We began with the idea that we would be specifically writing a program to assist sales associates for AB InBev in Quebec. Hung up on the details, we noticed the distinct kinds of foods consumed -- such as moules frites -- and recognized the potential to encourage beer and food pairing. About a year ago, AB specifically recognized the importance of the opportunity to focus on providing convenience to consumers. In our case, we are focusing on retailers as opposed consumers, but believed our partnerships could still leverage insights from our data on cross-selling for beer and food. In addition to this, the rise of popularity on craft beer in the industry suggests a potential need/opportunity to emphasize the premium quality of beer in creative ways - in our case, beer and food pairing.

What it does

This lead us to a pivot from simply discovering food and beer pairings that could be suggested for bars in Canada to building an algorithm that uses the provided data as training data for a location agnostic program that can discover potential beer and dish pairings. Sales reps will have an interface (currently desktop, but with the potential to be put on the current mobile app they use) that shows (bar/restaurant) buyers of AB InBev what consumers tend drink with meals that a given buyer sells. The key is that the program will suggest type and an amount of beer the buyer should purchase that they are not currently doing so already to seize the cross selling opportunity.

How we built it

BeerSquad split into back end and front end teams. The front end determined the best methods for visualizations based on the needs of both sales reps and buyers.[React/Redux, Chart.js, Axios, Node.js, Express] The back end team began by going through a standard process of further exploratory analysis, data cleaning and data selection. [mySQL, Python]The back end then created algorithms to set rules for beer-food pairing analysis that could then send suggestions to the front.

Challenges we ran into

Although data cleaning and selection proved to be an unexpectedly time consuming process, our main challenge was determining how to train the program to accurately discover meaningful relationships between beer purchases and food purchases. In response to these challenges, we decided to attack the problem from two angles: creating a system for analysis and creating a tool for beer-food pairing. In order to create an effective system for analysis , we took the raw data and wrote an algorithm that [normalized/fixed] inconsistent user input for product data in the food category. For the pairing tool, we overcame this challenge through narrowing our training data to only include the beer purchases where the brand type was (in the top 20th percentile of occurrence) to ensure accuracy. To add a layer of potential, we added a more flexible interface that allowed the sales rep to manipulate data visualizations in a way that provided more customized recommendations for buyers.

Other expected challenges were quantifying beer-food pairing characteristics and figuring out how our front end could most effectively consumed our back end program as an API.

Accomplishments that we're proud of

While the idea behind and functioning of our program exceeded expectations, we are most proud of our ability to identify and solve a problem as a team with both a clear direction and the flexibility to adapt to new ideas.

What we learned

We learned the importance of learning about data before beginning the cleaning process. This was critical for this particular situation because salience of each variable in building a program that learns from different sets of data would have a disproportionate negative effect on results if chosen incorrectly. The next thing we learned can more accurately be called that we remembered the importance of asking questions. By asking the mentors for clarification, we were able to not only clarify the problem, but also learn about the goals of AB InBev and how we could make sure our program was helping achieve this goal.

What's next for BeerSquad

While we our proud to say our program has proved the ability to learn beer-food pairing suggestions from existing data, we would like to perfect conclusions made from seed data. Furthermore, the core value of our program is the idea behind it: using a program that provides beer-pairing suggestions to increase convenience for the consumer and highlighting the premium qualities of a given beer. With additional data from AB InBev on the measurable characteristics of its beer brands, we (can) also increase the number of beer-food pairing and equip sales reps with additional information to assist them in increasing sales to distributors.

Built With

Share this project: