How to not get screwed on Amazon

Three months ago on Amazon, a pair of two-toned Wolverine oxfords sold for $160. Last month, I snagged them for a mere $12.54, even though they still retailed for $70-plus on competing discount sites. Had I waited until last weekend to buy them, I could have saved another $2. But if I had waited any longer, they would have gone back up, to $28.06. In suede, my new shoes’ price fluctuations were even more erratic, wavering up and down by more than 20% every few days.

You may not not realize it, but shopping on Amazon is more like playing the stock market than shopping at your local big box store. Prices don’t just change with the season or when something goes on sale. They fluctuate as often as every 15 minutes.

That’s because Amazon and the site’s third-party sellers are engaged in an algorithmically enabled pricing war. Amazon has long determined prices using lots of fancy math. But an increasing number of the more than 2 million third-party merchants in the Amazon Marketplace also rely on algorithms to take stock of supply, demand, and prices offered by their competition, then adjust their own prices, in real time. That can mean steals like brand new iPads for $250 or getting screwed when you buy a Vacation Getaways Lego Creator building kit for $99 instead of $55, because you weren’t savvy to the constant fluctuations.

Prices seem to vary most on trendy, big-ticket items. Take the DJI Phantom 4 Quadcopter drone. It’s selling at Best Buy right now for $1,200, but in Amazon’s Marketplace, it fluctuates between $2,100 and $699. On Wednesday, it started off selling at $1,065.61, but by the afternoon was down to $999. On Amazon, waiting just a few hours to purchase something can often mean serious cost savings.

Is this all reminding you of the last time you planned a vacation? It should. Thanks to these pricing algorithms, buying products on Amazon is starting to involve the same planning, price monitoring and exquisite timing required to get a good deal on a flight.

I recently installed Keepa, a Chrome extension that tracks Amazon pricing history, to dig through recent purchases in my Amazon order history. I found out I’ve overpaid for an awful lot of things over the past year. Last time I purchased shampoo and conditioner, I shelled out a whopping $46.93 for Pureology’s Reviving Red Shampoo & Conditioner. (Don’t judge! Christina Hendricks swears by it.) Little did I know that my already over-priced hair gunk was trading at a high. Had I bought a week earlier, I could have saved $12.93. If I’d been especially prescient, I would have stocked up last November, when for one brief week the price dropped down to just $29.90, a price difference of more than 30%.

I bought contact solution for $21 in September when right now it’s selling for $17. Tupperware I bought for $37 is now $26. Just a few days after I bought new mixing bowls for $34.88, the price dropped to $28.13.

I’ve always thought of myself as a savvy shopper—I monitor airfare prices obsessively for weeks before buying and often pick my vacation locations based on wherever has the best deal on flights. But it turned out that when I scoured Amazon for the best deals on the internet, I was missing a crucial piece of information. For me, a good deal is a trophy, a source of pride. Realizing how much I could have saved with a little more attention to timing (and some patience) wasn’t just a bummer, it was a crisis of faith. My approach to online shopping was all wrong.

If you’re a third-party seller in Amazon’s Marketplace, there are at least a half-dozen options for pricing software to give your business a competitive edge. It works like this: the software scrapes data from Amazon’s API to figure out how many of a certain item exists on Amazon, what Amazon and other merchants are selling it for and how many people seem to be interested in buying. Sellers can set parameters, like how low they are willing to go on an item, or how often they’d like to change the price. Based on all of those factors, the algorithm works to constantly evaluate the competitiveness of an item in the Marketplace, and automatically reprices it accordingly, sometimes as often as every 15 minutes.

Sellers that use algorithmic repricing software do so, mostly, for a better shot at winning what’s known at the “Buy Box.” The Buy Box is the button you hit to add an item to your cart. On a listing for an iPad, say, several sellers compete to be the seller feature in this prime Amazon real estate.

A recent ProPublica investigation found that the competition to get into this box is fierce, and that it often privileges items sold by either Amazon or sellers who opt to pay Amazon to fulfill their orders over other third-party sellers. This means that for many sellers, repricing software is an increasingly important part of their Amazon strategy.

In December 2014, two weeks before Christmas, a glitch in a popular repricing software used by Amazon Marketplace sellers accidentally dropped the price of thousands of Amazon items to as little as one cent. It was an Amazon Marketplace flash crash, creating panic among sellers who worried the error would mean deep losses on goods an algorithm had autonomously decided to give away practically for free. In 2011, dueling runaway algorithms similarly caused the price of an obscure developmental biology book to be listed on Amazon for $23,698,655.93 (plus shipping). These stories grabbed headlines, demonstrating acutely the consequences when behind-the-scenes decision-making software goes awry.

But extreme pricing volatility is the upshot of such software even when it’s working as it should. In the 1990s, airlines began constantly changing prices based on competition and the number of seats left on a flight, creating an entire micro-industry devoted to reading the algorithmic airfare tea leaves. Now the prices of shampoo and conditioner are beginning to be regulated with the same degree of complexity.

