A scientific approach with AI/ML modeling to improve planograms in the retail and consumer products industry

A blog post by Chida Sadayappan, lead specialist, cloud, data, and machine learning, Deloitte Consulting LLP and Dinesh Kumar, principal machine learning and deep learning engineer, Deloitte Consulting LLP

A planogram is a model that specifies exactly how much product should be displayed on store shelves to maximize sales and enhance the customer experience. Traditionally, the retail and consumer product industries relied heavily on their past statistics for forecasting and used heuristic methods and human judgment to perform a planogram product assortment, resulting in lost sales due to out-of-stock product(s), generating greater waste due to decayed product(s), and entailing high service levels in stocking up merchandise. Retail and consumer product companies have begun to realize that the traditional approach has its limitations and needs to be reimagined in a rapidly evolving and highly competitive market. That said, artificial intelligence (AI) and machine learning (ML) have started playing a critical role in the planogram assortment by helping to rank and recommend the products to maximize sales.

The three key aspects that help to build a successful ML-based planogram assortment are:

1. Do we have historical sales data available in an accessible state to adopt for AI/ML?
2. Should we build or buy an AI/ML-based planogram?
   
3. How do we ensure the successful adoption of the ML-based planogram?
   

Do we have historical sales data available in an accessible state to adopt AI / ML?

Before taking the journey toward implementing AI/ML on a large scale for a planogram assortment recommendation, retail and consumer product companies need to consider two things:

1. Deriving insights from data: Historical sales performance data is a key element in building an AI/ML-based planogram recommendation. The most important consideration, therefore, is to focus on building a centralized data repository to make data available in a public cloud platform, along with other historical sales signal data such as promotions, discounts, and demographics, among others. Historical sales data, combined with external sources such as weather, holidays, etc., will help predict and recommend the ideal planogram assortment by uncovering past sales patterns.
2. Revamping legacy systems and human judgment with data-focused strategies: Though traditional forecasting techniques and predictive models based on human judgment have limitations, and tightly coupled legacy systems make it difficult to adopt AI/ML without rethinking and strategizing the tech spend, it is still unrealistic to replace these legacy systems in one swoop. Instead, the industry should implement a phased approach, starting with better data management and governance; rationalizing IT architecture; adopting cloud to gain better computing power to enable AI/ML; and introducing MLOps and DevSecOps to automate the development process for continuous integration, continuous delivery, innovation, and monitoring.

Should we build or buy?

After overcoming initial hurdles and deciding to move ahead with an AI/ML-based planogram solution, it’s time to answer the million-dollar question: build or buy?

Build: Although there are commercial off-the-shelf AI/ML-based planogram solutions available in the market, it is important to evaluate the solution and make the right assessment before deciding. Key drivers to help decide may include:

• Unique factors specific to the industry, such as the mix of existing and new products for assortment changing very frequently, new or innovative products introduced in stores almost every month, and costs and complexity associated with integrating an AI/ML product with a custom-built visual assortment mobile application or other prescriber applications.
• Gaining full ownership of the code and model is critical to avoid compromising trade secrets and retaining a competitive edge.
  
• Industry is making a long-term commitment and perceives deployment of AI/ML technology as a differentiator.
  
• To an extent the pros of building and operationalizing a custom solution using MLOps outweigh the cons of buying a solution and customizing it to meet current needs, considering the frequency of probable retraining of the models.
  

If it is a consumer products company or a manufacturer is looking for the ML based planogram assortment solution, then custom-built solutions may work better because of the constraints and limitations they will have to deal with at physical store locations.

Buy: For small and midsize retailers, it is wise to buy a commercial planogram solution instead of building, as these applications will deliver quick value and can take advantage of pretrained models built by the product vendor using real-world data, resulting in higher accuracy. No time is wasted by investing in hiring and training the right talent and supporting infrastructure. Even if small and midsize retailers choose to build their own AI with open-source tools, it will entail significant monetary investment and may take months to train ML algorithms to do what most vendors have already achieved.

Commercial off-the-shelf AI planogram solutions provide advantages by helping understand localized customer behavior patterns and operational processes, thus enabling a data-driven decision-making process. They also help reduce time spent on cleansing data, smoothing production issues, and avoiding reinventing the wheel while deploying models daily or building in documentation and best practices to enable process automation.

How do we help ensure successful adoption of an ML-based planogram?

All the excitement of implementing an AI/ML-based planogram will be short-lived if companies do not establish proper evaluation criteria, consult subject-matter experts for their knowledge, train field staff to adopt and use the ML recommendations, and—most importantly—integrate the output of the planogram solution with the consumption system or application.

We can help ensure readiness for adoption of AI/ML-based planograms by: 

• Defining appropriate metrics for evaluation: AI/ML planogram stakeholders play a significant role in ensuring the success of AI/ML-based planograms. Throughout the journey, from initial conceptualization to building a proof of concept (POC) and the rollout of the solution into production, their expertise is needed to define objectives and key performance indicators and evaluate the AI/ML-based recommendations against them. To measure the success of recommendations, it is essential to evaluate measure the lift in sales in stores using AI/ML recommendations versus that of stores not using the planogram and compare growth rates from the prior week and the same week last year as well as measuring the reduction in logistics-related expenditures.
• Test and learn: The ML planogram initiative’s stakeholders must communicate and explain how the ML-based recommendation is key for business growth and is a more scientific way of executing planogram assortments to maximize sales, optimize logistics, and align with overall organizational strategy. A test-and-learn activity can be conducted in bite-size steps to measure and evaluate against the KPIs defined by picking a handful of test stores. ML product recommendations for the selected stores should be made available via the integrated mobile delivery and schematic app and could be consumed by in-store employees for executing a planogram assortment. The test-and-learn activity can be carried out over a few weeks to a few months, depending on the defined objectives and goals. After a successful test-and-learn activity milestone is reached, a decision should be made to roll out the solution across all stores.
• Consider leveraging ethics frameworks: An ethics framework such as Deloitte’s Trustworthy AI, comprising six dimensions for organizations, needs to be factored in when designing, developing, deploying, and operating AI systems. The framework manages common risks and challenges related to AI ethics and governance.

To summarize, AI/ML has come a long way to help retail and consumer product companies gain data-driven insights, enhance sales, stay ahead of their competitors, and, more importantly, better understand customer behavior, leading to better customer satisfaction.