $250 million Series D valued the company at $5.1 billion. The company reported year-over-year recurring revenue growth above 100 percent and customers across Europe, Asia, Australia, and North America.
In April 2025, it reclaimed the Abnormal AI name. The circle was tidy, though the market around it was not. Generative models had made polished, personalized language cheap. A plausible note from a supposed colleague no longer required a patient fraudster or even fluency. At the same time, defenders faced more alerts than security teams could examine by hand. The company said its technology was making millions of autonomous decisions each day, protecting more than 3,000 organizations, and had blocked nearly $1 billion in fraud.
The human endpoint
Reiser's argument has widened with the product. Email is only one place where people approve payments, share credentials, and trust a familiar name. Abnormal has extended its behavioral approach toward cloud applications and identity. Its stated ambition is to understand the human layer across the digital workplace, including services such as Slack, Workday, ServiceNow, and Zoom. Email remains the data-rich beginning, not the border.
That expansion brings an uncomfortable bargain. To catch subtle deception, a system needs considerable context. To act autonomously, it must earn confidence in its reasoning and accuracy. Reiser has increasingly talked about explainable AI and trust, particularly as agents begin doing security work rather than merely recommending it. An autonomous system that saves analysts from repetitive triage is useful. One that cannot explain a consequential mistake merely automates the postmortem.
He also sees an asymmetry. Attackers can use AI to produce more tailored attempts at greater speed, while defensive teams remain limited by headcount and attention. “The rate of acceleration is insane,” he told Axios in 2025. At a later discussion about agentic security, he framed the imbalance in labor terms: defenders already have more work than people. His answer is not to remove humans from security. It is to reserve them for judgment, strategy, and the ambiguous cases machines should not settle alone.
GamerNook
A loved gaming community teaches the difference between devoted users and a sustainable company.
Bloomspot
Revenue, a team of more than 100, rising acquisition costs, and an eventual sale to JPMorgan Chase.
AdStack, TellApart, Twitter
Behavioral modeling moves from a startup into a machine-learning business operating at global scale.
Abnormal
The same instinct for patterns is redirected toward social engineering, compromised accounts, and cloud behavior.
The value of an untidy résumé
There is a temptation to read Reiser's career backward from the valuation and make every detour look preordained. It was not. GamerNook did not fail in order to provide a charming opening anecdote. Bloomspot's economics were not a rehearsal staged for a future investor meeting. Twitter was not simply a finishing school for cybersecurity. Each episode had its own stakes, and several could easily have led somewhere else.
What connects them is Reiser's habit of retaining the lesson without preserving the plan. User affection matters, but value needs economics. Models can reveal behavior, but only when attached to a useful decision. A talented team needs explicit expectations. A product name must meet buyers where they are. Trust is not a speech; it is the accumulated result of correct decisions, clear explanations, and the humility to notice when the pattern has changed.
He has said that if he won the lottery, he would keep doing the same work. It is an earnest founder line, almost suspiciously so. Yet his record gives it texture. Reiser has been building internet companies since the years when a server bill could push dinner toward the dollar menu. He wants Abnormal to remain independent, employ thousands of people, and keep moving into problems where behavioral data can protect rather than persuade.
The student who once failed because he could not stop exploring the internet now runs a company designed to notice when the internet behaves strangely around a person. There is wit in that symmetry, and a warning too. Normal is not a permanent setting. It is a living pattern, revised every day. Reiser's career suggests the same is true of a founder.