A customer buys a candle after visiting your market stall, watching a creator’s video, and searching for your shop. Your analytics report may show a search visit. The customer may say a friend introduced them. Neither account necessarily describes the whole decision.
A post-purchase attribution survey adds something click reports cannot supply: the buyer’s remembered explanation. Used carefully, it can uncover discovery through markets, gifts, shops, and conversations. Used carelessly, it becomes another report claiming the same orders.
The practical goal is not to replace analytics. It is to ask a clear question, preserve the answer, and use disagreement to decide what to investigate next.
Ask about discovery separately from the purchase trigger
“How did you hear about us?” can mean first discovery, most recent contact, or the reason someone ordered today. Those are different parts of the customer journey.
Start with this optional question after the order is complete:
Where do you remember first discovering our brand?
Offer a short set of relevant choices: a friend or family member; a market or event; a shop carrying the product; a social platform or creator; a search engine; another source; and “I don’t remember.” Allow an optional detail field for the event, shop, platform, or creator name.
If you also need to understand the closing action, ask separately:
What most influenced your decision to place this order today?
Possible answers include trying the product in person, a recommendation, a reminder, needing a refill, a particular product becoming available, an offer, and another reason. Include an uncertainty option here too.
Do not combine the answers into one “winning channel.” A retailer may introduce the brand while a refill need triggers an online order months later.
Make the question easy to answer honestly
Keep participation optional and separate from completing the purchase. Do not require a response to view the order confirmation or obtain support. Explain the purpose simply: “This helps us understand how people find our products.”
Avoid asking, “Which of our ads convinced you to buy?” That assumes an ad was involved and that the buyer can identify its effect. Someone who remembers Instagram may not know whether they saw a paid ad, an unpaid post, or a friend’s shared story.
Keep those distinctions out of the first question unless customers can answer reliably. “Instagram, type unknown” is more useful than a guessed paid-social classification.
Pew Research Center’s survey-question guidance emphasizes clear question development, testing, and consistency when measuring change. Apply that discipline at maker scale: ask a few buyers what they think your question means before using the responses in a budget review.
Once the wording works, keep it stable. Record the date when you change choices, placement, or wording so a reporting shift is not mistaken for a marketing shift.
Create an answer record that preserves uncertainty
Use a small internal worksheet rather than overwriting the source already recorded by your store or analytics tool. Suggested fields are:
- Order reference and order date, where linking is appropriate under your privacy practices.
- First-order or repeat-order status using a consistent business definition.
- Survey version and whether the invitation was shown.
- Original discovery answer and optional detail.
- Normalized discovery category, including unknown and ambiguous categories.
- Purchase-trigger answer, stored separately.
- Review note explaining any uncertain classification.
Retain the original wording. If someone writes “a video my sister sent,” do not silently turn that into paid social. You may classify it as a shared recommendation with a video detail, while recording that the original publisher is unknown.
Keep access limited to the people who need it. Do not request unnecessary personal details, send raw free-text answers into analytics, or place customer identifiers in tracking links. A useful marketing measurement record does not need a customer’s full personal story.
Read response rates before channel shares
Suppose a soap business shows the survey to 200 eligible first-order buyers. Fifty answer the discovery question, and 20 of those respondents name a market or event. These are illustrative figures, not research findings.
Response rate = completed discovery answers ÷ eligible buyers shown the question. Here, 50 ÷ 200 = 25%.
Market share among respondents = market answers ÷ completed discovery answers. Here, 20 ÷ 50 = 40%.
The defensible statement is: “Forty percent of respondents remembered a market or event.” It is not: “Markets generated forty percent of all first orders.” The other buyers did not provide that evidence, and respondents may differ from nonrespondents.
Keep “I don’t remember” inside the completed-answer total and report it separately. Keep unanswered invitations separate again. Removing uncertainty from the denominator makes the named channels look more certain than they are.
If you do not know how many buyers actually saw the invitation, label any order-based response ratio accordingly. Do not present it as an invitation response rate.
Compare survey answers with analytics without forcing agreement
Place three views beside each other: actual orders, reported traffic sources, and remembered discovery. Do not add their order or revenue totals together.
Google Analytics distinguishes first-user, session, and event-scoped channel dimensions. These answer different questions. Its reporting identity documentation also explains how cross-device reporting depends on available identifiers and settings. A buyer’s recollection is different evidence again.
If market discovery appears repeatedly while analytics shows search purchases, keep both observations. Check whether the named events occurred, whether the answers come from first-time buyers, and whether the pattern persists beyond one enthusiastic response.
The next action might be testing clearer event links or collecting a longer period of consistent responses. It should not automatically be doubling the market budget. For the broader distinction between credit and cause, use the small-business marketing attribution guide.
Frequently asked questions
Should I offer a reward for answering?
Start without one. A reward changes who participates and the cost of collection. If you introduce one, document it and avoid treating the new response pattern as directly comparable with the earlier survey.
Can I ask customers at a market instead?
Yes, but keep those answers separate from online post-purchase responses. The setting and eligible audience differ. A conversation at your stall cannot represent everyone who later buys online.
What if a buyer names several sources?
Preserve all named sources without counting multiple customers. If you report source mentions, label that denominator clearly; mention percentages may total more than the customer count implies.
Should survey answers change my CAC calculation?
Not automatically. Preserve a consistent cost scope and new-customer denominator in your customer acquisition cost calculation. Use survey patterns to question channel assumptions, not to create extra acquired customers.
A practical next step
Pilot the two questions with a few buyers. Write your coding rules, unknown categories, eligible audience, and review date before collecting the next group of answers. You will have a modest but useful second view of discovery—not a claim to know everything that caused a sale.