“At some point more people will catch on and people are going to fundamentally change how they view online shopping,” Christo Wilson, a computer scientist at Northeastern University told me. “It just doesn’t obey the same rules as shopping in the real world.”

Earlier this year, Wilson and two colleagues published a study that looked at the prevalence and consequences of algorithmic repricing software in the Amazon Marketplace. Looking at four months of data for merchants selling more than 1,500 best-selling products, they were able to uncover algorithmic pricing strategies adopted by over 500 sellers. Conservatively, they estimated that about 2.4% of sellers use such software, accounting for 60-70% of items offered in Amazon’s Marketplace. But Wilson told me he thinks the number of sellers using repricing software numbers is more like 10%, accounting for 90% of merchandise. And it’s growing.

Algorithmic sellers, they found, tend to be more successful, offering fewer products but seemingly doing a much higher sales volume and “winning” the Buy Box more frequently. Items with algorithmic sellers, they found, were also much more volatile. Prices on an item could change thousands of times. Most of the time, those changes were small, a few dollars or less. But sometimes they would jump dramatically in just 20 minutes, the price changing by tens or even hundreds of dollars.

“When we shop, we operate under the assumption of an open marketplace and that’s not true here. It’s unlike other marketplaces,” Wilson told me. “Shopping for a toothbrush is going to be a lot more like shopping for airline tickets or hotels. It’s not a fun future. I, for one, certainly don’t like shopping for airline tickets.”

Jim Cockrum, a high-volume Amazon seller in Indianapolis who also consults with other sellers on how to best run their business, told me that for medium or big-sized sellers, repricing software has become an essential strategy.

He doesn’t view the economics of the Amazon Marketplace as all that different from those governing my local Wal-Mart. After all, retailers like Wal-Mart have long guaranteed to match prices, and prices vary online and off, and even from store to store. Pricing based on supply and demand isn’t an Amazon invention. It’s just basic free-market economics.

“You might buy something at a store and you go back a few days later and it’s half off,” he said.

The catch, he told me, is that on Amazon, everything is sped up.

“Online, you have real-time consumer demand and pricing data,” he said. “Every item on Amazon is kind of a commodity like you might see on the stock market. You don’t know if the price of your inventory is going to go up or down.”

The marketplace volatility, Cockrum told me, also creates opportunity. Often he’ll see an Amazon seller selling something for a steal, so he’ll buy it and then turn around and sell it himself for more money.

“It’s just supply and demand,” he told me. “The public determines the price, not me.”

Cockrum thinks that most of his customers don’t turn to Amazon for the best deal anyway—they’re looking instead for convenience and quick delivery.

“Personally, I don’t spend a lot of time nickle and diming prices,” he told me. “We live in a Prime Now area, and I am willing to pay a premium to have something be on my porch a few hours from now.”

Jean-Charles Compagnon, a former eBay seller who created the repricing software ki Magic, told me that when there are sudden, unbelievable dips in price (like, say, a $10 iPad), those are typically the result of two sellers who don’t know what they’re doing whose algorithms get into a pricing battle.

“Someone can set their software to always be 10 cents lower than the other person. If two people are doing the same thing then that’s when you have the race to the bottom,” he said.

That, he said, can be damaging for sellers.

“You have a good margin and then someone else is too greedy and the price goes down to a point where people aren’t making any money anymore,” he said.

He told me that more experienced sellers don’t aim to necessarily be the cheapest in order to be the most competitive. But in his study, Christo Wilson found that 70% of algorithmic sellers set their price within $1 of the lowest price. This, in part, is what makes for such drastic volatility.

The huge fluctuations in price on Amazon are a sort of open secret. Amazon, by both making a version of repricing software available to sellers and by opening up its API for pricing data to be scraped, encourages algorithmic repricing by Marketplace sellers. From the company’s point of view, increased competition means better deals for the consumer.

“Our customers expect to come to Amazon and find the lowest prices,” a spokesperson told me in a statement. “We do the hard work for them, finding the best prices out there and meeting or beating them for all customers, every day, across our entire selection.”

But Wilson’s concern is that the amount of turbulence in Amazon pricing is still largely unrecognized by the consumer. He worries that as competition drives more sellers to seek algorithmic assistance the instability of the market will only increase.

Adam Wierman, a CalTech professor who studies network economics, told me that for now most basic goods on Amazon don’t seem to vary much more in price than they would at the grocery store. But as the algorithms get better at gaming the system to win the Buy Box, it could lead to significant pricing variability on everything from iPads to toothpaste.

There is an upside to all this: While most Amazon operations are opaque, its pricing data is not. You can download a pricing tracker like Keepa or CamelCamelCamel and play the Amazon market, instead of letting it play you.