Beyond RFM – doing intermediate-level segmentation

By now, hopefully, you have seen the benefits of segmentation and the value of RFM as a preliminary tool.  But you have probably intuited a few of the flaws in it already:

  • The distinction between recency groups is sometimes artificial. Let’s say two people give $20 per year, one at the end of January and the other at the beginning of February.  The former would be at 13 months for a January mailing; the latter at 12 months.  A few days difference could mean that one gets mailed and the other didn’t.  (In fact, I’d love to test a segmentation out to 13 months instead of the standard 12 because of this; please comment or email me if you’ve done this, so I can share and illuminate myself).
  • The distinction between frequency groups is artificial. Your multidonors range from people who donate once a year over a series of years to people who donate to literally every communication they get.  RFM analysis gives these two people the same number of communications.
  • The distinction between monetary value is artificial. You probably saw this one coming, given the first two.  Which donor would you prefer – a donor who donates $10 ten times per year or a donor who donates $50 once a year?  RFM prefers the latter; I’m guessing you would prefer the former.
  • It carries literally no other information. The $25 dollar donation of today could be from someone who clipped coupons to raise the money to donate or Bill Gates divesting himself of what he found in one cushion of his couch.  What a person donates is an indicator of capacity, but it’s a blunt tool when a scalpel is possible.
  • You probably noticed I was talking only about the mail in examples. RFM doesn’t look at channels of donation, nor at the sensitivity of people to those channels.

Let’s address this last one first, as we look for ways to customize RFM more than using it alone.

Separate RFMs by channel. When you do your telemarketing list pull, you will almost certainly want to call deeper into your file to people who have made a previous telemarketing gift than those who you are trying to make their first gift on the phone.  Same thing in the mail: you may be willing to mail multidonors out to 48 months $50-99.999 for people who have given through the mail, but only to 12 months or 18 months $50-99.999 for people who have only given online or on the phone.  Separate RFMs by channel will help you make these determinations.  Remember that you are looking to make sure that you have a strong multi- or omni-channel program, but that doesn’t mean you have to be agnostic as to channel of origin or preference.

Cadence analysis.  This goes to your $10 10x versus $50 1x donor example above.  You have to figure out who likes/needs multiple communications to make a gift or gifts and who doesn’t.  Some ways you can do this:

  • Look at how many types of offers you have going to a person in rotation; hammering the same key on a piano over and over isn’t music, nor will the same approach be music to your donor’s ears.
  • Take a look at how many times you mail someone in acquisition. Sometimes chronic non-responders need a different offer (as with rotating your offers above), but you can also see whether someone after seven or 17 or 27 times of getting a message from you means they almost certainly won’t donate and you can bless and release.  The same holds true for your lapsed donors as well.
  • Test consistent communication versus no communication versus resting periods to see what happens when you try different cadences.
  • Look at frequency of gifts within a certain period of time, rather than ever. You can see dramatic results sometimes by decreasing your mailings to your one-time-per-year donors versus your multi-per-year donors.

Mission area: I mentioned this when discussing customization types, but tailoring your communications toward the area of your donor’s interest is also a legitimate segmentation target.  Why would you send that advocacy alert to people who care only about your work in schools or your calendar with pictures of dogs to your car people?

Income: If you don’t want to ask that millionaire for $17, a wealth append can get you the information you need to customize your ask strings and communications strategy.

Location: Zip codes are the poor person’s income append and can be free-ish, so that’s a potential win.  More often, though, you can use zip code modeling to breathe life into underperforming acquisition lists.  Simply find your top X percent of zip codes from your current file and ask for just those zip codes from the rental list.  It will be slightly more expensive to rent per name this way, but it can provide a 20%+ lift in response rates and/or average gifts.  This is also a way to test new lists while minimizing risk.  One caution: you don’t want to do this for all lists or you risk self-limiting your acquisition strategy.  If a list works well for you across zip codes, use the whole thing – that way you give yourself a chance to be wrong about a zip code long-term.

Demographics: Some nonprofits find that different messages work better for men versus women.  Age can also be helpful, as you work to avoid sending your one millennial donor your planned giving brochure (there’s optimistic and there’s delusional…).  With demographics, backtesting makes initial sense, where you see how people of different demographics responded to previous appeals and messages, then use that data to define your strategy.

Previous responsiveness: It sounds obvious, but it’s ignored by CRM: if someone like getting your calendar three years ago, they may like getting it again.  Replace “calendar” with member card, action alert, survey, etc., and you have the makings of a profitable add on to your usual list mix.

Those are some of the things you can add to spice up RFM.

I said at the beginning that this was intermediate segmentation.  Advanced segmentation is modeling.  The hacks above will help get you many of the benefits of modeling at a fraction of the cost, but it won’t get you all of the benefits, so definitely leave yourself open to building smarter and smarter donor modeling solutions.

Any other segmentation recommendations you’ve seen work?  Please leave them in the comments.

Beyond RFM – doing intermediate-level segmentation

4 thoughts on “Beyond RFM – doing intermediate-level segmentation

  1. Lauren says:

    I use both previous responsiveness and mission area, but combining them has proven most successful for me! We separate those who responded to that campaign to ensure they get that same piece again. Then we make sure to acknowledge their response to that campaign last year.

    I performed some testing around this technique, identifying respondents to two different package types at the same mail date then splitting those audiences for testing, one half getting the same package they responded to and the other receiving the opposite package. For both package types, the donors who got the same package they responded to last year performed better.

    This isn’t really a new recommendation, more a further validation of your recommendations, but thought it would be worthwhile to share!

    Liked by 1 person

    1. This is absolutely a great test and thank you for sharing it! The good thing about donor behavior is that the donor is tell you what they like and what they don’t like by “voting for” the communication with their donation. We just did this with our calendar and found the people who had responded to the calendar three years ago, but literally nothing since then, still had double the response rate of the donors selected by RFM audience. In the future, we’re going to be looking at whether you can go even deeper in previous respondents.

      However, I hadn’t split tested them (our quantities are very small of the previous respondents who weren’t already in because of their RFM). That’s a new wrinkle for me and it’s a far better test than what I’d run, so that’s really positive to hear. Thanks for sharing!

      Like

      1. Lauren says:

        Glad it was helpful. Another note on this: I have also used this testing as an added data point in the decision to roll out a test or retest as a control. So looking at if the test has the staying power year-over-year with donors who didn’t receive it last year and those who did. Not to be used on its own, but can’t hurt to add to the normal list of metrics we look at in that determination.

        Like

      2. Interesting… so if the test works well with the people who donated to the control, that lends support to the idea that the test should be rolled out with. If it doesn’t, you may have a segment of people who prefer the old version and perhaps you have dual controls on your hands. An interesting way of looking at it. Thanks!

        Like

Please share your thoughts.

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s