Incident 220: Facebook Mistakenly Blocked Small Business Ads

Description: Facebook’s AI mistakenly blocked advertisements by small and struggling businesses, after the company allegedly leaned more on algorithms to monitor ads on the platform with little review from human moderators.
Alleged: Facebook developed and deployed an AI system, which harmed small businesses on Facebook.

Suggested citation format

Giallella, Thomas. (2020-11-11) Incident Number 220. in Lam, K. (ed.) Artificial Intelligence Incident Database. Responsible AI Collaborative.

Incident Stats

Incident ID
220
Report Count
4
Incident Date
2020-11-11
Editors
Khoa Lam

Tools

New ReportNew ReportDiscoverDiscover

Incidents Reports

In mid-March, even as it encouraged its full-time employees to work remotely, Facebook's largely contract-based content moderators were still required to be in the office. Amid public pressure, Facebook eventually sent them home as well, saying it would rely more heavily on artificial intelligence to police its platform in the interim.

Just days later, Facebook mistakenly blocked users from posting legitimate news articles about the coronavirus pandemic.

Facebook said the issue stemmed from a technical glitch in its automated spam filters, raising concerns over whether the company's AI was up to the task of accurately and quickly sorting through the incoming tsunami of coronavirus hoaxes, scam medical products, hate speech, and election-related misinformation in the months ahead. On a call with reporters the next day, CEO Mark Zuckerberg addressed Facebook's pivot, hyping up the capabilities of its AI but also acknowledging that the company anticipated "some false positives" as a result of the new approach.

In the months since then, hundreds of advertisers claim they've gotten caught up in Facebook's AI dragnet despite not violating any policies, according to messages and screenshots viewed by Business Insider.

Because Facebook doesn't tell advertisers which specific policy they violated when it disables their accounts, it's difficult to know what proportion of those bans were actually the result of an error by Facebook — or how many additional errors have gone unreported. But in interviews with Business Insider, seven business owners and ad agencies said they've seen an uptick in bans that Facebook ultimately admitted were made in error.

"We know it can be frustrating to experience any type of business disruption, especially at such a critical time of the year. While we offer free support for all businesses, we regularly work to improve our tools and systems, and to make the support we offer easier to use and access. We apologize for any inconvenience recent disruptions may have caused," a Facebook spokesperson told Business Insider.

But those advertisers — all of whom run small or medium businesses, or run ads on behalf of them — told Business Insider that the slow, opaque, and inconsistent customer service Facebook provides to smaller advertisers has left them locked out of their accounts, sometimes for weeks or even months at a time, often costing them tens of thousands of dollars in revenue as they try to get erroneous bans reversed.

For those businesses, which often rely heavily on Facebook to reach customers, mistakes made by Facebook's AI are leaving them cut off from a key revenue source at the worse possible time: Black Friday, Cyber Monday, and the start of the holiday shopping season.

Facebook's errant AI police

Since March, there have been almost weekly reports of Facebook's "false negatives" — times it either failed to detect or refused to enforce what appeared to be clear policy violations, from calls for violence by President Donald Trump and white supremacists in Kenosha, Wisconsin, to conspiracies about COVID-19.

At the same time, Facebook kicked its AI into overdrive. The company reported that it removed 112 million organic posts in the first nine months of 2020, up more than 35% from the same period in 2019 (a spokesperson said there is no comparable report for ads). That's led to additional negative attention in recent months for wrongful action against users and advertisers that didn't violate its policies, the so-called "false positives" Zuckerberg warned about.

Days before the November 3 general election, a tech glitch "improperly" blocked political ads, affecting the campaigns of both Trump and then-Democratic presidential nominee Joe Biden. Barely a week later, another tech glitch caused major issues for non-political advertisers as well, resulting in their ads not getting approved and accounts being disabled.

Some advertisers have had their accounts deactivated for weeks at a time, while others have had accounts disabled multiple times this year. And a majority said that Facebook typically provides little information about why it disabled their accounts while taking weeks to review their appeals, before ultimately admitting it banned them in error and restoring the accounts. Even in some cases where Facebook took action against specific ads, screenshots and messages viewed by Business Insider showed that the ads appeared to have nothing to do with the policy they were flagged for ostensibly violating.

