Why should you rate your buyers? There are many reasons. One important reason is simple economics. If you are running a mature lottery, and have not been keeping track of lapsed buyers, you may be spending a huge amount of money for postage and materials on people who are not responding. You may want to know, year by year, how certain groups of people are responding. For example, how many new buyers are coming back the next year. You may want to send targeted letters to various classes of buyers at the start of your campaign.

Everybody knows, at least instinctively, that the most responsive lottery players are those that have bought tickets in the past. You know, because you see their names every year. But have you created a historical database of all the purchases? Can you look someone up in the database, and instantly see their purchase history? Can you send thank-you letters to your best customers?

RFM Analysis (Recency, Frequency, Monetary value) has been described in many places, including the pages of Canadian Fundraiser. For a quick recap: the theory goes that your most recent buyers will be the most likely to purchase again. You take your database, sort it by recency, split it into 5 equal parts ("quintiles"), and assign the numbers 5 to 1 to the quintiles; 5 being most recent, and 1 being least recent. The same goes for frequency. Within each recency quintile, the most frequent buyers are most likely to buy again. Take each of the 5 recency quintiles, split them again into 5 equal parts, and assign the numbers 5 to 1 in order of frequency. Do the same again for the monetary value. What you wind up with is 125 cells, coded as 555,554,... down to 111. Each cell has the same number of records, or close to the same. People coded as 555 are usually your best customers, whereas people coded as 111 are generally the least likely to buy again.

When applied to lotteries, ‘Recency’ equates to which lottery they purchased in; ‘Frequency’ to the number of previous lotteries the person has bought in; and ‘Monetary’ represents the total number of tickets bought. Because most people buy a single ticket per lottery, and each ticket costs the same, ‘Frequency’ and ‘Monetary’ values have a very strong correlation.

RFM coding is easy to implement if you have a reasonably proficient programmer or database administrator available. However, it is not useful by itself until your raffle has had at least 7 campaigns.

Experience rating will tell you, historically, what the response rates were for various groups of people, based on their buying patterns. It will let you answer all the following questions and more:

1. How many first-time buyers from Lottery 2 responded in Lottery 3? What percentage? How does this compare to the first-time buyers from Lottery 1?

2. How many consistent, long-term buyers do you have? What percentage bought tickets in the last lottery?

3. How much does the response rate drop if people do not buy a ticket in one campaign? Two campaigns?

Experience-rating can be done in all sorts of ways; our preferred method is to generate a ‘signature’ for each buyer, using their purchase history.

What is a signature? It’s simply a set of characters that represent whether or not a purchase was made. Each position represents one lottery campaign. Start with a blank field. For the first lottery a person buys in, record a ‘Y’ in the field. From then on, append either a ‘Y’ or an ‘N’ for each subsequent lottery they bought or did not buy in. You will end   up with a series of Y's and N's, which together are the buyer's signature.

Take a hypothetical example. Assume you have had three past lotteries - Lottery 1, Lottery 2, and Lottery 3. If 'Joe' bought in every one of the events, then his signature would be ‘YYY’. If he missed Lottery 2, then his signature would be ‘YNY’. If he only bought in Lottery 2, then his signature would be ‘YN’.

Every person on the master file has a signature. By analyzing these signatures appropriately, experience rating computes the historical response rates for each signature class. The historical rates may also be used to estimate what the next campaign's response rates will be, with certain limitations. The calculations and assumptions required are beyond the scope of this article.

In our work, we generally calculate the response rates for groups based on whether they are single or multiple ticket buyers, and an overall total. However, given sufficient data, groups can be broken down further by other factors such as gender or age. There are solid indications of gender and age differentiation in response rates. However, we have not performed a complete analysis on this. Nonetheless, we think it's important to track your buyers' age and gender.

The following table represents our hypothetical example. It shows the response rates for buyers from Lottery 1 over the course of three campaigns. Please note these figures are for illustration purposes only.

Looking at the table, you can see that the most recent and frequent buyers have the highest response rates. Additionally, compared to the overall total, multiple-ticket buyers have significantly higher response rates. However, remember that only perhaps 10% of buyers will purchase more than one ticket (at least with the standard $100 tickets).

To illustrate these concepts in more detail, we projected the hypothetical information an extra few years to create a model of a 'mature' lottery. We wanted to compare RFM analysis with experience rating, and used our model to calculate the average response rates for each RFM cell.

The following graph shows the experience-rated response rates (bars), and the percentage of the total predicted response (line) compared to the percentage of the total mailout. Each bar represents one RFM cell. That is; each bar contains 1/125 of the mailing list.

To show this chart in full detail, click here.

Note that the RFM codes in the graph above were chosen only by sorting the data according to recency, frequency, and monetary value, with no regard to the experience rated response rates. As expected, the most recent & frequent buyers have high response rates. That they show such a strong correlation illustrates the potential of simply coding your database for RFM.

Probably the first thing you should do once your database has been experience-rated is to identify all the lapsed buyers. That is, all the buyers that are no longer profitable to mail to. You can eliminate these people from the standard campaign mailings and save perhaps thousands of dollars in postage and materials.

You might think that as time goes on and your mailing list grows, that there will come a point where the responses from your mailing list alone will be enough to sell out your lottery. We believe this to be virtually impossible. In our most optimistic study models the mailing list alone produces at most 70%-80% of the total ticket sales. Under real-life conditions, we believe that a normal mailing list would produce at most 50% of the total sales.

We recommend using both experience rating and RFM when analyzing your data. RFM codes alone do not tell you much, unless you track results year over year and determine your response rates. The standard 125-cell breakdown is not useful until your lottery has been running long enough to create significant differentiation in your buyer profiles.

The data, response rates, and graphs used in this article are generated from models that we use for estimating response rates and should not be used to infer any specific information about your own mailing lists. Each lottery has its own unique characteristics and the data we presented here is developed using a general, non-specific model that attempts to illustrate basic industry patterns.