Congratulations on the Donation You Were Getting Anyway

November 20, 2026      Kevin Schulman, Founder, DonorVoice and DVCanvass

If you hired someone to bring guests to a party and instead, they show up and bill you for those showing up anway, you’d feel cheated.

A study in Management Science examined this in programmatic advertising, where platforms decide who sees your ads and how much to bid to reach them.  They worked with an electronics store and randomly assigning 208,538 website visitors to see either the retailer’s retargeting ads or unrelated charity ads.

The retailer ads worked, roughly 5.4% of those seeing charity ads (i.e. control group) bought vs. 5.6% of those shown retailer ads.  But the profit from those additional purchases covered only about 54% of the platform’s fees.  So, “yes” on incremental sales but “no” on it being a good use of dollars.

Why didn’t it work better?  The platform was charging the retailer for sales that were going to happen anyway, bidding more to reach what it deemed to be likely buyers.

Being likely to give and being more likely to give because you advertised are two different things. A platform can excel at identifying the first without doing much of the second. As prediction improves, it can get even better at finding people whose behavior needs very little changing. More sophisticated technology can make the same expensive mistake with greater precision.

We recently ran an geo-incrementality test of Google’s Performance Max with certain zips getting PMax ads and a matching set of zips not.

We excluded existing donors with a suppression list, refreshed that exclusion nightly and excluded branded searches.

The estimated gain was eight additional donors on $4,616 of spend: $577 per incremental donor. The difference was small enough that the chance of it not being incremental at all couldn’t be ruled out.  But taking the estimated gain at face value is another case of “yes” but, “no”.  The donor’s expected future value still has to justify the acquisition cost.

We also found that 94% of PMax spending went to Search. Despite access to Gmail, Display and YouTube, almost all the money went toward people already expressing interest through a search. That allocation makes sense when you’re optimizing for credited conversions. Search offers plenty of opportunities to reach someone shortly before they give and count the resulting donation as a success.

PMax reported a cost per donor of $83, that’s about 7x lower than the more accurate estimate.

What to do with this?  Let’s start with what not to do, don’t upload your best donors and request the machine find more people like them.  It will and you will pay for all of them, including those who were going to give without your ad spend.

Instead, hold a randomly selected group out of the advertising, or use a properly designed geographic test. Compare total donor and revenue results, including gifts made through other routes, then calculate what you paid for the difference. Judge that cost against a defensible estimate of donor value. Use the findings to change who gets advertised to, how much you spend or whether the campaign continues.

This requires accepting that fewer credited conversions might accompany a better decision.

Before approving the next budget, ask for the cost per additional donor and the evidence behind it. If the answer is another screenshot of attributed conversions, you still don’t know what you’re buying. You do know who’s getting paid.

Kevin

 

Leave a Reply

Your email address will not be published. Required fields are marked *

Leave a Reply

Your email address will not be published. Required fields are marked *