AWS for Industries

Powering solutions for retailers at NRF 2023: Amazon Forecast and AWS Supply Chain

We’ve talked about the exciting immersive commerce and in-store optimization ideas coming out of NRF 2023, but what about the stuff behind the “retail curtain”?

At this year’s big show, Doug Tiffan, head of worldwide solution strategy Apparel & Fashion, led an informative discussion talking about how and why AWS services and applications like:

  • Amazon Forecast, a time-series forecasting service based on machine learning (ML);
  • Amazon SageMaker Canvas, which provides business analysts with a visual interface;
  • AWS Supply Chain, a cloud application that unifies data and provides ML-powered actionable insights,

are helping customers power their retail solutions.

Let’s take a closer look at the background and benefits of these AWS solutions.

Forecasting with AWS: Retail-optimized forecasting

AWS offers low-code and no-code (LCNC) ML-based forecasting services to support retailers. Services such as Amazon Forecast and Amazon SageMaker Canvas allow organizations to select the forecasting framework that makes the most sense for their specific needs and use cases.

Amazon Forecast is a fully managed time-series forecasting service based on ML. With a low-code interface, this service puts the power of Amazon’s deep experience in ML-based forecasting into the hands of everyday developers. Retailers like More Retail and The Very Group, as well as partners like Anaplan, are using Amazon Forecast to power their solutions and support operations.

With Amazon Forecast, you can create forecasts based on a custom horizon, use price promotions, advertisement data, special events, and even weather to generate forecasts. Want to generate a forecast for items you’ve never sold before? The deep-learning models in Amazon Forecast are equipped to learn from patterns in the data and will use item metadata (characteristics shared between items) to generate forecasts for newly introduced items. In fact, at re:Invent 2022, AWS announced a new cold start feature that produces more accurate forecasts for items with no historical data.

Compared to traditional methods, Amazon Forecast can and should be used for several reasons:

  1. allows the use of large amounts of data that traditional models can’t use or evaluate together
  2. uses contextual data, such as promotions or item classification, and quantifies what data is impacting your forecast by showing you explainability metrics
  3. handles special scenarios like new product introductions
  4. picks up on subtle relationships between large amounts of items to better understand how one item impacts another
  5. combines multiple models, producing more accurate forecasts than any single model could alone
  6. provides probability-based forecasts, allowing you to select the appropriate service level for different inventory items

Amazon Forecast uses a combination of statistical and ML models to produce an optimal forecast at the item level. What this means is that all the models are considered for each item, not the whole dataset. From this, the best combination is selected. This ensemble modeling produces forecasts that are up to 40 percent more accurate than statistical models. In addition, Amazon Forecast provides a full set of accuracy measures (MAPE, WAPE, and wQL), as well as back-test metrics to allow users to compute their own customized accuracy metrics.

At NRF 2023, we presented Amazon SageMaker Canvas, a no-code ML application that allows business analysts to quickly create ML-based forecasts without writing a line of code or having any ML experience. Amazon SageMaker Canvas provides a graphical interface that allows users to import data from multiple sources, as well as build, train, and evaluate models using the same advanced ML algorithms as Amazon Forecast. Using Amazon SageMaker Canvas, analysts can rapidly build forecasts and “what if” scenarios that allow them to quantify the potential impact of different prices and promotions on demand.

By using Amazon Forecast and Amazon SageMaker Canvas, you can optimize service levels. This is because AWS LCNC ML forecasting services are:

  1. highly accurate: ML models can be up to 50 percent more accurate than traditional statistical methods. Unlike traditional methods, LCNC ML combines multiple deep learning and statistical algorithms for increased accuracy.
  2. no ML experience required: Do all the heavy lifting of building, training, and deploying custom models at scale without the need for ML experience.
  3. easy to integrate: Integrate into existing data lake, inventory, ordering, and supply chain systems.
  4. able to quickly iterate, explore, prepare data, and onboard for ML forecasting: No need to bring an entire data architecture onto AWS.

After employing forecasting capabilities from AWS, More Retail, The Very Group, and a large convenience store chain all saw improved metrics and financial results. The Very Group, the United Kingdom’s largest integrated digital retailer and financial provider, used AWS forecasting and artificial intelligence (AI) and ML solutions to accelerate and build new retail demand forecasting capabilities. It’s now expanding the model to other business areas, iterating with additional use cases across the organization and adding newer data to Amazon Forecast to continuously improve the model accuracy.

See what Amazon Forecast and Amazon SageMaker Canvas can do for you.

No more heavy lifting with AWS Supply Chain

AWS Supply Chain is a cloud application that mitigates risk and lowers cost through unified data, ML-powered actionable insights, and built-in contextual collaboration. Unlike Amazon Supply Chain, which focuses on the physical movement of freight, AWS Supply Chain operates in the digital world. This application easily connects to your existing enterprise resource planning (ERP) and supply chain management systems, without replatforming, up-front licensing fees, or long-term contracts.

Beyond serving inventory leaders, supply leaders, data stewards, and senior executives, AWS Supply Chain targets customers that are tier-1 enterprises operating in North America and Europe, manage inventory across multiple locations (for example, warehouses, distribution centers, stores, and manufacturing plants), and are disconnected ERP systems, or are in the automotive, chemicals, consumer products, aerospace, manufacturing, medical devices, and retail/wholesale industries.

AWS Supply Chain helps mitigate risks and lower costs by solving five problems:

  1. inaccurate or incomplete inventory visibility across multiple locations and sales channels: AWS Supply Chain’s inventory visibility shows how different supply chain nodes and facilities connect, displays current inventory levels, and highlights both the expected and actual flows of inventory between each facility.
  2. incorrect supplier lead time forecasts: AWS Supply Chain detects anomalies in supplier lead times and provides updated lead time forecasts to supply planners, improving lead time predictability for future orders.
  3. low fill rates: AWS Supply Chain Insights projects future inventory levels and risks for each facility by analyzing on-hand inventory, open customer orders, and current supplier lead times, sending alerts to managers if any facility is trending toward an excess or shortage of inventory as conditions change.
  4. high time, integration effort, and cost needed to adopt new supply chain solutions: AWS Supply Chain’s data auto-association tools use AI to do the heavy lifting of data mapping exercises so that customers can begin generating insights in days and not months, as is typical when adopting new supply chain solutions.
  5. predominately manual demand planning processes, or those relying on older statistical planning technologies: AWS Supply Chain Demand Planning saves customers time by automating formerly manual processes and improves forecast accuracy through the application of ML technology that uses learnings from

Considering AWS Supply Chain? The application provides unique value to customers by

  • helping improve planning assumptions by detecting anomalies in supplier lead times
  • looking for trends and anomalies and providing actionable recommendations to respond to demand and supply volatility (while customers continue to use their existing systems for implementation)
  • generating insights based on ML that process customers’ existing supply chain data to highlight emerging trends and risks that leaders are either unaware of or lack the information needed to chart a path forward
  • removing the undifferentiated heavy lifting needed to integrate with existing ERP and supply chain systems
    • This makes it easier for customers to adopt new supply chain solutions.

Discover the possibilities here.

Whether using Amazon Forecast, AWS Supply Chain, or both, AWS is here to help accelerate your retail transformation. Learn more at, and contact AWS today to get started.