At Steadforce, it’s part of the daily ritual: after lunch, bowls of sweet and salty snacks are placed in the kitchen for everyone to enjoy. It’s delicious, boosts morale—and sparked a burning question: which treats are the most popular?

The ChallengeCopied!

We wanted to create a vivid, hands-on showcase to demonstrate how data-driven projects work in the real world. Many use cases across industries share a common core, so we needed a relatable and tangible example to explain typical data science algorithms and workflows.

The Route to SuccessCopied!

The idea was born during a brainstorming session in our kitchen. Our Analytics team—Federica and Jonathan—decided to analyze the company’s candy consumption in detail. Why? Because this scenario reflects all the key elements of a classic data science project: data acquisition, modeling, prediction, and handling messy input.

To get started, they built their own scales using Raspberry Pi microcomputers, sensors, plywood, and cardboard. Each scale could measure weight changes and feed data directly into their models. The goal: predict when and which type of candy would be eaten.

The initial approach used a simple Baseline Model to deliver a first prediction. From there, more complex statistical models—borrowed from actuarial and medical fields—were applied to improve accuracy.

The Benefits for CustomersCopied!

Of course, this wasn’t just about sweets. The same methods used here are also applied in serious settings: predicting medication usage in hospitals or detecting maintenance needs in industrial systems.

This project offered a clear demonstration of how predictive modeling can be used to generate meaningful business insights—whether for procurement, logistics, or forecasting.

Data InsightsCopied!

Over four weeks of measurement, the team:

• Collected daily consumption data from four smart scales.
• Categorized and cleaned the data for analysis.
• Used predictive models with increasing accuracy.
• Achieved a prediction accuracy of about 70%.
• Discovered trends like higher candy consumption after salad days, and lower after pasta days.
• Identified milk chocolate and gummy bears as the top favorites.

Problems & SolutionsCopied!

Naturally, not all data was clean or easy to interpret. External factors like office visits, birthdays (with cake!), and fluctuating attendance (e.g., fewer people on Fridays) influenced the outcomes.

Some playful colleagues also introduced noise—like mixing candy types or removing bowls—mimicking real-world data manipulation and anomalies.

To deal with these issues, the team built smoothing software to filter out sensor noise and clean the datasets. Real-life analytics often means handling imperfect data, and this project was no exception.

Sweet ConclusionCopied!

The result? A fun, successful case study with ~70% prediction accuracy and actionable insights. Procurement now knows exactly which snacks to stock, and the whole team stays happy—and a little bit sweeter—thanks to data science.