FROST - Sales Forecasting & Analysis



FROST is a multi-location dessert concept known primarily for its unique varieties of doughnuts. Featured in national magazines and the Food Network, FROST has built a valuable brand in the growing specialty bakery space. As Creative Director, I am responsible for the design, marketing, and end-to-end guest experience for the brand.

While not part of my official duties, my extensive technical experience led to the creation and implementation of a sophisticated production tracking system designed to reduce waste, cost-of-goods, and labor. It would increase overall business efficiency, improve guest experience, and ultimately increase revenue.

One of the most frustrating challenges in a bakery business is determining how much product to make each day. Unlike a typical restaurant, a bakery selling items with extremely limited shelf life must estimate daily demand and produce product accordingly. Too much product means end-of-day waste, and too little creates guest frustration. In both cases, revenue suffers. Imagine a restaurant that had to guess how many steaks it would sell in a day, and cook them in advance. What could be done to help predict sales of steaks for that day to inform the cooks on how many to make?

My solution sought to gather as much relevant data as possible to help inform these daily decisions and to enable kitchen managers to more confidently estimate production.




The first priority for this project was to determine what data would be relevant in identifying sales trends. We know based on historical sales data that certain days are busier than others - but why? It wasn’t enough to say that Saturdays are always busier than Mondays because, while mostly accurate, there are several other factors at play. Is Monday a school holiday? Is there a big game on Saturday that will affect foot traffic in the area? What effect does the weather have on daily sales? Interviews with staff and management, review of sales against historical data, and industry research led to a list of factors that could better inform decisions made in the kitchen.

Once these relevant data points were identified, the next question was how do we gather this information and present it in a meaningful way? With the help of a software engineer, we created a routine that would gather both live and historical sales data from our point-of-sale systems, pull weather reports, and add calendar information to identify holidays and local events. We also gave managers the ability to manually add information from the store level that may affect traffic, such as mall closures or road construction.

Once the system collected the information, each data point was assigned a value that would increase or decrease the recommended daily production in the form of a percentage. With an average baseline of 100%, a recommendation of 125% indicates that the kitchen should increase production by 25%. The algorithm generating these percentages is easily tuned by adjusting the weight of each data point as we learn and collect more information making the system more accurate over time.

The next consideration was creating use-cases and identifying who would use this information, in what context, and on what types of devices. Some users will only consume the information while others may contribute to it. Based on job responsibilities, I identified three primary user types, each with increasing responsibility and access to the system. Iterating and testing with end users along the way, I defined both desktop and tablet interfaces that made viewing and understanding the information easy by revealing only what is necessary for the intended user.




Administrators use a web-based desktop interface which grants them granular access to all aspects of the system. Designed for the most technical users, the interface focuses on sweeping, system-wide changes and the maintenance of data sources.

Kitchen and store managers use a tablet interface to access daily recommendation numbers along with additional context to improve decisionmaking, and the ability to adjust or even override system recommendations. These users also have access to basic reports and are able to navigate between past and future dates, as well as input additional information that may be useful to bakery staff.

Bakery staff use a resistive touchscreen designed for gloved use in a busy kitchen. Their interface displays basic production information and allows the user to navigate to past and future dates, as well as view any additional information added by their managers.

A system of this type is only as effective as the data that drives it. Initial implementation required frequent adjustments and estimates in cases where there wasn’t enough data to support accurate results. As historical data grew, the system became more accurate and required less intervention by users.




Initially the system required more “care and feeding” than expected. This led to frustration for users, however its value was quickly understood as less intervention was required and more accurate numbers were provided over time. It also increased manager confidence when making production decisions, and allowed them to schedule production days in advance, freeing up time for other tasks.

In its first year of implementation, the system reduced product waste by nearly 25%, and by its second year nearly 40%. Store revenues have also increased by 15% due to properly projecting required inventory, and this number is expected to grow as the system gets smarter.




In the future, it is intended that this system will connect with other facets of the business such as assisting buyers with setting accurate par levels for raw ingredients, and providing deeper data on labor to help managers decrease costs.




The necessity for a system such as this could only be understood through experience in the industry and its unique challenges. By carefully studying the business over time and identifying its pain points, I was able to design and implement a system that removed a great deal of uncertainty and guesswork which has led to reduced waste and increased profits.


NOTE: Screens of this work are not yet available as the project is in development.