One advertiser, who runs a dog products business that predominantly relies on Facebook to reach customers and requested anonymity out of fear for retaliation from Facebook, told Business Insider that his ad for a dog poster was slapped with a warning label after an independent fact checker determined it was in violation of Facebook's policy against false news.

He told Business Insider he managed to contact the fact checker, who acknowledged the error, and the ad was restored temporarily. But then, Facebook's algorithm subsequently blocked the ad again. Since advertisers can max out the number of appeals they file, he gave up on the ad to avoid broader repercussions for his account (Facebook said it doesn't penalize accounts if it's aware it made the error).

In one Facebook group, advertisers claimed Facebook's AI has also mistaken necklaces for adult toys and arcades for gambling content. In October, Facebook blocked an ad for onions after incorrectly determining that the ad violated its policy against nudity.

But in Facebook groups and in interviews with Business Insider, advertisers' complaints suggest that these erroneous bans are far more widespread than the one-off tech glitches Facebook has acknowledged after outcry from advertisers, politicians, or news media outlets.

On hold with customer support

A Facebook spokesperson told Business Insider that both the company's AI and human reviewers make mistakes, and that advertisers can appeal decisions.

But those appeals appear to be moving at a significantly slower pace during the pandemic. Facebook reported that appeals concerning organic posts had all but vanished — down 95% in the first nine months of 2020 compared to the same time period last year — adding that "due to a temporary reduction in our review capacity as a result of COVID-19, we could not always offer our users the option to appeal."

Facebook refused to disclose how many appeals it fielded concerning ads, but advertisers say they're encountering a similar bottleneck.

Simon Wagner, who runs an online women's jewelry business, has had his account mistakenly disabled by Facebook six times this year — in each case Facebook reversed its decision after he appealed, but in some cases that review process took more than a month. In the meantime, Wagner estimated that he has lost around $35,000 in revenue as a result of being unable run ads during those periods.

Wagner was among dozens of advertisers who reported waiting a month or more for Facebook to review an appeal, even though a customer support representative told him via chat that the process usually takes 24 to 48 hours, according to messages viewed by Business Insider.

All seven of the advertisers who spoke to Business Insider said this was because, as smaller spenders, they felt they (or their clients) received second-class customer support.

Larger advertisers who, according to AdAge, spend a minimum of $10,000 per month with Facebook, are eligible to receive a dedicated account representative to help them navigate issues, such as erroneous bans. Those who don't — which Facebook has increasingly touted as it faces growing antitrust scrutiny — only have access to chat support with a randomly assigned customer service agent.

"If you're not spending millions and don't have a rep or if you're not working with an agency that has access to a rep, you're basically going to be treated unfairly, and they don't give a s---," said the owner of the dog products business.

Agencies also in the dark

Many businesses enlist ad agencies to help them create, target, and run ads on Facebook, and when those ads get removed by Facebook, they often ask the agency what went wrong. But even people who run those agencies say they have no more access to information than their clients.

"You don't really have anything to get back to your client with, and I think it really kind of fundamentally starts to fracture some of the core tenets of that agency-client relationship," Eric Allred, CEO of El Dorado Digital, told Business Insider.

"With COVID and shipping times, and it just being a kind of a weird year ... it's extra challenging to have your main point of distribution turned off without really a satisfactory explanation," he said.

"They're just getting stricter and stricter and are banning accounts more often and providing no explanation, no real appeal process, no proper customer support," Jay Topp, the founder of the Australia-based agency Lion Social, told Business Insider.

"How are we supposed to build a business on a platform that doesn't give a f--- about us, that can just take everything in an instant without explanation, no accountability," he said, adding that his clients spend around $1 million annually on Facebook.

No overnight changes

Justin Brooke, the founder of Ad Skills, a provider of digital marketing courses with more than 11,000 members, told Business Insider that Facebook's apparent spike in account bans and decrease in customer support quality resembles similar growing pains he went through with Google nearly a decade ago, when it banned 214,000 accounts in 2015 in an attempt to purge "bad ads" from its rapidly growing platform.

"Investors don't like to hear that 214,000 advertisers are no longer spending money, and so Google has gotten way better over the years — they've got phone numbers, we can call great reps, we can email ... love working with Google today," he said.

