Two weeks ago, in Part 1, I described how to build a recency, frequency, and monetary (RFM) model in a traditional retail environment. If you haven’t read Part 1, I advise doing so before reading Part 2. Not doing so would be like watching “Die Hard 2” before “Die Hard.” It’s nearly impossible to grasp the complicated plot of the sequel without first watching the original, experiencing the development of dark and mysterious characters such as Detective John McClane, Officer Al Powell, and Argyle. You get the idea.
Before we proceed into Part 2, I want to clear up some confusion I may have created last time. I referenced RFM models of a car dealer and a corner pharmacy. The car dealer and corner pharmacy can only define the parameters of the scoring system for each variable (RFM) by first performing the RFM analysis, as described in the “Building the Model” section of that column. Once the RFM analysis is performed, the car dealer and corner pharmacy analyze the cut-off points between the top 20 percent and middle 60 percent and between the middle 60 percent and bottom 20 percent to define the parameters of the scoring system for each variable. Daniel Riff let me know that this was not clear and deserves kudos for his attention to detail. Mea culpa.
Internet Media and the Point of Transaction
In Part 1, I promised to discuss how to apply the RFM model to an Internet media company. The key to applying the RFM model in this context is defining the point of transaction.
The point of transaction for a retailer is obvious. It’s the point at which a customer buys. When a customer opens up her wallet and plunks down cold, hard cash in exchange for a product. For an Internet media company, the point of transaction is less obvious. It’s when a site visitor either is exposed to advertising (impression-based ads) or makes a revenue-generating action (performance-based advertising). When a visitor sees a paid banner ad, a transaction occurs. When a visitor clicks on a paid search result, a transaction occurs. Defining the point of transaction is the key.
An RFM Model for Impression-Based Advertising
Let’s look at building an RFM model for an Internet media company selling impression-based advertising. First, define the transaction cycle. In today’s Internet media environment (which fosters attention-deficit disorder by offering access to mountains of information at the click of a mouse), if a visitor hasn’t returned in 90 days, it’s reasonable to assume he isn’t coming back anytime soon. He didn’t find what he was looking for or found better information elsewhere. Pull a list of all visitors in the past 90 days (the assumption here is that your site tracks visitors, most likely through a cookie). When pulling the list, request the following for each visitor:
- Date of most recent visit
- Number of visits in the past 90 days
- Average ad revenue per visit
Obtaining the third item may require some mathematical gyrations. Multiply the total number of page views for each visitor in the past 90 days by the percentage of ad inventory sold (assumes one ad unit per page view). This gives you a projected number of advertising impressions for each visitor in the past 90 days. Multiply this number by the average ad revenue per impression to get the projected total ad revenue generated by each visitor in the past 90 days. Divide this number by the total number of visits. You arrive at a projection for average ad revenue per visit for each visitor.
You now have the data you need to perform the RFM analysis. Start with recency. Sort the list of visitors by date of most recent visit, with the most recent visit at the top descending to the least recent. Assign a score of 3 to the top 20 percent of visitors, a score of 2 to the middle 60 percent of visitors, and a score of 1 to the bottom 20 percent.
Now move to frequency. Sort the list of visitors by number of visits in the past 90 days, with the highest number of visits at the top descending to the lowest number at the bottom. As before, assign 3 points to the top 20 percent, 2 to the middle 60 percent, and 1 to the bottom 20 percent.
Finish with monetary. Sort the list by average ad revenue per visit, with the highest at the top. Once again, use the same scoring method: 3 points for the 20 percent; 2 points for the middle 60 percent; 1 point for the bottom 20 percent.
As with the retail model, sort the list so visitors receiving scores of 333 (remember to always represent the numbers in RFM order) appear together to form a single segment, visitors scoring 323 appear together, visitors scoring 322 appear together, and so on. When complete, you will have 27 total segments of visitors, one segment for each combination of the RFM variables (see table below). People in the 333 segment are your most valuable visitors. Those in the 111 segment are the least valuable.
Analyze the cut-off points between the top 20 percent and the middle 60 percent and between the middle 60 percent and bottom 20 percent for each recency, frequency, and monetary. You will find it illuminating to understand how visitors fit into each of the 27 segments. For example, after performing the cut-off points analysis, you may determine that to receive a score of 333, a visitor must have visited the site in the last 15 days (recency), visited a minimum of 45 times in the past 90 days (frequency), and generated average advertising revenue per site visit of at least $2.00 (monetary). After analysis, you may also determine that to receive a score of 111, a visitor must not have visited the site in the last 50 days (recency), visited a maximum of 10 times in the last 90 days (frequency), and generated average advertising revenue per site visit of no greater than $0.05 (monetary). By translating the 333 and 111 scores into these terms, you can create successful marketing strategies targeted at each segment of site visitors. That’s where we’ll pick up in two weeks. Until then, in the words of Detective John McClane, yippee kay-ay…
Emily Ma, product director of Tencent’s advertising platform products department, was a keynote speaker at ClickZ Live Shanghai where she discussed the ... read more
In today's multichannel world how can marketers use data to ensure the experience a customer receives is relevant to them?
The terms that customers type into your site search function can help you to gain an understanding of user behaviour and can be used to optimise ... read more