Inside AI

Retail Pricing: Getting the Most Out of Machine Learning

Machine learning is here to help retailers become more productive and profitable

If you are reading this article, you are probably looking for new approaches to optimizing pricing. Simply put, traditional methods are not good enough anymore. It is true that they are still working well enough to let you survive, but they are insufficient to put you ahead of tech-savvy competition. In this article, I’d like to cover two things as Competera’s price optimization expert: the tremendous potential of machine learning in retail dynamic pricing, and how you can get ready for the adoption of the technology to take advantage of every opportunity it provides.

Machine learning in retail

Big names have been using machine learning in dynamic pricing for years. Among the brightest examples is Amazon, which was among one of the earliest adopters of the technology. This was, for sure, one of the factors which contributed to the company’s stellar growth in the market value: from 30 billion in 2008 to almost 1 trillion in 2019. Other players using machine learning include Walmart, Jet, and Newegg.

What makes the algorithms so efficient and popular? It is their ability to process massive amounts of data and enable humans to make real-time precise decisions. Today the retail market is getting increasingly dynamic, which means retailers need to change prices more frequently than ever before. At the same time, every business wants to craft optimal prices for every one of its products. Two major issues make price optimization difficult without ML:

• The impermeability of the future. When calculating prices, retailers inevitably need to factor in the reaction of demand for price changes for every item. But it is very hard to measure since it is non-linear, multi-factor, delayed (customers have shopping cycles and are exposed to new prices with a delay), and heavily biased by other factors like seasonality, distribution, advertisement, and competitive actions.
• The sheer number of products to monitor and reprice weekly or even daily, which is impossible for human managers anymore. Also, by altering the price of a particular product, retailers inevitably impact the sales of many other items in the product portfolio. Such an impact has to be taken into account for every pricing decision — and pricing managers simply cannot do that without using technology.

Tackling these two obstacles seems unyielding for humans — and is increasingly outsourced to self-learning algorithms. The algorithms process the infinite number of pricing scenarios (which equals the number of atoms in the universe) to choose the most relevant one in real time. They take into account thousands of latent relationships inside a product portfolio in order to recommend individual prices that altogether maximize sales and revenue of the entire portfolio.

I’ve gathered several examples of the machine learning-based price optimization case studies to better understand how retailers use it to grow.

Case 1

Goal: To boost revenue and sales across over 100 price zone

Sector: FMCG

Type: Offline

Results: A 15.9% uplift in item sales, an 8.1% revenue growth, a 9.8% gross profit front growth

Case 2

Goal: To maximize revenue without losing profit margin

Sector: Consumer Electronics

Type: Offline

Results: A 16% revenue growth, a 14% sales uplift

Case 3

Goal: To boost all sales KPIs

Sector: Giftware

Type: Online

Results: A 24.7% item sales uplift, a 9.3% revenue growth

So, self-learning algorithms in pricing can be extremely beneficial for your KPIs. What steps do you need to take to be ready for them?

Getting ready to embrace machine learning in pricing

There are four stages which you have to go through to deploy algorithm-based pricing solutions and make the most of them: to prepare all the necessary data, choose a provider or build an internal pricing system, teach your team, and improve the system every step of the way.

Data

Surprisingly, preparing and collecting data can be a big challenge since the data is often unstructured, stored in different sources and has many errors. Getting it ready can take several weeks or even months. In some cases, it would be even better to hire a contractor that is good at data management. But the good news is: you need to do it only once. Even if you are planning to use machine learning to automate all business processes, like logistics, inventory management, or dynamic pricing and bring them together into a single ecosystem, you need to prepare your data just at the beginning, as it will have a more or less similar structure regardless of whether you use it to adopt ML in pricing or logistics.

External or internal pricing solution

Imagine that you have your core competence — something which differentiates you from competitors and increases profit/revenue. But in addition to that, you need to be good at dozens of other things which are at the periphery of your core competence. For example, you manage to offer customers low prices and are a virtuoso of beneficial purchase prices, however, your profits depend not only on low purchase prices but also on optimal shelf prices. Therefore you need to learn to craft prices which are neither too low, nor too high. Being able to do this is a whole new skill which requires a heavy investment of several million dollars in R&D.

There are two options you can choose from in this scenario:

• hire a technological partner whose core business is to recommend optimal prices.
• pump millions in launching a pricing hub in your company and becoming a pricing guru.

When it comes to pricing, retailers need to know exactly which products are better to use machine learning algorithms or rule-based pricing for. How can you know that? It is essential to analyze your portfolio and indicate three groups of items:

• Turf protectors or the products which you share with all or most of your competitors. They generate your traffic since these are the items for which customers come to your store. In most cases, shoppers look for the lowest price for such products — and you have to deliver it to stand out from your rivals. Here you can do without machine learning. What you need is constant competitive prices monitoring and rule-based pricing.
• Exclusive products which you can sell at a higher price. Here you should use machine learning algorithms to change prices a certain way, influence demand reaction and reach a price optimum which allows for generating maximum revenue.
• Products which you share with your competitors, but which do not have to have the lowest price to attract customers. Since shoppers base their purchase decision on the power of your brand rather than price, you need to use machine learning to calculate how much your brand popularity can allow you to raise prices for these particular items.

Team education

Once your pricing system is ready, it’s high time to train your pricing or category managers. They need to understand what they are working with and accept that the logic behind every price recommendation provided by algorithms is not always clear. Basically, they need to learn to trust algorithms bearing in mind that there is always a way to put limitations to algorithms or test their suggestions before using them. At the same time, managers need to be aware that algorithms mostly come up with the right decisions: the system would perform even better if it knew everything your managers know. So, when coupled with the managers’ expertise, algorithms show much better results.

System improvements

You need to come up with a strategy and provide the pricing system with a set of goals it needs to hit within this strategy. You should monitor how the system proceeds and make corrections along the way. If your goals and strategy have changed, you need to adjust your pricing system accordingly.

On a separate note, the pricing system you use always has to be as cutting-edge as possible. Again you can either hire a technological provider that is responsible for its solution to be up-to-date or create a dedicated team to monitor all the technological advancements in machine learning and incorporate them into your system.

All-in-all, what I wanted to emphasize in this long-read is that machine learning can be a truly advantageous option for retailers which seek not only to survive, but thrive. To benefit from it, you need to do four things:

• Prepare your data.
• Build an internal solution or hire a technological partner.
• Educate your team.
• Update your pricing system constantly.