Post-purchase surveys as an attribution input

A post-purchase survey asks the buyer where they heard about you, at the moment they are most likely to remember. It captures demand sources no pixel can see: podcasts, a friend’s recommendation, a creator mentioned in passing. It is directional evidence about how people discover you, not a measurement system.

What is a post-purchase survey actually good at?

Its real strength is channel discovery. Tracking can only report on channels that produce a click or a view it can observe. A survey reports on the ones that do not. If a podcast read, a physical event or plain word of mouth is driving demand, the survey is often the only place it shows up, and it can surface a source you were not spending against and did not know existed.

It is also useful for catching movement. When answers shift steadily over several weeks, that is worth investigating even before you know why. A new source climbing the list can prompt a test. A source falling away can point to a dip in traffic that the ad accounts do not explain on their own.

Where does post-purchase survey data mislead?

Recency bias is the main problem. People name the most recent thing they remember, usually the last touch, often a branded search or a retargeting ad. A podcast heard six weeks ago rarely gets named, even when it created the demand.

Response rates are the second problem. Only a share of buyers answer, and they are not a random sample of the ones who bought. Answers cluster among engaged customers, which quietly overstates the channels those customers favor.

The option list is the third. Whatever you put on it gets chosen, and whatever you leave off does not. A list naming four channels plus “other” returns four channels’ worth of answers and a bucket you cannot use.

None of this makes the survey worthless. It makes it a source that needs context, and the way we fit it alongside pixel and blended data is set out at plaidtesting.com/services/tracking-attribution/.

How should you word the question and the options?

Ask one open question, phrased around discovery rather than the last click: “How did you first hear about us?” The word “first” does real work, since it pushes memory back past the retargeting ad. Ask it once, right after the order, on the confirmation page or in the first email.

Keep the option list short, concrete and stable:

• Six to eight options, since long lists get skimmed and the top entries win
• Name real sources a buyer would recognize, not internal channel labels
• Change the list rarely, since every edit breaks comparability with earlier months
• Leave a free text box, then actually read it once a month

Then leave the wording alone. A question rewritten every quarter produces a data set you cannot compare with itself.

How should survey data sit next to your other numbers?

Treat it as one of three imperfect witnesses. Platform reporting knows clicks. Blended numbers know total revenue against total spend but not which channel earned it. The survey knows what the customer believes, which is sometimes the only pointer you have.

Use it to raise questions, not to settle them. A channel that shows well in the survey and poorly in platform reporting is a reason to design a proper test, not an excuse to move money that week. Read it as a trend line over months, and never move a large budget on survey share alone.

Across six months with an athletic apparel brand, we grew sales 35.7% year to date to $9.27M while paid media investment rose 42% year over year, and blended ROAS held at 3.36x. That was a deliberate decision to scale rather than squeeze efficiency, measured on blended ROAS, not platform reported ROAS.

If you already run a survey, the next step is to freeze the option list for a full quarter so the trend becomes readable. Plaid Testing offers a free thirty minute growth audit if you want a second read on how your inputs fit together: tracking, account structure, and three fixes.