Guidance for Demand Forecasting for Restaurants on AWS
Overview
How it works
This reference architecture showcases an end-to-end pipeline to deliver restaurant order demand forecasts in a data format that non-technical users can update and consume.
Well-Architected Pillars
The architecture diagram above is an example of a Solution created with Well-Architected best practices in mind. To be fully Well-Architected, you should follow as many Well-Architected best practices as possible.
Operational Excellence
You can update changes (such as forecast frequency and forecast horizon) to prediction configurations for SageMaker Canvas to achieve required levels of granularity and explainability of forecasting outputs.
Security
The permissions for each user is controlled through AWS Identity and Access Management (IAM) roles. Additionally, Transfer Family integration with Amazon S3 server-side data encryption helps secure file transfers. Although the architecture is serverless, the Lambda components can run within your virtual private cloud (VPC) and be associated to IAM roles with minimal required permissions.
Reliability
This architecture will scale to meet demand based on the volume of data (such as data from restaurant receipts) you upload to Transfer Family. As you scale your workloads, consider opting for user-defined schedules for SageMaker Canvas models rather than re-forecasting and re-training models for every upload. Additionally, all components of this architecture are built on event-driven patterns, meaning the system will only run when an event or change occurs in Amazon S3.
Performance Efficiency
You can adjust data input into the architecture through direct integrations with your own systems using Lambda. You can add more data relevant to your business use case through Data Exchange and adjust the Glue Crawler to construct modified data sets with which to forecast. You can also adjust configurations for SageMaker Canvas predictions and training on an as-needed basis. QuickSight allows you to view and compare variations of forecasts, explainability data, and accuracy metrics in one place using an intuitive user interface.
Cost Optimization
Using SageMaker Canvas’ price-per-use approach, you can train a predictor for under $1 USD (assuming less than 3 hours training time) and produce 1,000 forecasts on updated data for $2 USD. The architecture also benefits from using Amazon S3 for cost-effective data storage. The price of individual architecture components can be isolated by re-using existing SageMaker Canvas predictors.
This Guidance offers a low start-up cost to trial forecasting for non-technical users that can be further customized with automation as your familiarity with AWS technology grows.
Sustainability
By default, the architecture’s resources are only activated when there are changes in Amazon S3 buckets. Additionally, by adopting a serverless architecture, you can scale based on usage so that you consume only required resources.
Disclaimer
Did you find what you were looking for today?
Let us know so we can improve the quality of the content on our pages