Articles
June,
2002
Segmenting Your Housefile
Most mailers are well aware of buyer file segmentation using RFM: Recency, Frequency, and Monetary Value. However, in my review of numerous circulation plans of small catalog companies, I have found that full RFM segmentation is rarely used. Additionally, when it is used, it is not effectively applied to mailing decisions. "My house file is too small for any segmentation beyond recency," or "I can mail all my buyers and get a great response," are the two reasons I hear most often for under-segmentation. These excuses frequently result in over-mailing or under-mailing of specific portions of the buyer file - a waste of valuable resources. In this article, we will explore the importance of segmenting even small files, ideal segmentation, and how to apply this segmentation in making better mailing decisions.
Recency
Recency is the most powerful determinant of whether a buyer will buy again. Table below shows the declining response ratio that results as buyers age for a hypothetical catalog company I shall call Typical whose buyer file was producing an average of $4.01 per catalog mailed. The most recent segment, Jul - Dec 2001, will produce 1.75 times the average for the buyers file. However, response quickly goes below the buyer file average as the file ages with buyers older than 1998 producing only 31% of the average response or $1.20 per catalog mailed.
Since recency is the most important segmentation element, it is important that it be done correctly. For two reasons, the segmentation should reflect actual date ranges rather than regressive spans of time. For example, use July - Dec 2001 rather than 0 - 6 months.
The first reason is that specific date ranges can encompass buying seasons. Very often spring buyers have very different response than fall/holiday buyers. If one is doing a merge in March, a 0 - 6 month buyer bridges two very different groups of buyers, the holiday gift buyer and early spring buyer. As one goes back in time, different mailing decisions should be made for these two groups. If multi-seasonal buyers are mixed in a single segment, differential response goes undetected and cannot be used to one's advantage. In selecting calendar periods, you should chose periods that reflect seasonality specific to your business, for example, spring outdoor décor buyers verses fall indoor décor buyers.
The second reason is that specific buyers do not migrate from dated ranges until they buy again, allowing you to control the number of mailings sent to a specific buyer. For example, buyer "Jack Jones" who last purchased on February 17, 2001 will remain a buyer in the Jan-June 2001 segment until he buys again. Therefore, if you want Jack to receive three catalogs during the year, you will mail the Jan-June 2001 segment three times. In contrast, if you are using regressive time periods, you will not know when "Jack" migrates from segment to segment and you cannot control the number of contacts.
Frequency
Frequency is the second most powerful indicator of response. For our Typical catalog, Chart below shows the total buyers that have purchased one time, two times, and three times. The portion of buyers who have purchased one time produced $2.90 or .72 times the average buyer response of $4.01 response. Two time buyers produced $5.54 or 1.38 times the average. Three time buyers produced $11.75 or 2.93 times the average. In making mailing decisions as files age, you can justify mailing two and three time plus buyers much more frequently than one-time buyers.
Monetary Value
Buyers who have made large purchases are more likely to continue to spend at higher levels. Chart below shows the portion and response of buyers at three different average order levels in our Typical catalog. A buyer whose average order level is $150 or higher responded at $7.04 or 1.76 times the average buyer of $4.01. In the $75 - $150 range, a buyer performed at $3.74 or .93 times the average, and finally, a buyer below $75 performed at $2.26 or .56 of the average.
As with frequency, a segment with a higher order value will justify more frequent mailings.
Dollar ranges should be selected that both encompass sizeable portions of the buyer file and demonstrate significant differences in response rate. This will vary depending on your average order. When applying segmentation, care should be taken to assign buyers to segments based on average order or mean order. Using "total dollars to date" will duplicate the meaning of frequency.
Combined Segmentation
Combining frequency and monetary segmentation results in the segments for each period of recency.
Note that I added a tenth segment to our table: "Buyers under $10." I have found when segmenting files for various catalogs, that for various reasons, buyers reside in this segment. Depending on the catalog, this is the result of buyers who have returned their purchase, buyers who are really "inquiries" who purchased their catalog, or file conversion errors where dollars were dropped. It is wise to isolate these buyers and analyze response to this segment separately. If you find you have significant numbers of these buyers, you may also identify a procedural or systemic problem that should be addressed.
Combined Ratios
By combining the ratios that we established for frequency and monetary segments, we have new ratios of how, on the average, each of these segments should perform in relationship to one another. The chart of the right shows that a 3X $150+ buyer should perform 5.16 times the average buyer response of $4.01 for our Typical catalog whereas a 1X under $75 buyer should produce only .40 times the average response. For most catalogs with small buyer files, we could not have established these relationships based on an analysis of the response rates of individual segments in any single time period. The quantities of buyers in any single segment and resulting orders are simply too small for statistical significance. But when examining the relationship over entire mailings, we are more likely to achieve statistical significance and repeatable results.
Having established these ratios on frequency and monetary value, we can then combine them with the relationship of recency periods. Simply multiply the frequency/monetary ratios times the ratio for each time period in Table One. For example, the table below, the combined ratio for a $75-$150 / 2x buyer is 1.28 and must be multiplied by 1.75 for the time period of July - Dec 2001. This yields a factor of 2.24 that is multiplied times the average of $4.01 with result of $8.98 as the expected response rate. After you have established all of the ratios for your catalog, you can produce a spreadsheet with an approximation of the expected response rate for every segment.
You are now in a position to make decisions as to mailing or not mailing segments that, in themselves, have too few names or statistically invalid past results to stand by themselves. This is particularly true with relatively young catalog businesses that have few buyers in their early years, and it is in these early years where you are most likely to discontinue mailing selected segments.
Additional Segmentation
While recency, frequency, and monetary value tend to be most important in influencing response, different catalogs may have other equally important factors that require inclusion in the segmentation scheme. Frequently, the gender of the customer will result in differential response. In one client's catalog that featured merchandise appealing to men, past women buyers responded at only 60% of the rate of men. In fact, when you combined this reduced response ratio with RFM ratios, you could not justify ever mailing another catalog to a woman who had purchased below $50. Considering that little profit was made on any order under $50, the client was temped to put a notice in the catalog that women buyers who were purchasing under $50 were not welcome!
We have occasionally found the following factors to justify additional segmentation: gender, male or female products, in-door or out-door products, multi-product or single product orders, full-price or off-price, original source of name, and geography. In reality, there are a limited number of factors that can be effectively managed in a segmentation scheme and you must pick the three to five that are most influential. When your file grows to several hundred thousand buyers and there is a multitude of factors to assimilate, it is time to move on to house file regression modeling. But for most small catalog companies, RFM+ segmentation is sufficient.
Conclusion
Micro-segmenting your buyer file, even when some segments may contain a small number of names, is highly productive. We have used ratios computed for a fictitious catalog. However, you can compute similar ratios for your catalog and put them to use. Based on what you know to be general ratios, you can suppress from mailing small numbers of names from recent periods that are likely to be unproductive and mail segments in older periods that will prove responsive. You can make this a mathematical process or simply have the ratios in your mind as you make "mail/do not mail" decisions segment by segment coming out of a merge-purge. Recognize that this is not a precise process and assumes a mathematical relationship that may only exist in degrees. However, it is a very useful tool. Generally, a 10% to 20% lift to response rates can be achieved for the same mail quantity when this approach is applied to a mailer that has taken an "all or nothing" approach based on recency.
Of course, we do not simply ignore mailing low-performing segments. In the next column appearing in the August issue of CatalogAge, I will address a variety of techniques to enhance the response of low performing segments using the merge itself or vendors such as Abacus.