But Brooke said those changes didn't happen overnight — it took until around 2018 for his experience with Google to become more positive. Similarly, he doesn't expect Facebook to act until enough advertisers leave or reduce spending on the platform to put a dent in the company's bottom line.

That could be awhile, given how Facebook has responded to past rifts with advertisers, and the fact that it has 10 million advertisers, meaning even a few hundred shifting some ad spend to Google or up-and-coming platforms like TikTok likely won't undercut sales (Facebook reported $21.5 billion in revenue last quarter, up 22% year-over-year).

In July, Zuckerberg told employees that Facebook was "not gonna change" its hate speech policies even after 500 advertisers — including major brands like Coca-Cola, Starbucks, Unilever, Verizon, Ford, and Ben & Jerry's — temporarily boycotted the platform. Many of those companies have since resumed advertising on Facebook, despite a lack of significant policy changes.

Facebook said it has long used AI (successfully, by its evaluation) and continually makes changes meant to improve its ad products and services, such as adding several thousand reviewers over the past year.

But even if Facebook never sufficiently addressed their concerns, smaller advertisers are typically even more dependent on the largest social-media platform on the internet, and more reluctant to shift their spending to other platforms than major brands.

"I can't really blame [my clients] for sitting there and being like: 'Okay, well we make, let's say, $100,000 on Facebook, but we make $25,000 on Snapchat or something, let's move all this spend over to Snapchat," Ameet Khabra, founder of Canada-based agency Hop Skip Media, told Business Insider.

Others echoed those concerns.

"[Facebook] is my main way of advertising and this is just so risky. Like it's a really, really risky platform to rely on, but it's also the most powerful, right? So it's kind of a love-hate relationship that we all have with Facebook," the dog products business owner said.

Still, Brooke said he has been increasingly urging his customers to switch to Google as more reports of glitches with Facebook emerge.

"Our Google courses are just the highest demand we have right now," Brooke said. "YouTube ads are just on fire right now."

Facebook's AI-fueled attempt to block bad ads is hurting legitimate small business owners — and its 'pay-to-play' customer support is leaving them stranded ahead of the holiday shopping season

Facebook Inc. recently issued a statement apologizing to small businesses for mistakenly blocking their ads. In a bid to curb ‘infodemics’ or misinformation and keep a check on policy violators, the company has appointed Artificial Intelligence (AI) bots. However, earlier this month, the bots malfunctioned, abruptly blocking advertisements from small and struggling businesses across the world.

How did it begin?

The issue first surfaced on November 11, when the social networking platform blocked an advertisement of HoneyGramz, a New York-based business who was struggling to make meets end owing to the COVID-19 crisis. Facebook, which had earlier cited police violations, restored her account in a few days, however, the company had already suffered a loss of $5000 until then. Another business was hit the same weekend in Ottawa with Facebook citing the same reasons.

Why did it happen?

According to a Bloomberg report, Facebook’s human moderators have focused on election and Covid-19 misinformation in 2020, enabling the firm to learn more about artificial intelligence algorithms used to monitor other areas of the platform. This has prompted more and more small businesses to get caught in Facebook’s automated filters and get their advertisements disabled.

What did Facebook say?

Later, the California based company issued a statement apologizing for the business disruptions. In a statement, the Mark Zuckerberg led firm said that the company regularly works on improving its tools and systems and that the malfunction was a result of it.

"We know it can be frustrating to experience any type of business disruption, especially at such a critical time of the year. While we offer free support for all businesses, we regularly work to improve our tools and systems, and to make the support we offer easier to use and access. We apologize for any inconvenience recent disruptions may have caused,” Facebook said in a statement.

Facebook’s AI mistakenly blocks ads of struggling businesses; tech giant issues apology

New York-based businesswoman Ruth Harrigan usually sells her honey and beeswax products in souvenir shops. But with COVID-19 pausing tourism, she’s been almost entirely dependent on Facebook ads to drive online sales.

On Nov. 11, this new financial lifeline was abruptly cut when the social media company blocked her HoneyGramz ad account for violating its policies. She couldn’t imagine what about her tiny, honey-filled gifts would have triggered the problem.

Friends told Ms. Harrigan to just wait a couple of days and the problem might resolve itself. She waited, until she lost an estimated $5,000 in revenue.

