ML-powered Demand Forecasting Solution
Blue Bottle Coffee increases ordering accuracy, cuts food waste through ML-driven demand forecasting
Blue Bottle Coffee (BBC) is a coffee roaster and retailer with an international network of cafes in the US and Asia that is dedicated to connecting the world to delicious coffee because delicious coffee makes life more beautiful.
Blue Bottle cafes order pastries several times a week to make sure customers have a sufficient and fresh supply. Because order estimates can be inaccurate, cafes either under- or over-order pastry, which leads to sell-outs and waste. Blue Bottle needed a new predictive ordering solution to reduce food waste.
Blue Bottle’s and Provectus’ engineering teams built an ML-powered predictive ordering system to generate accurate future pastry demand. Using inventory data, historical sales and growth projections, the solution suggests how many of each pastry each cafe should order daily or weekly.
Provectus has developed a predictive ordering system that accurately forecasts pastry order quantities to café leaders. By closely monitoring and following the predicted orders, Blue Bottle cafes were able to improve pastry ordering by 8% in July compared to the previous month.
Blue Bottle Coffee, a global leader in third-wave coffee, was looking to optimize the sales process for a variety of pastry items like croissants, cookies, and waffles across its chain of cafes.
Historically, all cafe leaders at Blue Bottle were tasked with ordering pastries several times each week to make sure that customers always have a sufficient and fresh supply. To satisfy demand for the coming week, they manually guesstimated orders by taking into account historical sales data, current inventory, and growth projections.
The manual ordering system worked when Blue Bottle only had a handful of cafes in the Bay Area. But with 70+ cafes across the globe, they needed a more scalable and precise solution to generate accurate future pastry demand.
Inaccurate orders could lead to pastries either prematurely selling out if under-ordered or contributing to waste if over-ordered.
- Underordering left some of Blue Bottle’s customers unhappy, increasing churn
- Overordering contributed to food waste and resulted in loss of profit
- Suboptimal utilization of pastry undermined Blue Bottle’s bottom line
Blue Bottle Coffee joined forces with Artificial Intelligence consultancy Provectus to develop a demand forecasting solution to increase pastry ordering accuracy, reduce food waste, and meet Blue Bottle’s sustainability goals.
Provectus proposed to build a demand forecasting solution powered by Artificial Intelligence and Machine Learning, a predictive ordering system that could generate highly accurate pastry ordering forecasts by assessing historical sales data, current inventory, and growth projections.
In close partnership with Blue Bottle’s engineering team Provectus reviewed and enhanced their machine learning models. The data that Blue Bottle’s team planned to utilize to train the demand forecasting model was also accessed for analysis, to improve training efficiency, increase ordering accuracy on new data, and eliminate bias.
Provectus proceeded to orchestrate computational workflows and data processing pipelines to ensure faster and more efficient data/ML manipulations in the system. A special “black box” library for Blue Bottle’s engineers storing call methods to prepare data, train models, and generate forecasts was developed to help them create new DAGs quickly and easily.
A new prediction ordering system consisting of three pipelines was built from scratch:
- Model Data Pipeline — to migrate data from Amazon Redshift to Amazon S3
- Training Pipeline — to train, enhance, and store the best performing demand forecasting models in Amazon S3
- Forecasting Pipeline — to generate demand predictions by using data stored in Amazon S3 and to push prediction results to Amazon S3 and Amazon RDS for reporting
In this ecosystem, ML models are trained and predictions are generated using Amazon SageMaker Training and Amazon SageMaker Batch Transform, based on data that cafe leaders upload to Amazon S3 once a week.
Blue Bottle Coffee’s team of developers received a separate Model Development Pipeline to test, compare, and select best performing ML models to deploy them in production with Amazon SageMaker.
All of those elements are connected to a user-friendly UI that allows developers to continue to improve prediction accuracy. It also allows Blue Bottle’s cafe leaders to check demand forecasts in the system and edit those forecasts manually if necessary.
This predictive ordering system built by Provectus has helped Blue Bottle accurately forecast pastry demand, which aligns with their zero waste goals and financial bottom line.
Blue Bottle Coffee has received an ML-powered demand forecasting solution to generate highly accurate order predictions for pastry.
By closely monitoring and following the predicted orders, Blue Bottle cafes were able to improve pastry ordering accuracy by 8% in July compared to the previous month.
More efficient and accurate ordering of pastry means less under- or overstock, which keeps both customers and cafe leaders happy. The first can enjoy a variety of delicious croissants, cookies, and waffles any time and day of the week. The latter improve food utilization, which means minimum stockouts, less waste, and cost savings.
Cafe leaders can quickly access the predictive ordering system to check the latest demand suggestions and order predictions. No longer do they have to spend hours guesstimating how many items of pastry to order for a given day or week.
The scalable and precise predictive ordering system has helped Blue Bottle to accurately forecast demand, cut waste, and improve food utilization. Overstock waste is reduced, resulting in cost savings and driving down the company’s ecological footprint. It helps keep customers happy and satisfied, enabling Blue Bottle to grow by spending less, and forging a direct path to further growth in new markets.