Pricing is hard - but AI can make it easier
We take a look at how to apply these AI techniques within your company in order to reap the benefits of pricing algorithms
Algorithms and artificial intelligence are at the cutting edge of pricing strategies at the world's leading companies. We take a look at how to apply these techniques within your company in order to reap the benefits of pricing algorithms and prevent the common pitfalls.
If you opened the news recently, you probably read something about Artificial Intelligence (and how it was going to destroy the world, fix the world, or just do nothing at all). While opinions differ on exactly how big the impact of AI will be on our everyday lives, there is clearly a massive renewed interest in a field that has lingered in the dark corners of academia for a long time. There's a good reason for this renewed interest. Computer algorithms are taking on progressively more complex tasks, and are now often outperforming humans in these tasks.
A prime example of a field where algorithms are outperforming humans is in pricing products - a field known as algorithmic pricing. You might know that if you have bought a plane ticket in the last two decades, the price was set by an algorithm, and not by a human. The idea of letting algorithms decide prices has since grown from the airline industry to pricing concerts, pricing taxi rides (think of Uber's surge price), pricing hotel rooms, and much more.
This growth in this field is primarily linked to the growth of online commerce. Indeed, on-line retail fulfills two of the fundamental needs of algorithmic pricing: 1. a huge quantity of (good) data, to allow pricing specialists to create robust statistical models of customer behavior, and 2. the ability to change prices quickly and efficiently. As the prevalence of on-line retail grows, so too does algorithmic pricing. In light of this, many of the major retailers and marketplaces have acquired companies that specialize in algorithmic pricing, to allow them to gain an edge over their competitors.
What is Algorithmic pricing
But how does algorithmic pricing work exactly? In reality, a pricing algorithm is a mathematical formula, where a given price is calculated based on several variables. A simplified example of a pricing algorithm could be this: the price of an apple in an on-line grocery store could be a function of the price of a pear in the same store (a substitute), the price of an apple in a competing store (a competitor), and whether there is a lot of demand for apples (by looking at the number of clicks this apple has received, relative to other products). The formulas usually attempt to maximize or minimize an outcome: the revenue of the company, the number of unsold products, etc.
Both the variables (price of pears, price of apples in other stores, demand) and the formula can range from very simple to extremely complex. Modern algorithms will take into account many variables.
When Algorithmic Pricing Makes Sense
There are good reasons why progressively more companies have moved to algorithmic pricing models.
Firstly, pricing has always been notoriously difficult, with much of the practice being seen as more of an art than a science. Algorithms help solve the age-old question of "what is the right price for my product", by taking into account various parameters (such as supply, demand, competitor's pricing, etc) in a structured way. This allows us to objectively measure the performance of one algorithm versus another algorithm.
Algorithms help solve the age-old question of "what is the right price for my product"?
Secondly, as pricing is a major lever on companies' bottom line, it makes sense for them to try to optimize this lever as much as they can. However, this often leads to more frequent price changes and more complex pricing structures, increasing the complexity of setting prices. Airlines, for example, might sell every single seat for the same flight at a different price, in function of demand, remaining capacity, cost of operation, etc. Once again, algorithms solve this complexity neatly, by taking over the heavy lifting of calculating and setting prices.
Thirdly, as business environments grow more mature, competitors tend to aggressively react to pricing changes of their competitors. Algorithms have much faster reaction times than humans, meaning that they can react in real-time to changes in market circumstances, such as promotions by competitors.
Fourthly, governments have become more and more aggressive in pursuing price-fixing in various industries. Algorithms, however, leave a clear audit trail and show intention explicitly, which makes for an excellent legal defense in these cases (but also an excellent proof against your company, if you are rigging prices).
Fifthly, algorithms are repositories of knowledge. As they are mathematical equations, and not knowledge or skills of a specific employee, they can be understood by, and tailored by anyone with the required skills.
Finally, in the last decade or so, algorithmic pricing has improved in performance, due to data becoming progressively easier and cheaper to obtain. Furthermore, the rise of relatively cheap computing means that these algorithms can be run quickly at low cost.
Algorithmic pricing's bad reputation
And yet, algorithmic pricing still suffers from a bad reputation.
Airlines lost money for decades trying to understand how to use algorithms efficiently. Customers' dislike of being charged different prices for the same product is well known (just Google "Uber surge price", and you will know what we mean). Finally, stories of algorithms gone haywire are now legendary (such as Amazon's 23-million-dollar book listing).
The three main reasons why companies fail to successfully implement algorithmic pricing are due to a technical difficulty, lack of customer acceptance, and lack of proper safeguards. None of these are insurmountable, given the right strategies.
These examples show the three main reasons why companies fail to successfully implement algorithmic pricing: technical difficulty, lack of customer acceptance, and lack of proper safeguards. None of these are insurmountable, given the right strategies.
Algorithmic pricing Best Practices
How do companies best minimize the risks of algorithmic pricing? We'll give some of the best practices to implement algorithmic pricing in your own company.
Don't start price wars: Having algorithms that respond too aggressively to changes in your competitor's pricing risk generating price wars, which might destroy value for your company. Carefully monitoring the response of your competitors to your own price changes is a crucial element of successful algorithmic pricing strategies.
Don't optimize for short-term revenue: While it might be tempting to create algorithms that only aim to maximize revenue, it is important to think of other, long-term drivers of your business. Take into account other factors, such as customer satisfaction and your brand image when deciding on pricing algorithms.
Use the maximum price as anchoring point: There is a clear consensus within economic and psychological circles that people strongly prefer avoiding losses to acquiring gains (a tendency called loss aversion). Make sure your clients have the impression of winning by implementing algorithmic pricing, rather than them losing out. More specifically, showing your customers the highest possible price point for a product, and the current (lower) price, is generally a much better strategy than the opposite (such as is the case with Uber's widely hated surge price).
Market pro-actively: Use algorithmic pricing to your advantage, by being explicit about your pricing strategy. Helping customers understand how you price (without giving strategic details that might help your competitors) generates trust.
Set price boundaries: setting both a floor and a ceiling for your pricing ensures that your algorithm cannot generate prices that would not make sense for your customers, such as Amazon's 23 Million dollar gaffe. Taking into account the cost of the product, whether a product is a loss leader or not, and setting a fixed multiple of the price as the maximum price are all potential variables algorithms could take into account.
Algorithmically price the long tail first: The "long tail" of your products refers to the products which have a relatively low impact on your overall business. Algorithmically pricing these products first allows you to offload the burden of pricing large sets of products to a computer, while retaining the control over your key products, while the algorithms are being tweaked.
Run Trials: Run trails by algorithmically pricing products at certain sales venues and compare the performance of these venues to those where you do not algorithmically price. Alternatively, you could algorithmically price on certain days, and not on other days, to compare the performance of algorithms. This strategy allows you to remove the guesswork out of the performance of your algorithmic pricing strategies.
If you are curious to find out how you can leverage AI in your organization, just leave us a message!