“I was getting a little anxious thinking, ‘Oh my God, Black Friday is around the corner, most of my sales for the year happen in November and December and that’s it,‘” she explained. “I said, ‘If I’m shut down any longer than this, it’ll cripple me.‘”

Ms. Harrigan is one of millions of small-business advertisers who have come to rely on Facebook because COVID-19 has shut down many traditional retail channels. The social media giant has provided new sales opportunities for these entrepreneurs, but also exposed them to the company’s misfiring content-moderation software, limited options for customer support and lack of transparency about how to fix problems.

Facebook’s human moderators have focused on election and COVID-19 misinformation this year, so the company has leaned more on artificial intelligence algorithms to monitor other areas of the platform. That’s left many small businesses caught in Facebook’s automated filters, unable to advertise through the service and frustrated because they don’t know why.

The same weekend Ms. Harrigan’s account went down, Ivonne Sanchez, who runs a permanent makeup clinic in Ottawa, found her ads were blocked too, for what Facebook said was a “policy violation.” Her business, which had to shut down between March and June for the pandemic, was relying on Facebook to recover financially. The account was restored the next day without explanation, but “in the middle of a crucial shopping season, it left us shaken,” she said.

“This experience makes us very nervous about investing dollars into a system that is operated seemingly by a bot.”

Even if an ad account gets restored, businesses lose crucial momentum. Facebook’s advertising algorithm takes a couple weeks to figure out which users may be interested in an ad, to refine the targeting. Jessica Grossman, chief executive officer of digital marketing firm In Social, said when her clients get hit, the hardest part is telling them their campaigns have to start over and their money won’t go as far.

“Facebook almost doesn’t realize the impact of their own algorithm and what that means,” Ms. Grossman said. There seemed to be no logic to the account bans imposed on In Social’s clients, she added. A pizza vending machine company, a reusable water bottle company, a coffee delivery service, a business coach and a hair weave company were all suspended.

“We know it can be frustrating to experience any type of business disruption, especially at such a critical time of the year,” Facebook said in a statement. “While we offer free support for all businesses, we regularly work to improve our tools and systems, and to make the support we offer easier to use and access. We apologize for any inconvenience recent disruptions may have caused.”

Facebook often touts its commitment to small businesses as it defends its ever-larger hold over their economic future. On a recent earnings call, CEO Mark Zuckerberg said this was a “major focus” that’s “more important now than ever” as COVID-19 shifts commerce online. During a July ad boycott by major brands, Facebook’s revenue still grew, bolstered by small businesses rushing online to try to survive. The company added more tools this year for small businesses to sell directly to customers through its site, hoping these virtual shops become advertisers, too.

But while business owners agree that Facebook is a lifeline during the pandemic, they say it’s also an unreliable partner.

Facebook’s ban on political ads around the U.S. election, for instance, affected companies that have no connection to politics, like a business selling bracelets to benefit refugees. A seed company was also blocked for sharing a picture of Walla Walla onions — which were “overtly sexual,” according to Facebook’s AI.

The company’s policies against cryptocurrency frequently trapped ads from a solar roof company, Human SOLR, because some of the acronyms used by the business are similar to cryptocurrency tokens. After that issue was resolved, Human SOLR’s ads were banned again for using phrases like “see if your roof qualifies.” Facebook’s software guessed the company was selling financial products, which are more regulated. After enough flags on the account, Brett Lee, who runs the business, gave up on Facebook ads. “My business is at a complete standstill,” said Mr. Lee, based in St. George, Utah. “My employees’ lives are at a standstill.”

In some cases, the business impact is hard to quantify. Matt Snow, co-founder of an apparel business called Boredwalk, said Facebook’s automated systems inadvertently flagged 40% of his company’s product catalog as unsafe late last month. That left Mr. Snow targeting the wrong products to potential customers. He eventually noticed and quickly resolved the issue with a Facebook sales manager, but Mr. Snow doesn’t know how long the products were banned, or even which other items were being advertised in their place.

Facebook has been automating content moderation for years, a transition it highlights in a quarterly report detailing how much content the company removes. In more nuanced categories such as “hate speech,” Facebook removed almost 95% of violating posts automatically in the third quarter, up from just 53% two years ago.

