Data science use cases and examples for e-commerce

Data science and machine learning are incredibly useful tools to use in retail, especially e-commerce retail. In this space there are a multitude of use cases that are available, from natural language processing to behaviour prediction and demand forecasting.

Which e-commerce data science projects should I work on?

Data scientists and their teams often have a list filled with potential ideas, some are crazy in scope and size, others are just a new KPI. The skill of a data scientist is to look at that list of ideas and decide on which project will bring the most value for the least effort, the sweet spot. This post seeks to help that decision by outlining the key use cases that data scientists should look into for e-commerce retail.

Customer KPI reporting

Before the ‘fun’ projects where we can use machine learning it’s important that the basics are covered. Without an understanding of the KPIs (key performance indicators) we want to affect we will never know if the changes made had a positive or negative impact. It’s therefore vital that a set of KPIs are decided upon and are easy to track over time. Examples of KPIs that are relevant for an e-commerce setting are:

  1. Average order value
  2. Abandoned baskets
  3. Average orders per customer per week/month/year
  4. Percentage of customers that are active or inactive

Customer segmentation

Customers very rarely act in a completely unique way, giving us the ability to group customers into segments that describe a certain type of behaviour. Examples of the types of customer segmentation we can create for e-commerce customers are:

  1. Product interest. What generalised types of products is a customer buying? Knowing this helps to personalise communication towards a group of customers, and avoids situations where a customer is receiving irrelevant information.
  2. Pricing. What level of price is a customer typically purchasing at? Using this information one can adjust the products shown to a customer to better suit them.
  3. Order revenue. How do you know if a customer is amongst the highest or lowest spenders? Segmenting customers on their order revenue will allow us to dynamically know if a customer is of high or low value.

Customer purchase and revenue prediction

Predicting if a customer will purchase in the coming weeks or months and the corresponding revenue from those purchases allows us to better forecast demand, personalise communication, and understand how customer behaviour changes over time. One thing to note with these models is that the accuracy will vary greatly from industry to industry, with those in a high frequency industry having the best data for these types of models.

Demand forecasting

Having a good understanding of future demand makes purchasing, planning and operations easier for e-commerce businesses. Building demand forecasting models at an aggregated level is often the first task, but it is equally beneficial to create corresponding models for specific products and services, such as:

  1. Delivery method
  2. Customer service contacts
  3. Return volume
  4. Products down to SKU level

How to put data science solutions into production for e-commerce

All of these use cases are important for a data science team to look at in the e-commerce space, but we also need to think about how they are used. Is the outgoing data just to be used for reporting and analysis, in which case we need to think about how best to display and structure this data in a visual tool. Or is it going to be used operationally in your e-commerce or marketing systems?

These are important questions to answer when you’re planning your project, which if overlooked can lead to wasted time. As an example, if you create a segmentation model but have no way of using the resulting data to change your direct communication then it’s probably not worth working on.

The impact of data science on e-commerce

Data science and machine learning have the power to provide immense value to e-commerce. Understanding customer behaviour at scale through segmentation and behaviour prediction can help to personalise communication, and demand forecasting helps those working in operational roles to better plan for the future.

Stephen Allwright

Stephen Allwright

I'm a Data Scientist currently working for Oda, an online grocery retailer, in Oslo, Norway. These posts are my way of sharing some of the tips and tricks I've picked up along the way.
Oslo, Norway