The world of Amazon is vast and intricate, and a little unnoticed detail can spell the difference between profit and shortfall. Many Amazon vendors tend to operate in quick bursts and short terms because it’s difficult to look further out. Forecasting demand doesn’t seem as appealing as making actual sales, so it’s often overlooked.
However, properly anticipating needs is vital for your business, whether you’re a big brand or a local venture. After all, we’re in the era of TikTok, YouTube, and internet of things. Supply chains and consumer demands are ever-changing, often on short notice.
Amazon rolled out its Probability-Level Demand Forecasting as part of its Amazon Brand Analytics (previously Amazon Retail Analytics on Amazon Vendor Central) to help sellers and vendors make informed decisions about their products, especially in inventory demand. While this creates opportunities for vendors, it also compels them to make critical, and sometimes drastic, decisions that will directly affect their bottom line.
Let’s take a refresher about Amazon’s Probability-Level Demand Forecasting — its rules of thumb, how it affects vendors, the challenges to overcome, and where analytics can fill the gaps.
What is Amazon’s Probability-Level Demand Forecasting?
Probability Level (P-Level) Demand Forecasting is a forecasting model that enables vendors to see the inventory levels they need to set in the near future. It’s a percentage of probability in terms of weekly customer demand as calculated by Amazon.
There are four probabilities in the forecast, which include:
P70. Amazon estimates a 70% probability that weekly consumer demand will be at or below the number of units shown. There’s a 30% chance they’ll buy more.
P80. There is an 80% probability that Amazon will purchase the level of demand indicated (or less), and a 20% chance Amazon will buy more.
P90. There’s a 90% probability that weekly consumer demand will be at the level indicated, and 10% chance Amazon will buy more.
Mean. This is used for vendors that import or whose manufacturing and production are overseas and need a five- or six-week waiting period to deliver products. In most cases, the probability percentage is around 50 – 60%.
The ideal P-Level would depend on many factors, including the nature of your business and the products you sell. Opting for a P70, for example, gives you a buffer against stockout, but you could run a bigger risk of chargebacks. You may also go with a P90 forecast to ensure the orders will always be fulfilled, but too much inventory also has its drawbacks, e.g., storage costs, expiring products, etc.
So how do all these relate to sales?
In the US, over two-thirds of Americans shop online. Unprecedented events like the COVID-19 pandemic pushed these consumers to flock to Amazon for their household staples. The surge left many sellers out of stock and miss out on sales while competitors take over and Amazon scrambles to sort out its disrupted supply chain.
Interestingly, many retailers and manufacturers in China responded by anticipating hoarding and stockouts and shifted from traditionally offline distribution channels to e-commerce platforms. Food and beverage producer Master Kong, for instance, regularly reviewed consumption habits and tracked other market dynamics to forecast demand. This enabled their supply chain to recover by 50% just weeks after the outbreak.
Anticipating demand based on empty shelves isn’t going to boost your profitability. By expecting what, where, and when the demand will be, businesses can properly allocate their resources and better manage risks.
P-Level Demand Forecasting: Caveats and impact to vendors
Perhaps the biggest — but arguably the subtlest — disclaimer about the P-Level demand forecasting is that the probability levels are simply forecasts, not guarantees. The risk tolerance is on the vendor. Even if Amazon provides a demand forecast, it is not a promise of purchase order.
The whole forecast is a rolling 26-week estimate. Amazon also intentionally stopped projecting safety stocks to make their operations more efficient. Safety stocks are also no longer calculated into the demand forecast. Vendor lead times (VLTs) can be estimated based on the vendor’s own needs.
What do all of these mean to vendors?
For one, more responsibilities are shifted onto the vendor. It is up to the vendor to decide their tolerance for carrying inventory based on needs estimated by Amazon.
With safety stocks no longer in the forecast’s equation, Amazon expects vendors to deliver the products to their warehouses or distribution centers on time, at the exact spot, in their exact quantities and mint conditions.
Vendors may also be compelled to straddle between prioritizing operational efficiencies, preventing loss of sales, and mitigating chargebacks. VLTs, for instance, can be a double-edged sword. On one hand, the vendor can freely determine a longer lead time to secure against chargebacks. However, Amazon also figures in VLTs when calculating actual purchase orders. VLTs also play a part when Amazon analyzes vendors to work with, so they may end up buying products from competitors if you have a longer VLT.
You can also opt for the mean forecast. It could be a cheaper option, but it may be less efficient for the vendor. Amazon also doesn’t offer this to all vendors.
