Most DTC stores I audit either don’t run a post-purchase survey or run one that asks too much. The common version is a multi-question form that loads after checkout, asks for product feedback, attribution, NPS, and how the customer wants to be contacted. Response rate sits at 8 to 12%. The data is messy and gets ignored.
The version that works is one question. “How did you first hear about us?” That’s it. Multiple choice with 6 to 8 options plus an “Other (please specify)” field. Response rate jumps to 30 to 40%.
The reason this matters for attribution is that it gives you a self-reported channel mix that doesn’t depend on pixels firing or platforms claiming. When 35% of customers say they first heard about the brand from a friend, that’s 35% of acquisition you can’t see in any ad platform dashboard. When 15% say “TikTok” but TikTok Ads claims 35%, you’ve got a real read on where the gap is.
The mistakes I see operators make with this data.
Treating the survey as ground truth. It isn’t. Customers forget. They round up. They say “Google” when they actually clicked an Instagram ad and then searched the brand name. The survey is one input, not the answer.
Mixing the survey with pixel data without separating them. If you average the two, you get a number that’s wrong in both directions. Better to look at them side by side and ask why they disagree.
Not segmenting by customer LTV. The channels that bring in high-LTV customers aren’t always the channels that bring in the most customers. A 5% channel by acquisition that has 2x average LTV is a different story than a 30% channel that’s all one-and-done.
I run survey responses through ThoughtMetric so I can cross-reference them against pixel-based attribution and look at LTV by reported channel. The interesting numbers are usually the ones where survey-reported share is much higher than pixel-attributed share. That gap is where the brand is actually working, and where the platforms can’t see it.
Set up the survey as a single question, present it on the order confirmation page, and run it for 90 days before you trust the read. That’s enough volume to start seeing the patterns hold.
Leave a Reply