But that increase comes with more corrections. Facebook removed 22 million posts for hate speech in the third quarter, more than three times as many as a year earlier. The number of posts it later restored jumped by 40%.

Advertisers have been particularly hurt by these automated decisions in recent months. “It just exploded. They turned up the AI recently — somebody changed something — and all of a sudden everybody was getting shut down,” said Justin Brooke, founder of Adskills.com, which teaches businesses how to market on Facebook. “What are these small businesses going to do? They’ve got families to feed.”

One of Mr. Brooke’s own Facebook ads has a small written disclaimer saying it wasn’t open to those trying to sell adult content. That got flagged and taken down. Facebook’s automated explanation? The post didn’t follow the company’s community standards on “nudity/sexual activity.”

The overreaction by Facebook’s AI is a side effect of the company taking more responsibility for the content on its platform, according to Guy Rosen, Facebook’s vice president of integrity. “As we take more action, we remove more content, there’s more opportunities also for those to be in error,” he said during a recent press call.

That’s what HoneyGramz’s Ms. Harrigan was told happened to her account. She eventually got desperate enough to Google names of Facebook employees who might help. She found Rob Leathern, the company’s director of ad products, and sent him a message on Twitter. Miraculously, he responded. A few hours later, Facebook sent an email restoring her account.

“They just said they turned it off in error,” Ms. Harrigan said. “They didn’t give me any feedback. They just reset the whole thing as if it never happened.”

Facebook's AI mistakenly bans ads for struggling businesses

New York-based businesswoman Ruth Harrigan usually sells her honey and beeswax products in souvenir shops. But with Covid-19 pausing tourism, she’s been almost entirely dependent on Facebook ads to drive online sales. On Nov. 11, this new financial lifeline was abruptly cut when the social media company blocked her HoneyGramz ad account for violating its policies. She couldn’t imagine what about her tiny honey-filled gifts would have triggered the problem.

Friends told Harrigan to just wait a couple of days and the problem might resolve itself. She waited, until she lost an estimated $5,000 in revenue.

“I was getting a little anxious thinking, ‘Oh my God, Black Friday is around the corner, most of my sales for the year happen in November and December and that’s it,’” she explained. “I said, ‘If I’m shut down any longer than this, it’ll cripple me.’”

Harrigan is one of millions of small business advertisers who have come to rely on Facebook Inc. because the coronavirus has shut down many traditional retail channels. The social media giant has provided new sales opportunities for these entrepreneurs, but also exposed them to the company’s misfiring content-moderation software, limited options for customer support and lack of transparency about how to fix problems.

Facebook’s human moderators have focused on election and Covid-19 misinformation this year, so the company has leaned more on artificial intelligence algorithms to monitor other areas of the platform. That’s left many small businesses caught in Facebook’s automated filters, unable to advertise through the service and frustrated because they don’t know why.

The same weekend Harrigan’s account went down, Ivonne Sanchez, who runs a permanent makeup clinic in Ottawa, found her ads were blocked too, for what Facebook said was a “policy violation.” Her business, which had to shut down between March and June for the pandemic, was relying on Facebook to recover financially. The account was restored the next day without explanation, but “in the middle of a crucial shopping season, it left us shaken,” she said. “This experience makes us very nervous about investing dollars into a system that is operated seemingly by a bot.”

Even if an ad account gets restored, businesses lose crucial momentum. Facebook’s advertising algorithm takes a couple of weeks to figure out which users may be interested in an ad, to refine the targeting. Jessica Grossman, chief executive officer of digital marketing firm In Social, said when her clients get hit, the hardest part is telling them their campaigns have to start over and their money won’t go as far.

“Facebook almost doesn’t realize the impact of their own algorithm and what that means,” Grossman said. There seemed to be no logic to the account bans imposed on In Social’s clients, she added. A pizza vending machine company, a reusable water bottle company, a coffee delivery service, a business coach and a hair weave company were all suspended.

“We know it can be frustrating to experience any type of business disruption, especially at such a critical time of the year,” Facebook said in a statement. “While we offer free support for all businesses, we regularly work to improve our tools and systems, and to make the support we offer easier to use and access. We apologize for any inconvenience recent disruptions may have caused.”

