These practices are starting to attract the attention of governments around the world. In February the European Commission opened three investigations into suspected anti-competitive practices in e-commerce.
Just a month later, EU Competition Commissioner Margrethe Vestager warned Germany's national competition regulator about the risks of pricing algorithms.
She revealed that two-thirds of retailers use algorithms to track the prices of their competitors and that some of them also use the software to autonomously adjust prices. The systems can also automate collusion between companies by driving up the prices of their products for their mutual benefit.
Policymakers, regulators, and competition authorities are in a constant game of catch-up as data troves deepen and algorithms grow more sophisticated. Businesses engaged in algorithmic pricing have an automatic head start. There's no record of pricing decisions when decisions are made autonomously so evidence of illicit behaviour is difficult to produce, and there's no digital equivalent of whistleblowers to protect consumer rights.
A group of leading academics and European Commission officials discussed the subject at a recent meeting of the Oxera Economics Council, which resulted in a discussion paper on the impact of algorithmic pricing.
They found that they can also yield positive results for consumers. They can use algorithms to find better prices and benefit from the emergence of disruptors, as faster price adjustments to market changes can increase competition by giving new companies more opportunities to offer better value.
Algorithmic collusion first hit the headlines in 2015 when e-commerce executive David Topkins pleaded guilty to rigging prices for classic cinema posters sold on Amazon.
Topkins admitted to conspiring with his competitors to artificially inflate the market for their products using software that autonomously maintained non-competitive prices at the expense of their customers.
The first antitrust e-commerce prosecution ended with the retailer earning a $20,000 fine for his crime, but future algorithmic collusion could be far harder to prove.
Algorithms could "tacitly collude" to coordinate pricing by monitoring movements in the market and the behaviour of their competitors and automatically reacting to it. Such collusion would be more effective and better concealed than how a cartel typically operates through humans sharing information.
"The tacit side is very hard to prove and it’s not obvious that it's illegal either necessarily," says Dave Jevons, the principal author of the Oxera paper.
"If you ask an algorithm to profit maximise and you're using artificial intelligence which monitors your competitors' prices, the algorithms could all start to work out that the best thing for them to is to not engage in a price war. And of course, price wars benefit consumers because they get lower prices.”
As algorithms are constantly learning and adapting to the environment in which they operate any collusion is challenging to detect. Competition authorities would have to test the impact of an algorithm using specific data points on changes in the market. Firms could be asked to run their algorithms in synthetic environments to see how they respond to being fed different pieces of information.
Another growing concern around the use of algorithms is their potential to discriminate against certain groups.
In December 2012, EU gender equality legislation came into force that made it illegal for car insurance companies such as Sheila's Wheels to charge different prices based on a customer's sex. But it can’t prevent firms from combining pieces of personal information to accurately predict somebody's gender.
A solution could be for companies to focus on outcomes rather than methods to understand the effects are equitable across different groups of people. It's an approach that can be replicated by regulators.
Flat pricing is not necessarily fair either, as certain groups incur fewer costs than others and suffer from equivalence. Music streaming services typically don't differentiate by the amount of music you play, although it could quite easily predict this depending on your historical use, personal characteristics and purchasing habits.
Personalised pricing remains a relatively rare model in the market, partly because when it has been introduced the customer reaction has been negative. Consumers can punish firms if they compare their prices to competitors on the market and take their custom elsewhere.
"The famous example of this was Amazon back in 2000 when they experimented with different prices for different consumers on DVDs and consumers reacted very negatively and they had to stop the experiment in the end because their customers were going to their competitors," says Jevons.
"You see it also with Uber and surge pricing, people reacting negatively to different pricing structures at different points in time. There's almost a self-disciplining role of the consumers who vote with their feet if they perceive the structure to be unfair."
Positive pricing models
The consequences of algorithmic pricing are not wholly negative.
If competition authorities can look at outcome probabilities to spot unscrupulous behaviour, then so can potential investors and upstart companies in the sector. The information can help them to spot signals of which markets could be ripe for investment and disruption.
Algorithms could also become a dimension of competition in themselves, as companies develop algorithms that make better decisions and increase efficiencies in the market. Search engines, for example, compete to provide the best results. Digital platforms are increasingly harnessing information to better personalise their services.
The concern would be if all of them used the same algorithm and led development to a stalemate. The prospect of this could increase if more companies begin to buy algorithms off the shelf instead of developing them in-house.
The possibilities for algorithmic pricing have made Jevons cautiously optimistic about the future.
"There are definitely upsides to algorithms such as the faster response, and it can ease entry into the market," he says. "We shouldn't be unconcerned about the potential risks but if we design markets well now we should be able to get good outcomes for consumers in these markets."