Netflix Once Paid $1 Million for a Math Problem

Long before Netflix streamed anything, it mailed DVDs in red envelopes and quietly obsessed over one number: how well it could predict what you'd rate a movie. That prediction engine, called Cinematch, was decent but far from perfect, and the company wanted better.

So in 2006, Netflix did something almost no company does with its core technology. It published nearly 500 million real customer movie ratings publicly online and invited the entire world to try improving its algorithm.

The offer was simple and enormous: whoever could beat Cinematch's accuracy by just 10% would win one million dollars in cash. Thousands of teams from over 180 countries jumped in, including statisticians, physicists, and students with laptops in dorm rooms.

The competition ran for nearly three years, and progress crawled at first. Teams inched toward the 10% threshold fraction by fraction, some improving accuracy by mere hundredths of a percent at a time.

In 2009, a merged team called BellKor's Pragmatic Chaos finally crossed the line, combining over 100 different algorithms into one system. Netflix paid out the full million dollars, exactly as promised, to a group that had never worked for the company.

Here's the twist nobody saw coming: Netflix never actually implemented the winning algorithm. By the time the prize was awarded, the company had pivoted hard into streaming, and DVD ratings mattered less than viewing behavior.

The million-dollar formula ended up shelved, but the contest itself reshaped how tech companies think about data. It proved that sometimes the smartest solution to your own problem is sitting outside your building, waiting for the right invitation.