You want to find out who is going to be the next president.  You don’t talk to every likely (or eligible) voter – too expensive.  Instead, you find a representative sample of Americans, ask them and extrapolate, building in error percentage for this extrapolation.

This type of sampling is mighty.  You can learn what segments to use, identities to target, and messages to roll out with.  We use it in our pre-test tool, where sampling 400 donors or prospects can give you a read of how donors and prospects will respond globally, and our commitment studies of donor populations.

But sampling fails in two different circumstances.  One is obvious: when you need to count people.  That is, since you need the number of people total to extrapolate, you can’t extrapolate the number of people there are total.

That’s why the census is a census.  While there is sampling involved, some basic questions need to be asked of everyone for a nose count.

The other is when you need to get specific information mapped to a specific person.  If it’s vital to your program to know who is a cat person or dog person, because that’s your primary donor identity, it doesn’t help much to know that your file is 53% cat, 47% dog (unless you want to send out a mail piece that alienates about half of your file).  You need to know that John Doe gets the dog letter and Jane Roe gets the cat letter.

Donor feedback is part sample and part census.  For benchmarking your satisfaction rates, analyzing where you have areas for improvement, and testing your brand, a sample is sufficient.  You want to get a feel for these things and to see which direction they are moving.

But there are some variables that you want to do by donor, for every donor.  These likely include:

• Donor identity for your most important segmentation points (if you don’t know what identities are most important, you can brainstorm them from yesterday’s Agitator post)
• Commitment to your organization
• Donor preferences by channel and frequency

These are all valuable, vital for segmentation, and relatively unchanging.

To get these answered, a simple two-point plan suffices: 1) ask until you get them; 2) stop asking once you get them.

Consider how much effort you put into getting LYBUNT (last year, but not this year) donors to give again, lest they lapse.  That’s the level of effort to put into getting these key differentiators, as they may be the difference between giving ago and not.

But we’ve also seen mail reply forms that have an email capture printed on them… all of them.  Even if you have given your email address and are enjoying the emails you are getting.

This is a clear sign that you don’t know the donor and you haven’t customized the experience to her/his liking.  Moreover, it’s an indication that any feedback you give will be treated with similar contempt.

What other variables do you think enter the pantheon of census variables?