s
Data Science in the Bakery
Running a bakery involves more than just production. It also includes managing stores, handling logistics tasks like deliveries and returns, maintaining inventory, and promoting sales. Additionally, there’s a significant responsibility toward customers who buy our products, ensuring quality, timeliness, and a complete product range.
If we want to be more precise, we can divide the production process into many smaller processes, such as supplying energy, raw materials, and semi-finished products. Managing employees is also a complex task, as is maintaining machinery and property. We’re talking solely about production, yet we also have vehicles that must be kept in excellent technical condition and a team of drivers. Between production and distribution lies the process of order fulfillment, where human resource management plays a crucial role. Sales involve overseeing anomalies, such as cash register shortages or inventory discrepancies.
Running a bakery is a highly complex business and is undoubtedly a demanding and intricate task. The development of artificial intelligence (AI) can significantly ease these burdens. However, to simplify and modernize operations, one must first invest considerable additional effort, all while handling daily responsibilities. Ignoring modernization and optimization will gradually erode competitive advantage, making operations even harder and less profitable. Therefore, it’s unwise to bury one’s head in the sand. Fortunately, some applications of AI are exceptionally affordable and easy to implement, like fleet management systems.
Application 1: The Traveling Salesman Problem and Transportation Optimization
One of the most efficient systems for managing and optimizing logistics processes, particularly route management, involves solving the traveling salesman problem. In a bakery, one of the most thankless tasks is that of the dispatcher, who is responsible for planning daily routes for multiple vehicles. Large bakeries usually have two delivery shifts: morning and afternoon. Routes change daily, as different product assortments are delivered each day, often requiring specific sequencing for delivery locations. Planning optimal routes for several vehicles is a labor-intensive task. AI assists with this by applying the traveling salesman problem and transportation optimization, both based on the simplex method, a known algorithm in operations research and econometrics.
In the past, I calculated optimal vehicle routes manually using a computer or calculator and paper. This manual approach is highly labor-intensive and prone to error. Today, there are companies (including some in Poland) that offer route calculation services for each vehicle for just a few dozen PLN per month. This is a highly effective solution that requires virtually no investment and is very profitable. The dispatcher receives an application in which they enter all the necessary data for current transport orders, and the algorithm calculates the optimal routes and sends them to drivers’ smartphones.
Application 2: Anomaly Detection in Transport
Anomalies are rare events that negatively impact the bakery’s financial health. By anomalies, I mean minor thefts or sales fraud. In the past, business owners would monitor vehicle odometers and calculate fuel consumption manually. Today, remote monitoring systems oversee routes and fuel usage, easily detecting fuel siphoning, unauthorized routes, and unattended vehicles. Similar to route optimization, everything can be monitored via a mobile app. Implementing such a solution is relatively inexpensive and highly profitable.
Application 3: Production Volume Optimization
This solution is more complex but also far more profitable than the previous two. In a bakery, the rule is: what isn’t sold today must be discarded or allocated for disposal, which equates to wasted labor costs, raw materials, energy, and transportation expenses. Ideally, we would produce exactly the amount we can sell, but achieving that is challenging. However, we can create a model that can calculate daily sales of specific products with a high probability, enabling production planning to minimize waste.
How do you build such a system? First, you need to consolidate all sales records from past years into a single database. Sales records from cash registers contain the date, time, items sold, and prices. Compiling these records across years is simple and provides a treasure trove of information. Sales data can then be combined with a calendar, linking specific days of the week and hours with seasons. Weather conditions from particular days and times can also be integrated. This consolidated database is fed into a machine learning model to build a prototype, which is then deployed to the cloud—a process called productionalization. Production managers access an app that connects directly to the cloud-based model, predicting the daily sales volume for each product at each store. Building such a system requires high-level data science expertise but can be funded with EU or KPO funds, which may be unlocked after upcoming elections.
Application 4: Automated Stores and Facial Recognition
This advanced project involves creating an autonomous store where customers serve themselves. Since bakeries usually sell low-value items, there’s no need for machines or special dispensers for bread or pastries. However, some oversight is necessary. AI can assist by implementing cameras that recognize customers, not necessarily by greeting them by name but by identifying regular customers. Based on their purchase history, personalized shopping suggestions, incentives, and free products as gifts can be offered. Another goal of facial recognition in the store is to remember faces of those who behave dishonestly, taking items without payment. Detection of such behavior would be fully automated. If someone is caught stealing, they would be notified of their previous unpaid goods on their next visit.
Setting up a facial recognition system requires qualified data science professionals. While face-detection applications are free, implementing and deploying such a solution incurs significant costs.
Application 5: Measurement of Production and Storage Processes
Every machine in a bakery generates numerous readings, such as gas, water, and electricity consumption, as well as time series data on temperature, pressure, and humidity. Since machines have interfaces, it’s relatively easy to establish a system for collecting technical information. A database of technical parameters over time serves as an excellent foundation for building machine learning models. These models can detect anomalies signaling potential future breakdowns, reduce raw material or energy consumption, and improve product quality.
Application 6: Recommendation System
Recommendation systems are commonly used in online sales. Until recently, only long-shelf-life products were available online, but now bakeries, pizzerias, and patisseries (selling items like cakes and pastries) are also online. A recommendation system is built on a customer database. Typically, customers must log in and create a profile to make online purchases. Sales analysis systems identify customers and analyze their buying behavior, enabling automatic customer segmentation based on preferences, financial resources, and behavior. The next step is to optimize the product offering for selected customer groups. Customer segmentation allows for various solutions, such as suggesting a second product based on the customer’s group and selected product. Building such a recommendation system requires highly skilled data science experts. These sales systems can increase sales by up to 30% without additional promotional or advertising expenses.
How to Finance It?
As I mentioned earlier, investment in digitalization and AI development can be funded by the KPO. Poland is set to receive 158.5 billion PLN, of which 33.7 billion PLN must be spent on artificial intelligence, digitalization, and IT. The fund expires on August 31, 2026, leaving three years to use these substantial funds. Given the limited time, funding applications are likely to be less rigorously reviewed, making it relatively easy to secure funding for AI development in a bakery. It’s worth trying.
Wojciech Moszczyński