Vintage Analysis — A Visual Primer
Techniques for measuring portfolio improvement
This article makes the assumption that you have an understanding of vintage and the concept of vintage analysis. For a short overview, you can click here. This article will focus on techniques in visualization.
Vintage analysis is a tool for analysis that quickly enables you to align groups of assets based on the time that has past since some major starting point. For our example here, that starting point will be acquisition and our asset will be new customers. Put another way, we will measure groups of new customers based on the month they joined our site or product. This often referred to by the name cohort analysis as well.
The chart below is typically called a Vintage Wedge, although I have no doubt it goes by other names in other places. The wedge lists each monthly vintage on the rows, beginning with the oldest at the top.
Along the columns, we use a dimension known as month on book (MOB). MOB is the number of months since our vintages acquisition point. Our example starts with MOB 1 and progresses through MOB 17. Some companies use MOB 0 as a starting point. Since the customer has not officially had a full month on book until the last day of any MOB, this can seem more intuitive. Unfortunately, it makes it difficult to use the MOB as a denominator for point in time averages.
Within the cells, we have calculated an activation rate. Rates and averages are common in vintage analysis, but so are nominal metrics like spend, visits, clicks, etc. The key to any vintage average or rate is that the denominator does not change. In our model, each rate is based on the cumulative number of customers who activated over the total number of customers acquired in that vintage month.
To close, reading his chart we learn that 17 months after acquisition 24.4% of the new customers acquired in October of 2014 have activated. Better defining exactly what activated means might benefit the client of this wedge, but really doesn’t advance this article.
Visualizing the Data
I am not a huge fan of the wedge as a final product. It is required to graphically display your data. Below is a simple line graph used to illustrate the numbers above. In some industries, this graph is referred to as a seaweed graph or possibly a spaghetti graph. The former is more common at high growth companies… but more on that later.
Reading a vintage line graph is easy as long as you sample properly. If we had graphed all the rows above, this graph would have been unreadable. The chart above selects only the first monthly vintage of each quarter. For this data, that represents a fairly unbiased sample (that may not be true if your data has a bias around quarters, perhaps you are measuring a quarterly tax prep service?).
The ‘seaweed’ graph above shows near perfect quarter over quarter improvement in the activation rate. Each younger vintage sits higher than the prior vintage denoting a higher activation rate at various ages. This can be highlighted by selecting a point on the x-axis and converting those points into a bar chart that highlights performance at a specific age of the vintage.
Note — this is a rather low-grade effort. This graph should have better labeling, the y-axis is truncated which visually exaggerates the monthly lift, and would benefit from numerous other additions that we will cover in later articles. That said, this visual allows you to include all the vintages (rather than sample) and focuses the user on the very steady improvement we are seeing.
Only the improvement isn’t steady, there is a hidden event.
Looking back at the wedge graph, you may have noticed the small green triangles slicing across the graph. These were added for this call out, I doubt the numbers in the graph actually drew your attention.
On the line graph, you were more likely to notice the odd kink in the Oct 2014 vintage (blue) line at the bottom. Though only because the other lines lack any real noise. Surely, you did not notice that the improvement in the second vintage bar over the first was smaller than others (though a split axis showing lift would have helped).
This is because the vintage wedge actually works to hide seasonal influence. Any graphic style you choose will help you see some things while hiding others. There is no perfect graph!
Enter the a new Wedge
The wedge below represents the same data now arranged by calendar month. Looking at the column for March 2015, you might notice and odd bump in the numbers. When I generated this sample set, I coded in an extra 75bps increase in the rate for that calendar month. If you still don’t see, no worries.
The line graph below focuses on the six vintages impacted by that event. A typical event might be associated with tax season, basketball's March madness, a social media program by your marketing department, or (sadly) a data error in your warehouse.
Looking at the graph above, you may still be struggling to find the event in what you see. Put simply, this wedge is NOT the best visual tool to investigate point in time events. I promise to cover better tools in later articles.
I also promise you that after a few months of making vintage visualizations a primary tool for your analysis, that graph will be sufficient. A little practice will go a long way to making you more sensitive to subtle changes. The trick is then to confirm them and communicate them in more direct way.
Thanks for reading and stay tuned! We hope to make this articles a weekly feature on our new Charting Ahead magazine. Please follow and recommend.
Learn more here:
This content is also shared on LinkedIN through a partnership with the CiA LinkedIN group. If you are interested in joining Career in Analytics — click here.
