AI is the latest massively disruptive technology making waves for businesses everywhere. But it’s not quite new to retailers. They’ve been leveraging AI for a number of years in the form of chatbots, product recommendation engines, self-checkouts and many other use cases.
However, with the advent of generative AI, things are changing quickly today. Even many tech-savvy retailers are struggling to keep up. Generative AI is giving birth to exciting new use cases, but as with any new technology, there are serious questions about how to properly wield it. Using it properly means significant business value, but misusing it can mean wasted time and money.
When most people hear the words “generative AI,” they immediately equate it with Large Language Models (LLMs). But LLMs aren’t the only game in town. Large Graphical Models (LGMs) are another gen AI technology – and they’re particularly well-suited to enable retailers to uncover patterns, predict granular business outcomes and make good decisions as a result. While LLMs are ideal for text-based use cases like customer support, personalization and content generation, LGMs are best for analyzing structured data so enterprises can forecast broader trends across their organization.
LGMs can be useful in a variety of retail applications today, and merchandising is one of the best examples.
Advertisement
The Rising Role of Retail Merchandising
AI and machine learning pilots and proofs-of-concept (POCs) are now pervasive in modern retail, aiming to build on the investments made in automation, supply chain optimization, mobile shopping apps, customer data platforms, ecommerce platforms and various analytics solutions that have brought the industry forward. Despite these advancements, many retailers still struggle in one key area: merchandising.
Merchandising plays an essential role for modern retailers. Broadly speaking, merchandising refers to all efforts and strategies to market and sell products to customers once they’ve entered a store (either brick-and-mortar or an ecommerce site). Ultimately, the goal of merchandising is to narrow or eliminate the gap between shelf and customer, but stockouts, overstocking and missed product opportunities persist.
For traditional retail, merchandising strategies cover questions like:
- How should I organize products?
- Which items should I put on endcaps or on display in the front of the store?
- How much shelf space should I dedicate to a given section?
Of course, answering these questions is tricky. These are complex problems dependent on many variables.
LGMs Take the Guesswork out of Merchandising
If retailers are able to forecast trends and detect patterns, they can answer the type of questions above, master merchandising and ultimately drive greater profits. LGMs were built for exactly that type of forecasting.
Predicting the impact of merchandising efforts on a granular level involves the modelling of tabular time-series data. This type of data refers to data that is kept in databases and has a specific time associated with it. In a retail setting, this would include point of sale data (any information collected during a customer transaction – i.e., SKUs sold, total transaction cost, discounts applied, etc.). Analyzing this data in detail, along with historical sales data and external data sources like market trends and social media sentiment, unlocks insights that tell retailers how merchandising strategies are performing.
Large Graphical Models are designed specifically for tabular time series data. They capture relationships across many dimensions, making them ideal for forecasting and modeling data of all kinds, and especially effective for tackling highly complex multivariate scenarios like inventory planning, pricing strategies and product placement that typically characterize merchandising efforts.
LGM technology can simulate various scenarios to test the effectiveness of different merchandising strategies, all while gaining insights into the business impact of those strategies across key drivers such as cost, inventory, labor and suppliers.
For instance, retailers can use LGMs to model and predict merchandising outcomes like, “How will replacing shaving cream products with deodorant products on an endcap impact sales of both and total revenue for the store?”
Making Better Decisions in a Competitive Industry
Retail is a cutthroat industry with razor-thin margins. Given this context, precision can make the difference between rapid growth and sudden bankruptcy. LGMs equip retailers with the insights to closely shape their merchandising efforts.
Devavrat Shah is the Co-founder and CEO of Ikigai Labs. He successfully combines academia and entrepreneurship. Shah is a professor and a director of Statistics and Data Science Center at MIT. He previously co-founded Celect, a predictive analytics platform for retailers, which he sold to Nike. Shah holds a Bachelor and PhD in Computer Science from Indian Institute of Technology and Stanford University, respectively.
