CPG Companies: Improve Demand Forecasting to Boost Sales on Amazon

For consumer packaged goods (CPG) companies that sell finished goods on Amazon, improving demand forecasting can be a large step towards maximizing revenue.

CPG companies must accurately forecast and anticipate variable product demand to provide sufficient time to manufacture and deliver products to Amazon’s fulfillment centers. Many CPG companies already use Amazon Vendor Central, yet some of them want to store and analyze their data with their business intelligence tool of choice. To assist CPG companies with their needs, AWS services help automatically ingest Amazon purchase order (PO) data and then use machine learning (ML) algorithms delivered by Amazon Forecast to generate forward-looking, fine-grained, time-series demand forecasts that can help CPG companies improve on-time, in-full (OTIF) performance. The architecture featured in this blog enables CPG customers to source, analyze, act on, and syndicate data directly in their own environment.

Selling Consumer Goods on Amazon

Companies can sell products on Amazon in two ways:

• First-party (1P) sales—Amazon purchases goods from a CPG manufacturer through POs and then warehouses product inventory in regional Amazon fulfillment centers according to expected consumer demand.
• Third-party (3P) sales—3P sellers can sell products directly to consumers on Amazon. The selling partner owns the inventory, sets retail prices, and manages inventory, shipping, and returns. Some 3P sellers choose to use Fulfillment by Amazon (FBA) to assist with warehouse management operations and instead use their time to focus on unique product innovation and growing their business.

Although some large sellers use both 1P and 3P models for different products or product lines, the key difference is who owns the inventory. But regardless of whether you’re a 1P or 3P vendor on Amazon, the goal is to serve the needs of consumers by ensuring desired products are available and in stock.

CPG Industry Challenge: Accurate Demand Forecasting to Meet OTIF Deliveries

Many large CPG companies are 1P vendors on Amazon, and Amazon electronically sends POs to manufacturers for quantities of different stock keeping units (SKUs). Once the CPG manufacturer accepts the PO, they are obligated to deliver the accepted quantity of assorted SKUs by specified dates to specific Amazon delivery points. In order to meet these obligations, companies need ample time to plan, manufacture, and stock products for delivery to Amazon, and they need reliable data insights—driven by sophisticated ML algorithms—to accurately forecast demand.

According to a recent McKinsey & Company article, “Online sales of consumer packaged goods have soared during the pandemic … and the trend shows no signs of fading. McKinsey’s consumer sentiment surveys reveal that US consumers plan to continue spending more of their money online even after the COVID-19 crisis subsides.” This is a signal that a focus on the management of operational performance to delivery points within the Amazon supply chain has become increasingly important.

Amazon: Refining Demand Forecasting

Amazon has been on a decade-plus long journey to develop a forecasting model that makes accurate decisions across diverse product categories. In the beginning, Amazon used rules-based statistical forecasts, which evolved to ML algorithms and eventually to deep learning algorithms that allow Amazon to predict variable consumer demand far more accurately.

AWS customers took notice and asked to purchase the same forecasting solution used at Amazon. As a result, AWS released Amazon Forecast, a fully managed cloud-based service that does not require deep expertise in time-series forecasting or ML.

Automate Data Analytics and Apply ML to Accurately Predict Variable Product Demand

Many 1P vendors manually collect forecasting data. The architectural framework shown below uses several underlying AWS services to automate the data collection process. It uses vended APIs to programmatically source historical and current PO and product catalog data, as well as diagnostic reports from Amazon Vendor Central, to understand actual demand. Automated integration with Amazon can help vendors improve and maintain their performance at scale and grow their businesses on Amazon. Then, using ML, the solution helps predict demand for variable vendor POs for the weeks ahead. Finally, instead of using point forecasts, vendors can match correct probabilistic forecasts to Class A, Class B, and Class C SKUs.

Figure 1: Reference architecture to automate data collection for Amazon selling partners

Get Started Today

As they accurately predict order demand week after week, CPG companies can start to improve outcomes at both the top and bottom lines.

This solution can help companies:

• Import Amazon PO data: More easily import Amazon PO data to better access, govern, and share data seamlessly across the organization.
• Increase forecast accuracy: Improve forecast accuracy using ML from Amazon Forecast to create a probabilistic forecast to predict demand for variable vendor POs weeks ahead of time.
• Track product change impacts: Track product weight or dimension changes to ensure they adhere to manufacturer standards.
• Reduce OTIF penalties: Improve OTIF metrics to enhance organic ranking on Amazon.com, reduce chargeback expenses, and help ensure products are in stock for consumers to buy.
• Ensure high velocity SKUs are in stock: Match correct probabilistic forecasts with SKU sales velocity to ensure key products will stay in stock and improve revenue.

If you’re ready to automate and improve demand forecasting, AWS is here to help you overcome this and other critical challenges to improve operations. Contact your AWS account team to get started today.