Nothing describes the Amazon product reviews problem as well as Goodhart’s law, named after British economist Charles Goodhart.
“When a measure becomes a target, it ceases to be a good measure.”
– Goodhart’s law
Loosely following it, when product reviews is picked as an indicator of a product quality, then it ceases to function as that indicator because people start to game it. Since product reviews is a key indicator not only to customers, but also to the search algorithm, gaming them is potentially very lucrative.
Product reviews is as old of a concept on the internet as the internet itself. But the incentive to fake them on a marketplace like Amazon has become increasingly higher. Reviews on Amazon help to determine how well products rank in search results, thus creating an incentive to outrank competing products by getting more and better reviews. On regular e-commerce website this is rare, because many more factors are at play.
For many years Amazon has been an online retailer using the marketplace to get brands they couldn’t get otherwise. Sellers in a way were the middlemen allowing them to expand beyond their own reach. But as more of those sellers went looking for opportunities beyond reselling established brands, and instead setting up their own brands, often through private label, the focus on reviews changed.
The rise of seller-owned brands meant products would get launched on the marketplace without external credibility. Sellers thus had to find ways to establish a brand new product, since customers were not willing to experiment buying an unknown product even at a great price. Product reviews is that way. Today it would be hard to find a “How to Launch a Product on Amazon” guide which wouldn’t include a section on product reviews.
This industry was developing faster than Amazon was giving it credit, which has allowed brands to easily game the system. Known as incentivized reviews, thousands upon thousands of reviews filled the system as brands figured the ultimate hack - buying reviews. While illegal, and unethical, it worked. It was the silver bullet everyone was looking for.
Luckily for customers a year ago Amazon put a stop to incentivized product reviews, even going as far as deleting some left previously. But while before it was clear when a review was incentivized, today there are just as many incentivized reviews, but without a clear indication anymore.
In theory incentivized reviews are no longer allowed, but in practice they are used just as much as before. But in ways which are non-obvious, and hard to track.
Some brands even go as far as leaving fake negative reviews on competing products. Customers are often aware that overly-positive reviews are to be taken with a grain of salt, but fake-negative reviews is a whole different problem. They mostly affect the search algorithm, tanking the competing product in search results.
And yet there is no solution to this problem. In a perfect world product reviews is a way to decide on a product’s quality, and value though relying on other people. But the world known as Earth is far from perfect. The gains to be made by faking reviews are substantial, and detecting fake reviews is often hard.
Some argue that machine learning and artificial intelligence are the solution. As a tool to filter out fake reviews. Amazon is already doing some work on this, but fixing it perfectly is likely impossible - a well crafted review has no signs of manipulation. Both Facebook and Google are hiring thousands of people to address content issues themselves, they too couldn’t automate it. And thus it will continue to be a target to be gamed.
In “All Systems will be Gamed: Exploitive Behavior in Economic and Social Systems”, W. Brian Arthur writes:
“There is a general rule in social and economic life: Given any system, people will find a way to exploit it. Or to say this more succinctly: All systems will be gamed. This is not a universal rule and it is certainly not a physical law; it is merely an observational truism. Given any governmental system, any legal system, regulatory system, financial system, election system, set of policies, set of organizational rules, set of international agreements, people will find unexpected ways to manipulate it to their advantage.”
Eventually customers will learn not to trust reviews at all. This will take years, but it will inevitably happen. But then there will be a new metric to game.
The question is then what metrics can customers and search algorithms use which are actually useful. An obvious one is how many people have purchased a product, but popularity is even easier to fake given enough capital to execute purchases. This exactly why sometimes a New York Times bestseller book is not actually a bestseller. Instead it was bought to be a bestseller.
The point of this is that all metrics are inherently targets to game them.
Thinking long-term Amazon needs to be in a position of trust. Customers need to trust Amazon that it is doing its job in building a marketplace of dependable sellers, which are sourcing products without counterfeits, which then are ranked in a fair way. None of these are always true today, time will tell if this gets worse.
There is no such thing as a 5-star rated Floor Standing Hat and Coat Rack. People want to rate the new Star Wars on IMDB. No one wants to rate a $19.95 coat rack they just bought.