Facebook Apologizes for Overstating Video Metrics

Facebook apologized on Friday for an error in the way it measured video viewership, a miscalculation that greatly overstated how much time, on average, its users were spending watching videos.

Facebook apologized on Friday for an error in the way it measured video viewership, a miscalculation that greatly overstated how much time, on average, its users were spending watching videos.

The incorrect numbers were displayed to partners, including advertisers and publishers, for more than two years, as Facebook aggressively challenged YouTube for online video dominance and urged partners to embrace video publishing and advertising on the platform.

While Facebook called the problem a “discrepancy,’’ it is a troublesome admission for major advertising agencies and publishers that rely on Facebook’s metrics to assess their investments on the platform — whether a 15-second ad or a lengthy video production.

It is also likely to add to advertisers’ long-held frustrations with Facebook and other internet platforms over how closely they guard valuable audience data, and the limits the tech companies place on outside measurement companies.

“About a month ago, we found an error in the way we calculate one of the video metrics on our dashboard — average duration of video viewed,” wrote David Fischer, a vice president of video and marketing partnerships for Facebook, in a post on Friday. “While this is only one of the many metrics marketers look at, we take any mistake seriously,” he wrote. “Our clients’ trust and belief in our metrics is essential to us, and we have to earn that trust.”

Facebook acknowledged in an update to its advertiser support page last month that it had been excluding videos watched for fewer than three seconds from the average viewing times. That greatly inflated the figures presented to partners. (Facebook publicly counts video views after three seconds. By comparison, YouTube counts a view after 30 seconds; Vine, where videos are limited to six seconds, counts after a full video loop.)

Advertisers, however, buy Facebook video according to different metrics: either by “impressions,” which are measured as soon as a user sees a video in the news feed, or in 10-second views.

“Average duration of video viewed” is neither a public metric nor does it determine how much advertisers pay for video placement. But the figure is displayed prominently in partners’ dashboards, and it has helped assuage discomfort about the platform’s generous view-counting methods.

While Facebook’s update last month largely passed under the radar, Mr. Fischer’s post on Friday was prominent and contrite. “We sincerely apologize for the issues this has created for our clients,” he wrote.

Facebook has made a number of stumbles lately that it has had to scramble to fix. In May, it grappled with reports that editors of its Trending Topics — which displays some of the most talked-about stories on the social network — were suppressing conservative content. The company later laid off the Trending Topics team.

And earlier this month, Facebook had to reverse itself after taking down a Pulitzer Prize-winning Vietnam War photo because the picture, which featured a horror-stricken and naked 9-year-old girl, violated the company’s content standards for nudity.

But none of those problems hit the core of Facebook’s business — digital advertising — the way the discrepancy over measurement of video viewership does. In July, Facebook reported blockbuster sales, up 59 percent from a year ago to $6.44 billion, and emphasized it was betting on video and video advertising for future growth.

“We see a world where video is first, with video at the heart of all of our apps and services,” Mark Zuckerberg, Facebook’s chief executive, said in a conference call with investors at the time.

Advertisers are unlikely to cut back or change spending on Facebook in a big way because of the metric error, but it could affect “larger brand-focused advertisers who may have been influenced to alter their budget planning’’ based on Facebook’s previous data, said Brian Wieser, a media industry analyst at Pivotal Research in New York.

It could also benefit companies with digital video inventory like YouTube, Twitter and Snapchat, he added.

Above all, it reinforces the importance of outside measurement when it comes to data from Facebook and other media companies.

“While publishers’ data has the potential to be more accurate, it also has the potential to be more inflated,” Mr. Wieser said. “This risk supports an ongoing role for third parties in the field of measurement.”

Jill Sherman, senior vice president for social strategy at the agency DigitasLBi, noted that the metric was one of many that influences investment decisions in Facebook video and that it remained a viable platform for brands. Still, she said, Facebook’s acknowledgment could make it harder for it to compete with YouTube.

“If the data suggested that people were staying with video longer on the Facebook platform, that’s an indication people were starting to use the Facebook platform similar to a platform like YouTube,” she said in an interview. That would be a crucial “behavioral shift,” she said.

With the new revelation about the error, she added, “the implication is really there that how we’ve been perceiving the evolution of the Facebook platform may be off.’’

One of the advertising agencies affected was Publicis Media, which was told by Facebook that the earlier metric probably overestimated the average time spent watching videos up to 80 percent, The Wall Street Journal reported, citing a letter sent to clients.

In a statement on Friday, a Publicis Media spokesman said: “We take the job of being stewards of our clients’ investments very seriously. As an industry, we need to further push for more third-party measurement.”

Omnicom, another major advertising agency, had a similar response, saying: “This is why we and many others have been repeatedly saying that third-party verification is a fundamental requirement. Publishers should not ‘grade their own homework.’”