Big Data

Using Affinities to Optimize Promotions- Part 1 of 2

By Stacy Klimkowski • March 23, 2016


Whether you are a category manager, merchandiser, retail buyer, or shopper marketer in the fast moving consumer goods industry, chances are you use a variety of data sources to make important business decisions. When it comes to promotion planning, deciding what items are best to drive dollar sales and co-promote together can feel like a guessing game. The lack of insight surrounding item cross-purchase and incremental sales impact often results in promotions being run solely to use trade funds. By using full-store basket-level transaction data to identify item relationships, retailers and suppliers can make more informed recommendations. Here we examine how category and item affinities can be analyzed to plan and maximize your promotions.

For instance, if I am a category manager for the fast-growing high margin energy category, and my retailer wants to partner with me to run a bundled promotion/event, I would start by understanding what categories and items are sold with energy drinks and the respective relevancies they have together. The measure of these interrelationships are what we will call “affinities” and calculates the likelihood of a purchase of a category, item, or group of items, based on the respective co-purchases in the basket. The key affinity metrics that we will use to conduct a market basket analysis are as follows:

  • Occurrences: the number of times that an item was purchased with an energy drink
  • Support: the % of total transactions that have an energy drink and affinity item in the basket
  • Confidence: of energy baskets, it is the % of transactions that have energy drink and the affinity item purchased together
  • Lift: determines the likelihood of the transaction for an item/category vs. any other item/category (i.e. breakfast sandwiches are 50% more likely to be bought with energy than any other category)
  • Conviction: describes the strength and directionality of the product correlation; a higher positive number indicates energy strongly drives the affinity item, while a lower negative number shows that the affinity item more strongly drives the energy drink purchase.


The chart below illustrates that even though juices and tonics are the most often purchased item with energy drinks, occurring in 1.6 million transactions, the negative lift indicates that it is 7% less likely to be purchased with an energy drink than a random category. On the other hand, breakfast sandwiches, while occurring less frequently, are 50% more likely to be bought with energy drinks than a random category. Now that we know there is a connection between those categories, we look to the conviction which shows a negative number, indicating that without a breakfast sandwich in the basket, the energy drink purchase may not occur.


So we have learned that breakfast sandwiches are likely to be purchased with energy drinks and drive the sales of energy drinks, therefore, we can conclude that a bundled breakfast promotion with energy drinks is most likely to have success. The next step is to examine which items would be most relevant to promote together.


Below we can see that Monster and Redbull are fairly equally likely to be purchased since the lifts and convictions are in the same range and are directionally the same, however, Monster pairs slightly better with breakfast sandwich #1.


STEP 3: 

Extracting this same data by SKU level, also shows that Monster 16 oz. cans are the leading energy item purchased with breakfast sandwiches, occurring in 44% of baskets vs, the second leading item Red Bull 12 oz. found in 35%. Therefore, it makes sense to run the promotion with the better selling and higher price-point Monster 16 oz. drink.



When time context is added using a daypart analysis, we can see the frequency of energy drink purchases broken down by hour of the week. The below graph shows that energy drinks are most often bought from 5 a.m. to 9 a.m. which validates why breakfast is so relevant in the affinities analysis. Perhaps McDonalds is on to something in their decision to test selling Monster 16 oz. energy drink along with all day breakfast. 



Putting it all together, these insights can be used by category, sales and marketing teams to improve promotions, optimize store layouts, understand shopper behaviors, and even enhance media targeting.

SwiftIQ’s recent launch of its Affinity application can conduct on-demand market basket analyses in seconds and provide purchase behaviors across, items, brand, and categories, as well as time of day, day of week, and average dollars and items in a basket. If you want to optimize retail execution, or learn more about SwiftIQ’s product affinities request a demo by clicking here.



SwiftIQ uses high scale data processing and machine learning to deliver contextually relevant insights and digital experiences for retailers and brands. Its platform unifies and analyzes data primarily from in-store transactions as well as online behavior and third party sources to predict and inform category captains, shopper marketing, assortment, supply chain and content delivery decisions. SwiftIQ’s unique ability to process billions of basket-level transaction records in near real-time and convert that into on-demand mobile visualizations, dayparts, affinities and attribution fosters a new level of retailer/supplier collaboration and innovation.

Since launching its category captain and consumer behavioral analytics platforms in late 2014, SwiftIQ now analyzes over $60 billion in offline, receipt-level point of sale data. The company serves 5 of the global leading category suppliers and several billion-dollar retailers. SwiftIQ, named a Top Innovator twice by DataWeek, has also been recognized by Forrester, Forbes, NACS, ProgressiveGrocer and ComputerWorld for its achievements. 

Leave a Comment