Predictive pricing supposedly came from the Japanese rice exchanges in the 17th century, analyzing rice stock to characterize stock market behavior and make predictions about future prices and trends. Today, retailers use the strength of algorithms to establish optimal prices and generate revenue. What are some of the biggest changes that machines have made in retail?
From the 19th century to today
Pricing recommendations lie between the methods that the Japanese had used and today’s machine-based products that make pricing recommendations. In the 19th century, Sir Francis Galton invented the initial machine learning algorithm called linear regression, which was able to set a one-way dependency of a variable from another one and then predict one of them with the help of the other.
Since then, people have continuously enhanced their computational capacity as they’ve moved on to more advanced machines with the help of gathered data. Also, due to today’s more complex nature of data, intricate algorithms are a must.
What change has come about from modern predictive pricing?
Today, “predictive pricing” has been coined as companies that use machine-operated algorithms combined with information to reach certain goals. Retailers’ bigger goals revolve around boosting revenue, therefore, they must alter their price management approach, read more here.
The three criteria that make something “optimal” include “the more, the better” like in sales, “the less, the better”, like refunds, and “optimal”, like prices. These cannot be infinity or zero.
Price analysis has evolved throughout time:
1. Companies attempted to price products with a margin benefit to themselves (cost-based pricing);
2. Free niches decreased, so the competition increased. Companies started using competitor price tracking tools so that they weren’t too high or low (competitor-based pricing);
3. Retailers use analytics to stay away from descriptive price comparison and accurately count the price change impact on sales (predictive pricing).
Qualitative and quantitative are the two kinds of prescriptive pricing. Qualitative pricing is described as, “My prices are too high in contrast to competitors.” However, what does “too high” mean and who are your competitors?
Algorithms handle historical data in predictive pricing which is about price and sales dynamics. They set a quantitative relationship between the two to suggest retailers’ optimal prices.
Pros and cons of predictive pricing for retail
Advantages:
1. More is being sold because customers see the prices as optimal, so sales increase.
2. When prices are properly managed, retailers always know why and can repeat it.
3. Algorithms improve and computational abilities speed up with more data. Faster pricing decisions are made and prices are reset frequently.
Disadvantages:
1. Prices change too quickly, so customers cannot compare them fast enough. They are buying more spontaneously versus based on well-informed decisions.
2. Algorithms, predicted prices, customer behavior, and everything else is getting more complicated to manage.
Customers often display less rational actions, which are however combined with normal behavioral patterns. Therefore, retailers have to look at the optimal price for each customer, which has resulted in personalization and hyper-personalization.
In addition, analysts and programmers do not yet have enough information to create algorithms that would not bother shoppers. However, offering optimal prices would make customers satisfied and bring retailers more money.
Eventually, algorithms will have the capacity to do so; they just need analysts to gather more customer data and create advanced software.
Algorithms influence analysts
Algorithms gather and structure data and make pricing recommendations. Therefore, analysts can focus on upper-level decision-making instead of manually going through spreadsheets and making errors.
Instead, analysts are thinking strategically and adjusting the algorithm according to proper business goals.
However, can algorithms replace people? Yes, once they are able to think strategically. Currently, though, they are “short-sighted”, so they can only predict one step further.
Data runs this century
Algorithms need lots of high-quality, complete information to make predictions; a minimum of hundreds of thousands of data points. Even more for neural networks.
Algorithms require data in the same structure and format with variables and tracking results from the past three years.
New retailer mentality
Today’s algorithms are more accurate and advanced, but how do they make decisions? Companies have to change their mentality to trust machines without controlling them and understanding how they work. Retailers should not manage prices, but rather the algorithm.
Algorithms have to be tried out in a pilot to analyze if they are more or less effective than a group of analysts. Right now, retailers can work without an algorithm, but when most of them are using it, the rest will not be able to compete.
Conclusion
Algorithms are completely changing retail. However, retailers are faced with two problems:
1. customers that are more demanding and unpredictable;
2. having to trust the software.
Predicting if algorithms will take control is difficult as machine learning approaches can still fail to outline and predict the continuously changing retail space in comparison to humans. The truth is, we simply do not know where retail is headed.