It’s the end of the year, and that means end of the year lists. Favorite beers, best beer bars, and so on. As a data geek, one I always enjoy is Google’s top searches lists, the only place you’ll probably find “What is 0 divided by 0?”, “How do you use beard balm?”, “Eazy-E” and “Philly Cheese Steak Recipe” on the same page of lists. In recent years, Google has included beer as one of their topics. This year they switched it up, and instead of top searches, included beer topics that were “trending”.
That got me to thinking about whether Google Trends data was useful as a sales/marketing analysis tool in the broader scheme of things. I often point to the data in my analysis, without really digging into what it means or how it might relate to sales. I am sure breweries also use Google Trends data from time to time to evaluate styles, marketing terms, or other data over time or across geographies. Obviously, Google Trends data is accurate for what it is, measuring the volume of searches in Google’s engine, but how well does it predict more tangible outcomes like sales?
To answer this question, I lined up Google Trends data for two searches, “craft beer” and “IPA” (with the beer style specific search selected), both versus IRI scan data. For the first (“craft beer”), I looked at searches in four week periods versus IRI’s craft sales during the same four week period. For the latter (“IPA”), I looked at the searches for 2015 across US states. As always, IRI data is a snapshot of the off-premise, and isn’t comprehensive across all geographies or channels, but has aligned pretty well with the overall market in recent years.
Overall Craft Sales and Searches
So how did Google Trends do? I’d rate the data as good, not great for predicting sales. While both searches certainly showed a positive correlation with the sales they were lined up against, there was a lot of additional variation. Let’s dig in a bit more to see what this means and where that variation might come from.
Here’s a graph showing the overall “craft beer” searches vs. IRI multi-outlet + convenience (MULO+C) data for various four week periods starting in May of 2013. I fixed the high point for each line as 100. You could fix other points at 100, such as the starting point, which would affect how the chart looks, but not the statistical relationship. The correlation between those two lines is r = .55, which is a pretty strong correlation for that many data points (n = 34). That said, there’s still a lot of unexplained variation left over, and while the search trend mirrors the overall growth of sales, the growth and decline in individual four week periods leaves a lot to be desired. Why would that be?
Source: IRI Group (2015) and Google Trends (2015)
For one, searches don’t always translate into sales (at least immediately). For example, there’s a spike in searches every May around American Craft Beer Week. Although sales also tend to increase in May, it’s clear that in the short-run the growth in searches far exceeds the growth in sales (though it may help more in the long-run; incidentally, the correlation is much stronger on a two-month lag, though I’m hard-pressed to come up with a causal argument about how searches two months ago drive purchase decisions today).
Secondly, “craft beer” may be a decent blanket term, but certain times of year people are likely searching for more specific items—for example, “pumpkin beer” in October. That will show up in beer sales, but not in searches. This may be why searches are down in the fall whereas sales are up.
IPA Sales and Searches by State
We see a similar general alignment when looking at searches by geography. The bottom axis is the search index for 2015 (for a few states with low search volume 2014 data was included to complete the data set). The y-axis shows the calculated share for IPA in that state (in all beer) based on craft’s share in the state and IPA’s share within craft. All states are either the multi-outlet or MULO+C channel with the exception of Colorado, which is the liquor channel.
Source: IRI Group (2015) and Google Trends (2015)
At the same time, the relationship between the search index and the sales data is far from perfect. It’s possible that some of this is a result of the data, and that off-premise scan data doesn’t accurately capture the level of sales at this focused a level (a style in a specific state). I’m more inclined to believe it’s simply due to variations in what searches represent as well as variations in market conditions.
The goal of this exercise was to evaluate a data tool that’s free, up-to-date, easy, and flexible and to align it with sales data, something that remains critical for breweries in the business analysis. The analysis showed that although search data certain picks up some of the same information, there remains a significant amount of variation between the two measures. As always, this particular data tool provides some information, but should be used in the broader context of different types of analysis.