Let’s look at the comparative performance of a traditional psychographic segmentation and PersonaLive™ in predicting liquor sales. First, here’s some background on this dataset.
Source: State of Iowa public data (1,741 stores)
Data: Liquor sales at the product-store level
Timeframe: Jan 2019 - Jan 2020
For this test, we selected 10 products from these stores to represent a variety of alcohol interests and preferences so we could show wide applicability.
The left column shows the breadth of liquor types, including flavored, fruity, stiff, high-end, whiskey, vodka, etc. In the next three columns, you can see which set of variables won in predicting sales outcomes for each of these products.
In 8 of out 10 tests, PersonaLive™ outperformed traditional segmentation. As a point of reference, we included age and income as another column.
As you can see, traditional segmentation does have a lot of value, increasing predictiveness from 8.3% to 21.8%. You can also see that PersonaLive™ has improved upon that, adding an additional 3.5% in error reduction.
Let's take a look at how the Proximity dataset does in combination with PersonaLive™. Because the Proximity dataset is based on different data, it captures a different signal, and so, we would hope that it would improve performance.
By combining the two datasets, we get a further 8.7% reduction in error—that's a 12.2% difference in overall error reduction compared to traditional segmentation. To recap, when predicting Iowa liquor store sales for these 10 products, age and income resulted in an error reduction of 8.3%; traditional segmentation 21.8%; PersonaLive alone 25.3%; PersonaLive and Proximity 34%.