How assortment optimization works in retail with AI

by | 8. July 2021 | Basics, Tech Deep Dive

The retail industry is one of the most challenging ones due to its complex supply chains, very large product assortments and low margins. Both traditional physical businesses as well as e-Commerce retailers experience cut-throat competition, and the only way to stay on top of the industry is by continuous assortment optimization and improving the way the business operates and staying ahead of their competitors.

One of the key factors that retailers have at their disposal in order to differentiate themselves from their competitors, is their product assortment: that is the set of products they offer for sale.

Using assortment optimization with advanced analytics and AI, retailers now have the tools to decide which products they should sell at each of their locations, customizing assortments to local customer tastes and different store sizes.

Our approach to Assortment Optimization in retail

We base our approach a five-step process:

  1. Quantifying the cross-selling effect between products
  2. Quantifying the similarity / uniqueness of products
  3. Understanding what product factors drive customer demand
  4. Forecasting the demand of products
  5. Optimizing the assortment

Quantifying the cross-selling effect between products

Let us start with an example here:
Super-fresh, a North-American supermarket chain owned by the grocery retailer A&P, delisted many of its low selling dry grocery items in order to expand their fresh produce offering. The eliminated products turned out to be essential to customers, and when the customers couldn‘t find these products, they went to do their shopping somewhere else, seriously affecting the company to the point of bankruptcy.

Traditionally, purchase managers have used simple performance metrics such as the sales or rotation numbers along with their many years of experience in the job to decide if they should list a new product into the assortment or delist a low selling product.

Unfortunately, these simple metrics used in the past are not enough information to decide if a product should be delisted. Looking at the performance of the product in isolation does not take into consideration the cross-selling effect that products have with one another.

Assortment optimization in retail with AI and advanced analytics have been turning to understand their customers’ behavior better and quantify the cross-selling effect between products. This is typically done by analyzing all historical purchase transactions coming from POS (point of sale) systems and determining the co-occurrence of product pairs.

For retailers is vital to understand what happens when a customer doesn’t find the product he is looking for. Will he choose another similar product? Or will he take his business somewhere else?

Quantifying the similarity & uniqueness

While some products might not have stellar performance numbers by themselves, they can be key factors in driving customers to your business motivated by finding these unique products.

On the other hand, there are products that customers can easily replace with similar products. This is what is known as the “cannibalization effect” between products that happens when customers have no distinct preference between two products which are similar.

Retailers can now use assortment optimization with AI and advanced analytics to quantify the cannibalization effect between products and their uniqueness, improving their listing and delisting process.

The uniqueness of products is estimated by analyzing the cross-selling relationship among products. For example, products are similar if they happen to be in similar shopping baskets. For example, milk A and milk B would be considered similar if the products which typically are bought together with them are similar too.

Using a similarity score between all the products in the assortment, retailers can determine a product’s uniqueness by how many similar products there are in the assortment.

Understanding what product factors drive customer demand

New products pose another difficult challenge, as there is no historical sales data on which a reasonable forecast to estimate the sales performance of the product can be based. This is what is known as the “cold-start” problem.

Assortment Optimization with AI and advanced analytics are also helping retailers to solve this challenge. Retailers have large volumes of data at their disposal which can help them understand which product attributes matter the most to customers, and what their clients will do if they do not find their preferred products.

Forecasting the demand of products

Armed with AI tools, modern-day retailers can now estimate the demand of products to select the most promising ones based on store characteristics. These demand forecasts are also performed taking into consideration the full assortment and the interrelationships between products, capturing the uniqueness of products and the cannibalization effects that happen when a new product is introduced to the assortment, as part of the demand of a product is transferred when a new similar product is listed.

The future is called assortment optimization with AI

Optimizing the assortment is a continuous process that is never finished. It is therefore important that retailers develop a culture to use advanced analytics in their decision making.

To determine if a given product should be listed or delisted from the assortment, an analysis is done across multiple dimensions.

  • The economic performance of the product
  • The uniqueness, cross-selling and cannibalization effects between products
  • Supply chain considerations
  • Strategic objectives.

Delisting tends to reduce the assortment complexity while improving margins. Most delisting however is motivated by the introduction of new products as store space is limited.

The assortment of a store needs to be considered as an evolving entity, that continuously should adapt to new products and customer preferences while taking into consideration the supply chain costs and aligning that to broader strategic objectives. For example, these strategic objectives could be a higher ratio of organic products, or more gluten-free products in the assortment.

From category allocation to in-category optimization

The assortment optimization process is a multi-level process that happens first at the category level.

As a first step, the space per store is allocated by categories analyzing their marginal contribution per meter and optimized using the trade-off between the different product categories.

As a second step, the optimization happens at the “in-category” level where an optimum balance between similar products and unique products is obtained.

The goal of the in-category assortment optimization is to find the best mix of similar and unique products so that it maximizes the sales of the store as well as some other supply chain and strategic objectives.

The main concept of the in-category optimization is that does not make economic sense to have many similar products in the same category competing for limited shelf space, and that space can be better allocated to products which have higher “uniqueness” and cross-selling effect, enhancing the breadth of the assortment while reducing the overlap between products.

Benefits of assortment optimization

Changes to the assortment have a great impact across the whole supply chain. Factors like the cost to serve the product, where end-to-end logistic costs are calculated, help in determining optimum assortments that reduce the whole supply complexity and costs.

Another important benefit of an optimized assortment is ensuring an optimal space allocation according to the store characteristics and local customer preferences.

Using advanced forecasting methods, retailers can assess the optimal assortment for each store taking into consideration local store characteristics and discover which are the factors that have an impact on the sales performance of products. This enables retailers to select which products best fit to local characteristics, customer preferences, supply chain factors, and strategic goals.


Getting the assortment right, is crucial for retailers. Better assortments lead to improvements in many areas of the business. Not only are sales increased by a better product assortment, but the reduced supply chain costs and therefore higher product margins ultimately have a great impact on the bottom-line of the business.

For a deeper introduction on what AI can do for your retail or e-Commerce business, take a look at our AI in retail whitepaper. It explains the basic terms you need to know, discusses application areas and real use cases from our client projects. In addition, you will get to know our data journey, the multi-level process we follow in our AI and data science projects.

<a href="" target="_self">Dr. Jose Manuel Berutich Lindquist</a>

Dr. Jose Manuel Berutich Lindquist

Jose Manuel holds a PhD in Artificial Intelligence and has been working as a Data Scientist for the past 10 years focusing on forecasting, deep learning and complex optimization problems for the retail, manufacturing and electricity industries.


You may also be interested in


Data Navigator Newsletter