Marking down, or discounting, a product when it is towards the end of its active life (shelf-life, code-life etc) is common practice in grocery retail. The objectives of markdowns are to maximise the revenue from the inventory and minimise the amount of inventory remaining at the end of the active life. While this article focusses on the example of grocery retail, many of the principles are wholly relevant to non-grocery retail and consumer products without expiry dates. Just take the end of line or delist date instead of the expiry date.
With thousands, or tens of thousands, of SKUs in a typical grocery store, it is not uncommon to see expired products left on shelves for a day or two. It is not practical for staff to check all SKUs for expiring inventory every day. To avoid this situation, reduce time spent checking for expiring inventory and increase the revenue from expiring inventory, retailers should proactively predict when products will need marking down. This requires visibility of the quantity of each product on shelves, and what expiry dates these units have.
Retailers typically lose sight of product expiry dates when inventory enters the store. To solve this problem, Purpose AI have developed a machine learning model which predicts with >90% accuracy the breakdown of expiry dates purchased each day for a particular product. There are two key points to note for this type of solution. Firstly, an organisation needs to collect data on consumer behaviour to use for the machine learning model. In our most recent project, we collected data for 10 stores over 7 days. This equated to additional work equivalent to 2 extra shifts per store. The second key point is that for some categories, the actual breakdown of expiry dates needs to be manually reviewed every 1-2 weeks.
Despite >90% accuracy, there are scenarios where the actual breakdown of expiry dates differs from the predicted breakdown more significantly. This is usually caused by atypical shelf-stacking practices, for example if the oldest expiry dates are not put at the back of the shelf. These scenarios result in different buying behaviour, and cause the predicted vs actual split of expiry dates on shelf to move out of alignment. This can propagate differences between actual data vs the predictions because the prediction is being based on different assumptions to what is on shelf. This is primarily a consideration for categories where there is the strongest consumer preference for longer expiry dates, such as bread, milk and fruit and can be corrected with manual expiry date reviews every 1-2 weeks for these categories.
This machine learning technique outputs daily, or even real time predictions of the inventory and expiry dates on shelf. It can also predict the closing inventory for the following day using the store demand forecast. This prediction highlights any inventory which is predicted to expire unsold and should therefore be marked down.
Before we discuss optimising markdowns, it is worth calling out that this technique of predicting the expiry dates of inventory on shelf is not just beneficial for markdowns. It can also be used to identify other actions that can avoid expiring excess inventory. These actions can be taken further before the expiry date to achieve more revenue from the inventory. There are more details on this in our Preventative Actions for Retail Food Waste Article.
With a prediction of inventory that will be expiring and unsold, store staff have a list of specific SKUs to review for markdowns. There are two main parameters to optimise for markdowns: when to markdown, and by how much. Some retailers have a set time for marking down products each day, others have multiple set times while some markdown on an ad hoc basis. To manage this, we run a range of scenarios covering the different timing and discount parameters available. We then identify the scenario with the best performance in terms of revenue, margin or a combination of the two. It is also possible to incorporate a penalty for any unsold inventory to cover the cost of disposal and the internal waste price (same principle as a carbon price). It should be noted that in some scenarios, we have found that marking down inventory the day before it expires has also been found to be the optimum action to take.
The optimisation of markdowns requires both process and technology. Firstly, it is important to have an owner of the markdown process in each store on each day. Secondly the owner of markdowns needs the technology to enable an efficient and accurate execution. There needs to be a clear list of markdowns for review, a simple way for the employee to confirm in the system how many expiring units are on shelf and a clear markdown price to execute.
While the calculation and logic are the same in the background, it is possible to implement the system in different formats. Web or mobile applications are the most flexible for use on different types of devices. Other organisations prefer to use their existing tech stack and the technology can be integrated into Microsoft Power Apps or Power BI, which also operate as mobile applications or websites. As is often the case, technology is not the limiting factor, and the choice is driven by preferred solution architecture.
A final implementation factor that is worth considering is the step between optimising markdown parameters and executing a new barcode on the product. This process is prone to human error whenever an employee has to take the markdown price from one system to another. So whatever the chosen technology and architecture, it should be integrated with the labelling hardware to automatically print the new barcodes. This will both improve accuracy and streamline the process, making it less manual for the employee.
While marking inventory down close to expiry is common practice, it is often not optimised and consumes considerable time in identifying products to markdown. By predicting which SKUs should be checked for markdowns, you can save hundreds of hours each month for a medium or large store. With machine learning, it is possible to increase the cost recovery from expiring inventory by 30% and reduce waste by 15+%. This benefits financial performance, sustainability and employee satisfaction, a true win-win-win.
Purpose AI is an AI and Analytics Consultancy delivering data-driven solutions for complex Supply Chain and Commercial decision making. We work across the Consumer Products and Retail Industries and have worked with many Top 20 CPGs and Retailers in Europe and North America. Our USP is that we have developed industrialised solutions to common industry problems, reducing the time and costs to deliver value. But unlike SaaS providers, there is no licensing costs, we customise our solutions to your requirements and embed a strong business process to ensure they deliver value.