Data-Driven Price Optimization: From Intuition to AI

The new era of buyers has a lot of information at their fingertips and companies must anticipate this situation, obtaining updated data on the competition and projecting demand and prices to be more competitive.

In this sense, airlines are pioneers and, since 1970, they have been implementing price optimization in real time. Another example: In 2004, a major international hotel chain abandoned the fixed pricing model to adopt a 100% dynamic, online pricing model. And today, in general, all retail has implemented different optimization strategies.

Data volume

We have basic data on both our operation and the market: our stock, our costs, competitor prices differentiated by location and time. But, in addition, optimization algorithms are influenced by data such as weather, events, traffic, social networks, news and all types of external data.

Take as an example the page www.preciosclaros.gob.ar of the National Government, which has a large number of prices online by product and location. We will only take this external source of information to understand the volumes of information available.

A supermarket could monitor prices on 1000 products, a price comparison on 3 competitors in 20 locations (assuming it has only 20 branches) and a 30-day window of time, initially.

This simple exercise gives us: 1000 * 3 * 20 * 30 = 1,800,000 data.

Now, with all this data, what do we do? How do we analyze them and what conclusions can we reach?

For a team of people it is a very complex task. But for an artificial intelligence algorithm it is ideal.

In short, if we have a lot of information and we do not have the ability to automate its processing, obtaining results will require a lot of effort.

Prediction

Many organizations end up doing summary sampling and simply estimate prices by intuition.

If we apply a discount or increase the price of a product, how will it affect the demand of our customers? How much should be the minimum discount to exhaust the stock?

These questions are really complex to answer, but in some scenarios it is possible to predict demand and for our AI algorithms to help us validate different hypotheses before being applied and not analyze their post-implementation success.

Optimization Process

To implement an optimization process we must execute a series of steps:

1. Get as much information as possible:

• • Know the competition. Collect competitor information in an automated and systematic way.
  • Understand the consumer. Obtain behavioral patterns and customer segmentation.
  • Have the cost structure as reference data.
  • Get external data: seasons, macroeconomic data, weather, events, holidays, traffic, etc.

2. Process, diagnose and automate:

• • Map products. Identify, classify and associate competing products with our catalog and identify similar ones that could be compared by price.
  • Identify the patterns. And this is where artificial intelligence comes into action. With all the information available, AI will detect behavioral patterns that will allow us to implement price recommendations.
  • Monitor KPIs. If we think about automation, we must have dashboards that show us the efficiency and evolution of key indicators.
  • Business rules. Once we have reliable recommendation rules, they can be implemented automatically.

Technologies

Depending on needs, a data scientist will be able to recommend the most appropriate technology.

For the first group of tasks (obtaining information) there are numerous technologies such as:

• Web Scrapping and RPA, which are based on robotizing and emulating human interaction on the web to obtain information.
• Crawling: Internet search engines similar to Google.
• APIS: integration via APIs against partners and public services that provide information.

For the second group (processing, diagnosis and automation) other technologies will have to be put into practice:

• Machine Learning. Regression algorithms are usually the most used to predict the price of a product, but there are numerous approximations.
• Business Intelligence, to create dashboards and KPIs that allow us to understand how the business is evolving, the decisions we are making, and adjust business rules.
• Business Rules. It will allow us to design business rules such as “we want to be no more than 5% above the competition”, “increase margin up to 15% depending on remaining stock”, “increase promotion up to 10% if projected demand drops by more than 30%”, etc.
• Product Matching. They are techniques and tools to associate products and, for example, when we enter an e-commerce, when looking at a product it offers us “similar products.”

Project approach

What optimization frequency do we want to have? Weekly, daily or online? Depending on the frequency, strategies may vary.

Nowadays, e-Commerce stores usually implement online price optimization while many large chains, such as supermarkets, choose to use daily or weekly offers.

The variables to take into account are many and the challenge is important. An approach will be required that allows us to get closer to a real, reliable and implementable solution.

For this it is necessary to have a qualified team in the matter, with data scientists and agile methodologies to approximate, refine and correct the solution.

It is advisable to go through short stages, with concrete results that can give way to new challenges. It is important to start with data ingestion, automate the collection of competitor information and other external data with minimum quality.

Once we ensure that the data is reliable and has volume, we can begin with AI algorithms that discover patterns and propose optimization rules. At that point, we must implement parallel approaches that can compare price optimization against our regular business.

In terms of price optimization, without a doubt the opportunities and associated benefits are enormous. But the challenge, too.

The inclusion of technologies such as AI is something that all managers should study and evaluate. They are increasingly accessible and economical tools. And other competitors are already using them.