Fired by Bot at Amazon: ‘It’s You Against the Machine’
(Bloomberg) - Stephen Normandin spent nearly four years speeding through Phoenix as a contract driver for Amazon.com Inc., delivering packages. Then one day he received an automated email. The algorithms chasing him had decided he wasn't doing his job right.
The 63-year-old army veteran was stunned. He had been fired from a machine.
Normandin says Amazon punished him for things beyond his control that prevented him from completing his deliveries, such as locked apartment complexes. He said he took the dismissal harshly and, proud of a strong work ethic, recalled that during his military career he helped cook for 250,000 Vietnamese refugees in Fort Chaffee, Arkansas.
“I'm an old school guy and I give 110% to any job,” he said. “That really pissed me off because we're talking about my reputation. They say I didn't do the job, even though I know damn well I did. "
Normandin's experience is a twist on decades of predictions that robots will replace workers. At Amazon, machines are often boss - they hire, rate, and fire millions of people with little or no human supervision.
Amazon grew to become the world's largest online retailer, in part by outsourcing its extensive operations to algorithms - sets of computer instructions designed to solve specific problems. For years, the company has used algorithms to manage the millions of third-party sellers on its online marketplace and to file complaints that sellers have been booted after being falsely accused of selling counterfeit goods and raising prices.
Increasingly, the company is also outsourcing its human resources work to machines and using software not only to manage employees in its warehouses, but also to monitor contract drivers, independent delivery companies, and even the performance of its office workers. Those familiar with the strategy say Chief Executive Officer Jeff Bezos believes that machines make decisions faster and more accurately than humans, which reduces costs and gives Amazon a competitive advantage.
Amazon launched its gig-style flex delivery service in 2015, and the army of contract drivers quickly became a crucial part of the company's delivery machine. Typically, Flex drivers process packages that have not been loaded into an Amazon delivery van before the driver leaves. Instead of keeping the customer waiting, Flex drivers ensure that the parcels are delivered on the same day. They also handle large numbers of same-day grocery deliveries from Amazon's Whole Foods Market chain. Flex drivers helped keep Amazon buzzing during the pandemic and were only too happy to make about $ 25 an hour when their Uber and Lyft gigs dried up.
But once they log in, Flex riders discover that algorithms monitor their every move. Did you come to the delivery station when you said you would? Did you complete your route in the prescribed window? Did you leave a package on the porch from the pirates instead of being hidden behind a planter as you wanted? Amazon algorithms scan the torrent of incoming data for performance patterns and decide which drivers get more routes and which are deactivated. Human feedback is rare. Drivers occasionally receive automated emails, but most of the time they are obsessed with their ratings, which fall into four categories: Fantastic, Great, Fair, or Endangered.
Bloomberg surveyed 15 Flex drivers, including four who say they were unfairly terminated, as well as former Amazon executives who say the largely automated system is not adequately adjusted to the daily challenges drivers face in the field. Amazon knew delegating work on machines would lead to bugs and harmful headlines, said these former managers, but decided that it was cheaper to trust the algorithms than to pay people to investigate erroneous layoffs while the drivers were running could easily be exchanged.
So far, Amazon has had no problems finding Flex contractors. According to AppAnnie, around 4 million drivers worldwide have downloaded the app, 2.9 million of them in the United States. And more than 660,000 people in the US downloaded it in the first five months of this year, 21% more than the same period last year, according to SensorTower, another app tracker.
Within Amazon, the Flex program is considered a huge success, the benefits of which far outweigh the collateral damage, said a former engineer who helped develop the system. "The executives knew this was going to bed," said this person. "That's actually how they present it in meetings. The only question was how much feces we wanted there."
In a statement, Amazon spokeswoman Kate Kudrna called drivers' allegations of bad treatment and unfair dismissal anecdotal, saying they did not reflect the experience of the vast majority of Flex drivers. "We have invested heavily in technology and resources to provide drivers with insight into their status and eligibility for further performance, and to investigate any objections raised by drivers," she said.
As independent contractors, Flex drivers have few options if they believe they have been wrongly disabled. There is no paid administrative leave during an appeal. Motorists can pay $ 200 to take their dispute to arbitration, but few do as they see it as a waste of time and money.
When Ryan Cope was disabled in 2019, he didn't bother to argue or pay for arbitration. At this point, Cope had already decided that he could not meet the requirements of the algorithms. He drove miles on winding dirt roads outside of Denver in the snow, often shaking his head in disbelief that Amazon expected the customer to receive the package within two hours.
"When there is a problem, there is no support," said Cope, who is 29 years old. "You're against the machine, so you don't even try."
When drivers question bad reviews, they can't tell if they're communicating with real people. Answers often include only a first name or no name at all, and the answers tend to relate to a variety of situations rather than a specific problem. Even with a name attached, a machine was most likely generating the first few email responses, according to those familiar with the matter.
