Success story: Coca-Cola Optimizes Vending Machine Site Selection with Synthetic Society

Download PDF

Summary

Coca-Cola partnered with Arima to tackle the 'Pick' or 'No Pick' Dilemma, seeking the ideal locations for vending machines to enhance customer service. Arima's Synthetic Society datasets were carefully vetted, and the analysis focused on 24 demographic variables out of over 25,000 available. As a result of their collaboration, Coca-Cola achieved a significant increase in the probability of selecting a "good location" (from 45% to 68%) and a 43% boost in average sales for model-selected locations.

Key improvements

Probability of choosing a "good" location: 68%            

  Average sales increase for model-selected sites: 74%         

      Sales index for chosen locations: 76%

The Problem: The 'Pick' or 'No Pick' Dilemma

Vending machines are important outlets for the distribution of Coke products, and while they only represent 1% of Coke's total revenue, it's also the most profitable business unit and one of the few channels where the organization has full control of. As such, a crucial question is: How can Coca-Cola improve their customer service through identifying ideal locations for placing vending machines? As the biggest beverage company in the world, Coca-Cola has a major presence in Canada with 25 different brands, over 50 production facilities, and more than 5,700 staff servicing 38 million customers. Their most popular brands include Coca-Cola, Minute Maid, Dasani, and Powerade - household names that virtually all Canadians recognize.

The Solution: Mobility Data

Coca-Cola underwent a thorough vetting process to ensure Arima's Synthetic Society datasets could help them achieve their goals of understanding how consumers move around locations and creating accurate profiles of these consumers.

Before the commencement of the project, they expected three things:

1. Mobility data has enough geographical and time granularity - that is, mobility data must be able to identify who has been physically near a vending machine and that they are an actual customer and not someone who walked by the area.

2. Mobility data can be tied back to Arima's Synthetic Society so that Coca-Cola can create profiles of each vending machine visit.

3. The sample of both datasets must be big enough to draw statistically significant answers.

The Synthetic Society is a very rich dataset with more than 25,000 variables about Canadians. Coca-Cola had the luxury of picking the most key attributes about their potential customers. The analysis started with using 24 demographic variables, with a concrete plan to expand the number of features in subsequent analyses.

Finally, what made Arima really stand out to Coca-Cola's team was how easy it has been working together. "We always know there's someone at Arima who can answer our questions, help us work through issues, and be receptive to our feedback," emphasized Nikita Medvedev, Director of Advanced Analytics at Coca-Cola. "This really feels like a partnership; Arima offers a level of service you don't get with other vendors."

“We always know there’s someone at Arima who can answer our questions, help us work through issues, and be receptive to our feedback.”

The Result:

Arima worked with the Enterprise Analytics team at Coca-Cola to develop predictive models based on their 1st party data (e.g., vending machine locations, machine type and revenue) and Arima’s Synthetic Society.

As a result of many brainstorming sessions and technical iterations, the team achieved the following results:

1. The probability of picking a "good location" increases from 45% to 68%.

2. On average, a model-selected location generates 43% more sales.

3. Locations picked by the model have a 76% higher sales index compared to the ones not picked.

Coca-Cola utilized the more accurate consumer and mobility data to drive efficient placement of vending machines, which, in turn, led to higher revenue from relocating existing placements and adding new vending machines.