Being a product manager is a full-time job. There are a lot a parts to this job that we wish we could have someone help us with. The good news is that we are living in the 21st Century and that means that we can harness the power of algorithms to automate parts of our job. However, it can be all too easy for us to hand a job off to an algorithm and then walk away from it. Perhaps we should not do this, do we really know what the algorithms that we are using do?
The Problem With Algorithms
Product managers know that it’s one thing for algorithms to set prices. It’s totally another thing for algorithms to unfairly raise prices. We need to be aware that a new study shows that this can indeed happen, at least theoretically. Product managers are increasingly using algorithms to help them set prices for such things as plane tickets, gasoline , ride sharing and a range of other goods.
Researchers have now conducted a series of computer simulations in which pricing algorithms learned through both trial and error and without communicating with product managers directly to collude in maintaining prices above levels that would be expected in open competition. The only instruction the algorithms were given as a part of the study was to maximize profits. However, the researchers didn’t tell them how to do this.
In order to prove that the algorithms were colluding and not just randomly landing, in unison, on inflated prices, the researchers forced one algorithm in each simulation to cut its price. The result was that the algorithms then engaged in a price war. Eventually they returned together to higher prices. The bad news for product managers is that experts say such behavior is a hallmark of collusion. Collusion occurs when companies that have agreed to raise prices above competitive levels occasionally cheat on the collusive arrangement but ultimately fall back in line.
Dealing With The Algorithm Problem
The question that product managers need to be considering is as we delegate key managerial tasks to machines, what should we expect? We can hope that markets will function better – perhaps allowing us to offer lower prices. Or maybe machines can find clever ways to extract more money out of our consumers’ pockets. We need to realize that collusion is a clever way of doing exactly that. The study is an important advance in the literature on pricing algorithms. This is a study that product managers need to see. This work is the first step toward creating algorithms can collude in reasonably complex environments.
Whether or not algorithms can learn to collude in real markets, however, is still an open question. When the studies were conducted the researchers used relatively basic algorithms. They were powered by the same kind of artificial intelligence that is used to play chess and used in environments that don’t exactly replicate existing online retail markets. The next step is to bolster the idea that collusion is possible. This can be done by allowing algorithms to be tested in more advanced and complex environments.
In the study when researchers imposed a price cut on one pricing algorithm, a second algorithm independently cut its price in response, before the two began raising their prices in unison. The findings raise serious questions for product managers regarding antitrust regulations in industries where algorithmic pricing is in use or is likely to be used in the future. Effectively the study’s algorithms engaged in what the researchers describe as “tacit collusion”. This is a sort of price fixing that takes place without any explicit communication. Experts say product managers could run into legal problems if they were to agree to use algorithms with the intention of allowing the algorithms to collude on prices. The problem is that these kinds of surreptitious arrangements could be hard to detect, primarily because parallel behavior among algorithms doesn’t necessarily imply explicit collusion.
What All Of This Means For You
Life is complex and product managers have to do a lot of different tasks. The good news is that in the modern era that we find ourselves living in now we can often turn to algorithms to help us accomplish a number of different tasks. One of the most common tasks that product managers use algorithms for is the pricing of their products. However, researchers have discovered problems with doing this.
The problem is that researchers have discovered that when algorithms are allowed to manage prices, they can end up colluding with each other to create prices that are higher than they should be. The researchers took steps to confirm that the higher prices were not just a random occurrence. Product managers need to understand that collusion is one solution to the problem that the pricing algorithms are trying to solve. It is still unclear if pricing algorithms operating in the real world would act this way. Product managers need to understand that if they use algorithms to set product prices they could get in trouble if they use algorithms with the expectation that they will collude with other algorithms.
I think that we can all agree that having algorithms take care of some of the more complex parts of our job is a big help. However, this cannot be a fire and forget type of action. We need to take the time to understand how our algorithms operate and how they may interact with other algorithms. When we know how they will behave in a given situation, we can make informed decisions about using them.
– Dr. Jim Anderson
Blue Elephant Consulting –
Your Source For Real World Product Management Skills™
Question For You: Do you think that product managers are responsible if their pricing algorithms collude with other algorithms?
Click here to get automatic updates when The Accidental Product Manager Blog is updated.
P.S.: Free subscriptions to The Accidental Product Manager Newsletter are now available. It’s your product – it’s your career. Subscribe now: Click Here!
What We’ll Be Talking About Next Time
As any of us who have gone shopping online know, online retail sites often make product suggestions based on our browsing history. New research is suggesting that this may be a mistake. The reason is that it turns out that consumers dislike being observed while they are deliberating. They dislike it to the point that many may react by not making a purchase if they think their shopping is being monitored, or perhaps by avoiding retailers that they know will watch them browse.