When human managers get involved, they usually do a hasty review - if they do one at all - because they have to meet their own standards of performance. A former driver assistance call center employee said dozens of low-skilled part-time seasonal workers have been tasked with overseeing the problems of millions of drivers.
“Amazon doesn't care,” said the former Amazon employee. "They know that most people get their packages and the 2 or 3 percent who don't get something at some point."
Amazon has automated its HR work more than most companies. But the use of algorithms to make decisions that affect people's lives is becoming more common. Machines can approve loan applications and even decide whether someone should deserve parole or stay behind bars. Computer science professionals are calling for regulations that force companies to be transparent about how algorithms affect people and give them the information they need to view and correct errors. Legislators have looked into the matter but have been slow to enact rules to prevent harm. In December, Senator Chris Coons, Democrat of Delaware, introduced the Algorithmic Fairness Act. To do this, the Federal Trade Commission would have to create rules that ensure that algorithms are used fairly and that those affected by their decisions are informed and have the opportunity to reverse mistakes. So far, his proposal has gone nowhere.
Neddra Lira of Arlington, Texas began shipping on the Amazon Flex app in 2017. The 42-year-old school bus driver and mother of three took a part-time job during the holidays and summer to earn extra money. She used to pay for her daughter's gymnastics classes. When the pandemic broke out and schools closed, Lira turned to Flex as her main source of income, delivering packages and Whole Foods groceries. She liked the flexibility and the ability to pocket about $ 80 for a four-hour route after pulling the gas on her Chevrolet Trax crossover.
Lira estimates she delivered approximately 8,000 packages and had an "excellent" performance rating for most of the time. Amazon algorithms evaluate drivers on the basis of their reliability and delivery quality, usually measured by whether they arrived on time to pick up parcels, whether they made the deliveries within the expected time frame and followed the customers' special requests. Flex metrics mainly focus on being on time, as opposed to ride hailing services like Uber and Lyft, which also prioritize things like the cleanliness of a car or the courtesy of the driver. Plus, Uber and Lyft passengers know when they're stuck in traffic, making drivers less likely to be penalized for circumstances beyond their control.
An Amazon customer has no idea what obstacles Flex drivers encounter on the way to their place of residence, nor do they know the algorithms that clock them. Lira says that sometimes there were so many drivers queuing outside the delivery station that she waited up to an hour to collect her packages, which put her behind schedule before she even started her route. When she discovered a nail in her tire, Amazon did not offer to pick up the packages, but asked them to return them to the delivery station. Lira was afraid the tire would go flat, but gave in to protect her position. Despite the explanation of the situation, her rating dropped from “great” to “compromised” for abandoning the route and it took several weeks for her to recover.
An email address was provided to Lira and she was asked to object to the termination within 10 days. She did so and asked why she was disabled so she could tell Flex Driver Support what went wrong. She never got any further details. She followed Oct 18, stating that she was a single mother who was fired from her regular job due to the pandemic and that Flex was the only thing keeping her afloat. Lira received what appeared to be an automated response from The Amazon Flex Team, apologizing for the delay and assuring her that her situation would be investigated by the responsible team.
Three days later, on October 21, she received a message from "Margaret" saying, "We are still investigating your complaint." Then, a week later, on October 28, an E signed "SYAM" read -Mail: "We have checked your information and looked at your history again. Our position has not changed and we will not restore your access to the Amazon Flex program ... We wish you every success in your future endeavors. "
Without appearing as a driver, Lira began struggling financially. She stopped paying her mortgage and her car was repossessed two days after Christmas with gifts donated for her children. Lira was forced to take a government flyer to pay her electricity, gas, and water bills. Eventually she started driving the school bus again and used most of a pandemic stimulus check to get her car back, paying $ 2,800 in missed payments, repurchase and storage fees.
"It just wasn't fair," said Lira. "I almost lost my house."
The computer engineers who designed Flex worked hard to keep it fair, taking into account variables like congestion and issues with accessing apartments that the system cannot detect, former employees said. But no algorithm is perfect, and at Amazon's size, even a small margin of error can internally be considered a big success and still cause a lot of pain to drivers. According to someone familiar with the program, Amazon Flex drivers deliver about 95% of all packages on time with no issues. Algorithms examine the remaining 5% for problematic patterns.
The Flex algorithms started out as blunt instruments and have been refined over time. According to a person familiar with the situation, early on the designers set too tight a time for the drivers to get to the delivery station. They hadn't considered human nature. Work eager drivers would promise to arrive by a certain time if they are too far away to make it. The bug caused good drivers to fail, the person said and was only fixed after a widespread dip in ratings. The system also uses GPS to decide how long it should take to get to a given address, but sometimes fails to take into account that navigating a country road in the snow takes much longer than crossing a suburban road on a sunny day.
