Improving On-Shelf Availability for Items With AI Out of Stock Modeling

One of the biggest problems CPG/FMCG manufacturers face in this fiercely competitive retail environment is on-shelf availability (OSA). Due to the ongoing pandemic's considerably higher demand volatility, inconsistent reorder points for vendor-managed supplies, bullwhip effect, and growing competition from private-label products in key product categories, which have shaken the category leadership for many established brands, the challenge's magnitude has increased significantly.

The need for demand sensing in retail on shelf availability has increased dramatically during the past 12 months. In order to better maintain optimal inventory across multiple echelons, maximize on-shelf availability, and meet their dedicated service levels without sapping too much working capital into the inventory and driving up their inventory carrying cost, organizations have experimented with machine learning and artificial intelligence (AI).

When they can't find the goods they want, 30% of customers move retailers, and 70% switch brands, according to studies. When products are out of stock, retailers lose $1 trillion in sales. Consumer confidence in the dependability of retailers and the availability of products is increased by high on-shelf availability. Retailers value CPG manufacturers or suppliers as well since they can assist them to increase on-shelf availability and minimise supply disruptions, which will enhance the customer experience and increase sales.

According to an IRI survey, 20% of all out-of-stock situations get unaddressed for longer than three days. According to a research, the typical OOS rate is around 8%. Accordingly, one out of every 13 products is not available when a consumer wishes to buy it at the store. One of the biggest issues in retail is OOS, but there are ways to fix it with real-time data and analytics. 

Problems with manual scanning and conventional methods of data collection

To track on-shelf availability, store associates manually scan and gather data, typically at a certain time each day. If there are biases present that make the time period chosen unsuitable for inventory scanning procedures, those biases are magnified across the whole data set. For instance, the majority of manual scans are performed first thing in the morning following the conclusion of restocking procedures, creating a perfect representation of inventory supplies.

Additionally, manual scanning is a time-consuming, repetitive task that necessitates continuous, intense concentration for an extended amount of time. Low accuracy levels are a natural result of human mistake. Due to the ongoing man-hour expenditures and lost opportunities from not switching to a more precise and effective method that enables you to boost OSA, it is costly in the long term. It is unrealistic to ask for manual tracking across a large-scale retail format when SKUs can number in the hundreds of thousands.

Unlocking the potential of out-of-stock (OOS) modeling for businesses

Regular inventory counts and the use of point-of-sale (POS) and inventory management software data to track inventory are not practical due to inconsistent data entry and variables like shrinkage. Although out-of-stock modelling is the better option, inaccurate data can cause false positives and negatives, making it challenging to apply model predictions. High volumes of data must also be processed quickly enough to be put to use.

While the OSA solution concentrates on increasing sales through better stock availability on the shelves, the more comprehensive Retail Supply Chain Control Tower addresses a number of related merchandising issues, such as designing store inventory, replenishing stores effectively, designing the network of stores for omnichannel operations, etc. Given how prevalent this issue is in the retail industry, we jumped at the chance to accept their offer.

Examining the occurrence of OSA issues in the historical data is the first step in resolving them. Past events indicate structural difficulties with suppliers and internal procedures that, if left unaddressed, will continue to be a problem.

The OSA solution makes use of past data to locate systemic problems with suppliers and internal procedures that have a negative influence on income. It addresses a number of issues, such as:

Phantom inventory - Phantom inventory is when the items reported do not match the items really in stock.

Retailers mistakenly assume they have more units on hand than they actually do due to poor tracking of replenishment units, unreported or undetected shrinkage, and out-of-band procedures, as well as irregular and occasionally inaccurate inventory counts. This phantom inventory could cause an out-of-stock situation if it is substantial enough to delay or even block the ordering of replenishment units.

Safety stock violations - When the bar for triggering replenishment orders is either too high or too low, safety stock violations happen.

Most businesses set a minimum inventory level for a particular product below which replenishment orders are placed. If the bar is set too low, insufficient lead times or even slight supply chain delays could result in an out-of-stock situation while new units are being replenished. 

On the other hand, if they are set too high, businesses run the danger of having excess inventory that could go bad, get damaged or stolen, or take up space and money that could be used more effectively elsewhere. Effective inventory management depends on determining the appropriate safety stock level for a product in a particular location.

Zero-sales events - Understanding and optimising for these criteria helps to drive stock availability because zero-sales events—periods when no units of a product are sold—help to drive stock availability.

The two most frequent reasons for out-of-stocks are phantom inventory and safety stock violations. No of the reason, out-of-stock situations show up when no units of a product are sold.

Zero sales events may not always indicate an out-of-stock issue. Some things don't sell every day, and for some products that move slowly, it may take several days for 0 units to sell while the product is still fully stocked. Understanding the likelihood that at least one unit of a product will sell on a particular day can help you analyse zero-sales events at the item level. After that, you may define a cumulative probability threshold for days that show consecutive zero-sales.

Situations when merchandise is available but operating poorly because of a variety of variables, such as inadequate inventory management techniques.

Recognizing situations in which things are theoretically available for sale but underperforming due to subpar inventory management techniques is as crucial to comprehending situations in which items are not in-stock. These merchandising issues could be brought on by poorly placed store displays, products that are stocked deep within shelves, a slow product transfer from the backroom to the shelves, or any number of other situations where there is enough inventory to meet demand but items are difficult for customers to see or access.