There are a lot more variables that influence how Amazon issues purchase orders — VLTs, cost and profitability to Amazon, and availability of products from other suppliers or distributors, just to name a few. Mismanaging them can result in a mismatch between the demand forecast and actual purchase orders.
Challenges in implementing Probability-Level Demand Forecasting
Amazon’s shift towards vendor-driven actions means more freedom to use different tools, technologies, and processes at the vendor’s disposal, but there are also more responsibilities to shoulder.
Amazon’s demand forecasts are great starting points, but they can be shortsighted. If you only want to know how many products you need to carry to meet your sales for a couple of days or weeks, then it’s probably fine. Note that Amazon’s P-Level demand forecast is based on consumer demand. Seasonality, promotional events, and advertising — which directly affect how you carry inventory — are outside its purview.
The demand forecast is also an estimate for weeks at most. If you’re a bigger retailer or a brand whose operations and manufacturing plan further ahead, it will be difficult to generate year-on-year comparisons and long-term estimates.
While you are visible to your own inventory, you may not know how long you need to have the stock in their warehouses or fulfillment centers to be able to achieve sales that Amazon forecasts. Transit times are not factored in. If you’re shipping from a different state (or country) or through other logistics providers, you need to know exactly when it will be received by Amazon so they can be available to sell. The process can be particularly complicated if you’re opting for the mean forecast.
There’s also the challenge of integration, a common woe when working with data. Technologies that businesses are already using may not be compatible with Amazon’s main forecasting systems and other tools. Amazon uses a rather convoluted formula that incorporates variables that may not be readily available or visible to vendors. Using Amazon’s data through your own systems and tools can be tricky, as these unseen variables can be crucial in the bigger picture but may not be accounted for by your system. There are also the hurdles of consolidating your own disparate and fragmented data from different sources that also need to be filtered and formatted to make them usable.
Because Amazon’s demand forecasts are not a promise of purchase, vendors need to set up business processes that will cushion risks. These data-driven processes should empower your overall strategy. They should enable the vendor to track demand and purchase, steer clear of chargebacks, and minimize situations that marks down prices in your inventory. They should also enable you to leverage alternatives like third-party and drop shipping fulfillments when you encounter stockouts or hiccups in your logistics. Vendors unfortunately don’t have full visibility into data that can help drive these strategies.
How analytics can help make smarter inventory decisions
Almost every company can access data that it needs to generate better and more accurate demand forecasts. While you may be sitting in a goldmine of your own data, however, it can be grueling and befuddling to sift through them, let alone use them for your business strategy.
Analytics can help fill the gaps so you can draw and see the bigger picture.
Starbucks is a good example. Beyond calculating inventories, they gather data points about their shops, suppliers, and other dynamics that will foretell changes in customer behavior and demand. Even the number of caramel syrup bottles in each of their stores is a helpful piece of information. With the help of machine learning, they aggregate, cleanse, and categorize these data points, then use them to ensure a reliable supply chain from which they can craft a steady dose of their coffee to their customers. By discovering patterns and interdependencies in their data, for instance, they can even concoct new recipes based on customer demand.
By harnessing analytics, vendors can plan for products with shorter life cycles, volatile demand, and changing market conditions. Fashion brands, for example, generally update inventories as fast as a couple of weeks, with each piece of apparel having their own demand. Machine learning can be leveraged to track data on sales, designs, and the latest fashion trends to predict what collection to put out for the next season.
As a vendor on Amazon, analytics can help you choose which probability level suits your needs. If you’re on P70, analytics can eliminate the guesswork in the remaining 30%, all while having visibility to hidden variables and other seemingly trivial data that could, in fact, make or break a sale.
With more robust data analytics, you can better navigate your way around the P-Level forecasts and gain a clearer view of your responsibilities as a vendor on Amazon. In turn, you can keep up with customer demand while also solving potential issues before they even happen.
Lingaro harnesses the power of data to drive transformative change and fuel business growth. Lingaro’s demand forecasting capabilities incorporate artificial intelligence (AI) and machine learning (ML) to automate workflows, aggregate isolated datasets, and seamlessly integrate other data sources to deliver bespoke, accurate, and intelligent demand forecasts that will optimize production, inventories, logistics, and finances.
Lingaro also provides supply chain analytics to help businesses gain full visibility across their supply chain, make informed decisions, and achieve operational excellence. By optimizing processes and harnessing cutting-edge technologies, Lingaro delivers tailored AI- and ML-powered solutions that map new opportunities and improve key areas in the supply chain — from demand forecasting, logistics networks, and warehousing to inventory management.