Facebook often touts its commitment to small businesses, as it defends its ever larger hold over their economic future. On a recent earnings call, CEO Mark Zuckerberg said this was a “major focus” that’s “more important now than ever” as Covid-19 shifts commerce online. During a July ad boycott by major brands, Facebook’s revenue still grew, bolstered by small businesses rushing online to try to survive. The company added more tools this year for small businesses to sell directly to customers through its site, hoping these virtual shops become advertisers, too.

But while business owners agree that Facebook is a lifeline during the pandemic, they say it’s also an unreliable partner. Facebook’s ban on political ads around the U.S. election, for instance, affected companies that have no connection to politics, like a business selling bracelets to benefit refugees. A seed company was also blocked for sharing a picture of Walla Walla onions — which were “overtly sexual,” according to Facebook’s AI.

The company’s policies against cryptocurrency frequently trapped ads from a solar roof company, Human SOLR, because some of the acronyms used by the business are similar to cryptocurrency tokens. After that issue was resolved, Human SOLR’s ads were banned again for using phrases like “see if your roof qualifies.” Facebook’s software guessed the company was selling financial products, which are more regulated. After enough flags on the account, Brett Lee, who runs the business, gave up on Facebook ads. “My business is at a complete standstill,” said Lee, based in St. George, Utah. “My employees’ lives are at a standstill.”

GFP Delivered, a Chicago-based produce company advertising a way for people to avoid the grocery store during Covid-19, had its Facebook ads shut down for two months without clear explanation, according to owner George Fourkas. He said he was able to fix the problem only after reaching out to old college friends who work at Facebook.

Yaniv Gershom, co-founder of digital marketing firm 4AM Media, said he had to cut 12 jobs partly because of Facebook ad account bans, which have lasted almost six months. “They give you zero feedback,” he added. “The only people who are OK are massive spenders who get a Facebook rep that can escalate issues and find out what’s wrong.”

In some cases, the business impact is hard to quantify. Matt Snow, co-founder of an apparel business called Boredwalk, said Facebook’s automated systems inadvertently flagged 40% of his company’s product catalog as unsafe late last month. That left Snow targeting the wrong products to potential customers. He eventually noticed and quickly resolved the issue with a Facebook sales manager, but Snow doesn’t know how long the products were banned, or even which other items were being advertised in their place. “Facebook is very black box about all their internal machinations,” he said.

Facebook has been automating content moderation for years, a transition it highlights in a quarterly report detailing how much content the company removes. In more nuanced categories such as “hate speech,” Facebook removed almost 95% of violating posts automatically in the third quarter, up from just 53% two years ago.

But that increase comes with more corrections. Facebook removed 22 million posts for hate speech in the third quarter, more than 3 times as many as a year earlier. The number of posts it later restored jumped by 40%.

Appealing these often-automated decisions has also become a lot harder. “Due to a temporary reduction in our review capacity as a result of Covid-19, we could not always offer our users the option to appeal,” Facebook wrote in its third-quarter report.

Advertisers have been particularly hurt by these automated decisions in recent months. “It just exploded. They turned up the AI recently — somebody changed something — and all of the sudden everybody was getting shut down,” said Justin Brooke, founder of Adskills.com, which teaches businesses how to market on Facebook. “What are these small businesses going to do? They’ve got families to feed.”

One of Brooke’s own Facebook ads has a small written disclaimer saying it wasn’t open to those trying to sell adult content. That got flagged and taken down. Facebook’s automated explanation? The post didn’t follow the company’s community standards on “nudity/sexual activity.”

The over-reaction by Facebook’s AI is a side effect of the company taking more responsibility for the content on its platform, according to Guy Rosen, Facebook’s vice president of integrity. “As we take more action, we remove more content, there’s more opportunities also for those to be in error,” he said during a recent press call.

That’s what HoneyGramz’s Harrigan was told happened to her account. She eventually got desperate enough to Google names of Facebook employees who might help. She found Rob Leathern, the company’s director of ad products, and sent him a message on Twitter. Miraculously, he responded. A few hours later, Facebook sent an email restoring her account.

“They just said they turned it off in error,” Harrigan said. “They didn’t give me any feedback. They just reset the whole thing as if it never happened.”

But Harrigan won’t forget. She printed off the email and pinned it to her office whiteboard. “It was really, really scary,” she said.

Facebook’s AI is mistakenly banning some small business’ ads