Did a robot reporter screw up the AP's Netflix earnings story?

Earlier today, Netflix reported profits of $26.3 million for the second quarter, on $1.64 billion in revenue. Those were better-than-expected numbers, reflective of a solid quarter, and they sent Netflix’s stock sharply higher in after-hours trading.

But if you read the Associated Press’s report on Netflix’s earnings, you might have been confused.

The AP (whose story was syndicated by the Los Angeles Times, CNBC, and other outlets) carried the headline “Netflix earnings miss Wall Street forecasts.” The AP reported that Netflix’s profits of 6 cents per share had come in well below analysts’ expectations of 32 cents per share.

But other outlets (correctly) reported that Netflix had in fact surpassed analysts’ expectations of 4 cents per share in profits.

What gives?

Well, for starters, the reason behind the discrepancy is likely due to the fact that Netflix recently split its stock 7-for-1. That means that the company issued six new shares for each share of existing stock, making each individual share worth only one-seventh as much. So if analysts had expected Netflix to earn 32 cents a share before the stock split, dividing those estimates by 7 would mean that the analysts were expecting about 4 cents of earnings per share of the new Netflix stock.

Maybe a reporter for the AP messed up the math on Netflix, using the old, pre-split estimates instead of the correct, post-split ones.

But there’s another possibility: maybe a robot is to blame.

Last year, the AP struck a deal with a company called Automated Insights, which makes automated reporting software that can write certain types of stories without human assistance. Among the things Automated Insights’ software can do is write simple corporate earnings stories, using numbers it pulls in from an automated feed. The AP now publishes more than 3,000 earnings stories per quarter with Automated Insights’ help.

We don’t know exactly what happened in today’s case of mistaken Netflix earnings. But we know that today’s incorrect AP story on Netflix’s earnings was aided by robot hands. On a version of the story picked up by CNBC, a disclaimer appeared at the bottom of the page:

It’s possible, in other words, that the Automated Insights software pulled in data from a source that contained the wrong, pre-stock-split earnings estimates for Netflix, and used those to generate its earnings report.

Lou Ferrara, the AP’s vice president of business news and sports entertainment, told Fusion in a phone interview on Wednesday that the source of the Netflix error wasn’t yet known. “We’ll have to go review that, but any time there’s an error we fix it,” he said. If Automated Insights had, in fact, pulled in the wrong numbers, he said, it would be “no different than when a human makes an error.” (The AP’s story has since been changed on most sites, including the LA Times and CNBC, to reflect the correct earnings estimates.)

Ferrara defended the AP’s decision to rely on Automated Insights for earnings stories, saying that it had “significantly” reduced the number of errors. “We have a system in which we’re generating up to 3,700 earnings reports a quarter,” he said. “By the end of the year, we should be up to 4,700. At that volume, there may be an error from time to time.”

It’s true that human reporters make mistakes requiring corrections all the time, and that the coming wave of software-based reporting assistance will be great for journalists in all kinds of ways. Still, robots aren’t perfect. And with high-stakes journalism like quarterly earnings reports, where millions of dollars can change hands in the stock market based on what’s reported, it’s clear that the robots of the future will still need some old-fashioned human editors.