Post-Purchase Survey Attribution for Long Buying Cycles

Ask a buyer on the confirmation screen and 55 to 70 percent of them answer. Ask the same question in an email two days later and 8 to 15 percent do, drawn mostly from people who already liked you. That one placement decision determines whether a survey is evidence or decoration, and it is the first thing this engagement settles. What the survey is for: It is not a second attribution model. Platform data records exposure. The survey records recall. Where the two disagree, and on brand search, podcast and offline they always disagree, the survey shows which of your paid channels is buying credit for demand it did not create. That is the question I design it to answer, and the readout is written so it cannot be quoted as anything larger. How the instrument is built: One question, first screen, before the receipt. Answer options rotate per session so the top option does not absorb position bias. Google is never an option on its own; it splits into search, video and a shopping surface, because the undivided bucket has ended more measurement arguments in the wrong direction than any other field on the page. A free-text box stays open and gets coded by hand each month; codes that recur become options the following quarter. Responses are written against the order id rather than pooled inside a vendor dashboard, so a channel's share can be re-weighted by revenue, market and product line instead of by headcount. Respondents skew, usually toward first-time buyers, desktop and one language. The weighting model corrects the skews we can see and the readout names the ones we cannot. Sample size, plainly: A channel needs roughly 380 responses in a period before its share moves outside about five points. Below 600 orders a month the survey is worth running for the free text and nothing else, and I will say so before you buy it rather than after. By sector: Travel and hospitality: the question sits in the booking confirmation, not the post-stay email. A survey sent after the stay measures the stay. Logistics: the question attaches to quote acceptance inside the CRM and is asked of the person who signed, who is frequently not the person who filled the form. With 30 to 90 day lead times, responses are coded to quote month. Pharma and medtech: patient-facing surveys go nowhere without medical and legal review, and an open text field creates an adverse event reporting duty someone has to staff. In most cases I recommend HCP-side collection through the rep portal, or not running one. What is not included: No incrementality claim. A survey is not a holdout and I will not let it be presented as one; where a causal read is needed that is a geo test and a different engagement. No licence resale. No integration with a checkout your team cannot deploy code to. Who this is not for: Teams that need the survey to confirm a budget decision already taken. The instrument answers whatever it is asked, and a leading option set hands you the answer you wanted, which is worse than having no data at all. What you get: The question set and rotation logic, the vendor configuration or a custom endpoint writing to BigQuery, the weighting model as SQL you own outright, and a monthly channel view placed beside platform-reported numbers with each difference named rather than averaged away.