The system worked well for Normandin for years. The Arizona native, who used to deliver pizza at night and newspapers in the morning, knew all the shortcuts and traffic bottlenecks. He also drove for Uber and Lyft, but did more flex work during the pandemic when the demand for trips fell and it became riskier to carry passengers than cart packages.
Normandin enjoyed excellent reviews and was even asked if he would like to train other drivers. He had a sophisticated system: sorting parcels before leaving the station, putting his first deliveries in the front seat, putting the next parcels in the back and putting the last load deep in the back of his 2002 Toyota Corolla. Normandin has been medically disabled for more than a decade due to a stomach and back condition that prevents him from sitting or standing in one place for long periods of time. He liked gig work because he could work a couple of hours straight.
Then, starting last August, Normandin had a number of setbacks that he claims are beyond his control. Amazon assigned him some pre-dawn shipments in apartment complexes when their gates were still locked, a common complaint among Flex drivers. In such cases, the algorithm instructs the drivers to deliver parcels to the control center, but it was not open either. Normandin called the customer as instructed - a long way to go as most people don't answer calls from unknown numbers, especially early in the morning. He called the driver support who could not reach the customer either. The clock was ticking now and the algorithm took notice.
“There are a lot of things the algorithms don't take into account, and the right hand doesn't know what the left is doing,” Normandin said.
Around the same time, he was asked to deliver packages to an Amazon locker in an apartment complex but was unable to open it. After 30 minutes on the phone with support, he was told to bring the packages back to the delivery station. Then his rating crashed. Normandin called support again to explain that a broken locker was responsible and said they had been told the problem will be fixed. "They never fixed it," he said, "and it took six weeks for my rating to go back up."
On October 2, Normandin woke up at 3 a.m., showered, and grabbed his phone to find a flex route, but couldn't log in. He checked his email and found a generic message from Amazon signed by "Gangardhar M". It was said that Normandin's reputation had "fallen below acceptable levels" and that he was being fired.
Then began a process familiar to anyone caught in an automated customer service loop - except in this case, Normandin did not request a refund for a damaged product. He was struggling to get his job back.
Normandin offered Amazon the usual 10 days to appeal, sent an email to Flex support and asked for his cancellation to be reversed. He stated that he had already informed Amazon of circumstances beyond its control and was promised that the violations would not be brought against him
Normandin received a reply from “Pavani G” the next day, thanking him for “using Amazon Flex to provide more context to your story”. Normandin replied to this email with additional information and received the exact same reply that promised to address the issue, but this time it was signed by "Bitan Banerjee". The email promised to respond within six days. Seven days later, “Arnab” apologized for the delay via email and promised an update as soon as possible.
Meanwhile, Normandin was making no money. He was counting on Amazon's annual Prime Day sale, which had been postponed until October to earn the money he needed to pay bills. With no response until October 19, Normandin wrote another message to Amazon, this time copying Bezos.
"I ask for specific details on how this decision to deactivate my account came about," he wrote. "I am confident that after a thorough review of my entire delivery history as an Amazon Flex driver, I will have a consistent history of performance at the highest level, from a sensible and prudent person."
About 12 hours later, he received a reply informing him that Bezos had received the email and instructed "Taylor F" to investigate the problem and respond on his behalf. On October 23, Normandin received an email from “Raquel” on the Amazon Flex support team informing him that his appeal was still being considered. Former Amazon employees who worked on Flex said escalating to Bezos is a common tactic among disabled drivers, but it rarely helps them.
The verdict came on October 28th from "SYAM", the same name in the last message to Lira. The email did not respond directly to Normandin's requests, but acknowledged the challenges of the contract and said, "We understand that every delivery partner has difficult days and that delays can sometimes occur and we have already taken this into account." But Normandin still didn't get his performance back.
After the shock subsided, he tried a few other delivery services but instead decided to use his pandemic stimulus money to start a small engine repair business. It was time to deal directly with people again. Of the people who developed the algorithms that tracked, rated, and eventually fired him, Normandin said, "You don't seem to have any common sense about how the real world works."
More stories like this can be found on Bloomberg.com
Subscribe now to stay one step ahead with the most trusted business news source
© 2021 Bloomberg L.P.
In this article:
You should check here to buy the best price guaranteed products.
A Waffle House waitress who worked a double shift while looking after her baby says she received a $1,000 tip from a country music star who was impressed by her work ethic
ShowBiz Minute: Wu, DaBaby, US Box Office
Vickers Venture Partners On China’s Regulatory Crackdown
Team USA survives battle with France, keeps Olympic win streak alive
Federal ban on evictions expires
‘Grandmas buying shotguns’: US dealers see ammunition shortage as sales surge