Cube.js Blog

Strategic Pricing in Retail with Machine Learning

Building a dynamic pricing strategy in retail

Author avatarAlexandr Galkin/Startups/August 23, 2017
Strategic Pricing in Retail with Machine Learning
Show Original

Today, retail faces new challenges and needs to learn very fast to process all available market data, gain useful insights, and evaluate outcomes. The easiest way to achieve this is by having a dynamic pricing strategy that uses machine learning techniques.

The Statsbot team asked the specialists from Competera to tell us about building a good strategic pricing in retail.

If you follow retail news, you’re aware of different tips and tricks retailers around the globe use to keep customer engagement high. They create cashier-free stores, self-driving shops, and use new engagement techniques in old-fashioned offline stores. But all these are lame attempts as long as the price is still a prevailing reason to buy stuff for 60% of shoppers.

That’s why modern retailers need an intelligent dynamic pricing strategy. Although new ways of collecting and processing data greatly help them with that, most specialists still use manual crawling techniques, Excel Spreadsheets, or simple price tracking solutions like Price2Spy/Prisync.

The aftereffect of such an approach is obvious: only 39% of retailers succeeded with useful insights.

Price adjustments made for the whole front store’s inventory in just a second, in response to real-time demand, is much more effective than those set manually with all the human mistakes. It is where machine learning steps into the room, giving retailers an option to optimize not only prices, but also business strategy, costs, and managers’ efficacy.

Keep reading to find out how exactly a retailer can employ strategic pricing and outperform his competitors.

Step 1: Errorless Data Collection

The very first thing a retailer needs to handle for strategic pricing is data. Before any pricing decision can be made, clean and useful data needs to be collected.

Before the rise of technology, data collection in retail was straight forward:

1980s — hire a person who will visit a competitor’s store and manually collect all data needed;

1990s — build custom crawler and collect all the data you need;

2000s — create more tools to allow a crawler to get past a competitor’s protection tools, and collect data from different marketplaces as well;

2010s — find and utilize a solution that will suit all of the above.

Nowadays, most items on the list are senseless. Custom crawler creation is too complicated, so the main goal for retailers is to find an appropriate solution that will act behind the described scenarios, and wisely use the machine learning techniques to:

  • Check itself for mistakes and double check data quality (anomaly detection to avoid parsers errors);
  • Automatically avoid website protection with self-learning human behavior imitation;
  • Inspect, clean, transform, and deliver data;
  • Produce data crawlers automatically.

All profitable and loss pricing decisions depend on the data quality. Auto crawlers, with a help of computer vision and machine learning algorithms — instance based learning — automatically extract structured items from the web page faster and more properly than by manual methods:

  • 1200% crawler initialization time reduction, from 12 min per item for manual collecting to 1 min with auto crawler;
  • ~70% cost reduction for auto crawler creation in comparison to the manual programming;
  • anomaly detection algorithms show deviations in the way crawlers work.

Dynamic pricing tools

To get started with strategic pricing, retailers can use one of the following solutions:

Competera is an advanced retailing solution based on a best-in-class data IQ and competitive intelligence family of products to gain leading edge pricing, promotions, inventory analysis, and management and optimization.

Wiser (QuadAnalytics) helps automate price monitoring, reprice your products based on customizable rules, and get actionable insights into current market trends. For enterprise-only retailers and brands.

Market Track is the leading provider of retail promotion. It offers both retailers and manufacturers competitive price monitoring, Excel-based reporting with custom filters, and more.

Upstream Commerce is a configurable, scalable, and accurate retail solution for price management automation and high data accuracy.

Step 2. Advanced Data Analysis

The next stage in handling a dynamic pricing system is data analysis that should lead retailers to correct and quick decisions based on direct price suggestions or recommendations.

To get more insights out of the collected information, there’s a need to visualize it. Good visualization provides the possibility to quickly grab all deviations and react on them. All the tools which we suggested for strategic pricing have this feature. Let’s take a closer look in the case of Competera Price Management.

A Price Index graph built by Competera Price Management

This graph shows the price position of your products, product categories, leading brands, and a whole assortment of data for your online store, regarding similar parameters of the competitors. This includes the market in general, especially if several competitors constitute it. In addition, the Price Index detects which of the competitors, and how intensively, influence your sales during a certain period of time.

Price Index and sales time series analysis determine the Price Index elasticity of sales, and key competitors whose prices affect the retailer’s sales. Such a sales elasticity model helps to create a short-term, e.g. one day or two, forecast of future sales. Machine learning time series analysis can be applied to different periodic fluctuations of any nature — by year, month, a day of the week, time of the day, etc.

When a retailer handles a qualitative visualization, he can get the best time for competitive price monitoring, analyze and improve sales performance, study customer behavior and demographics better, etc.

Step 3. Agile Pricing Framework

Can you figure out what price is right for your products without using advanced tools? The number of variables that needs to be taken into account before price setting is very wide: product seasonality, price elasticity, competitor prices, competitor markdowns and promotions, customer demand, retailer’s desired margin, etc. To apply all of them to a single product pricing, the IFTTT (“if this then that”) approach can be helpful.

Of course, all these variables can be considered manually, by category, or by using pricing managers with Excel or Google Spreadsheet, whichever is the best option for small or beginner retailers. Still, it’s nearly impossible to handle thousands of products in no time without automated rules-based pricing algorithms.

Moreover, after a dynamic pricing system and smart suggestions, crafted after algorithm processing, comes A/B smart price testing. This allows a retailer to measure pricing rules influences on performance and forecast gross margins.

One of our clients found a repeated pattern with A-category items. Machine learning techniques helped to find items with similar properties from B-category and apply the same pricing scenario. As a result, B-category sales raised.

More advanced, unsupervised machine learning can provide a retailer with:

  • searches for clusters with identical sales patterns to apply efficient pricing rules;
  • forecasting a product’s transferring between categories, e.g. for ABC-analysis;
  • creating more advanced pricing rules on the basis of previously tested ones;
  • anomaly detection.


Retailers with no strategic pricing are playing in a blind zone with no estimation engine that is able to calculate, analyze, adjust, and set prices without mistakes, and which allows testing all hypotheses on a tiny product range before applying them to the whole inventory.

But before implementing a dynamic pricing strategy, a retailer needs to set the traction of direct and mediated factors to calculate the efficacy of the approach. Direct factors can be marketing budget saving, margin growth percentage, inventory optimization index, micro- and macro-conversions of a product card, etc.

Mediated factors are not so obvious, yet equally important, saving managers hours of working time and efforts to allow focusing on the strategies and trends instead of a business-as-usual workflow.

Long story short: strategic pricing powered by machine learning is only the first stage on the transformational path for the retail we used to know. We believe soon, it will open the doors to prescriptive- and AI-based pricing with all the benefits of personalization of interaction with a shopper.

You may also like: