Use post-purchase surveys to understand where customers truly discovered your brand, then calibrate those self-reported signals against platform data.
Survey-driven attribution asks customers directly: "How did you hear about us?"
This self-reported data captures channels that digital tracking misses entirely --
word-of-mouth, podcasts, billboards, and organic social impressions that never generate a click.
While platform pixels track clicks and views, surveys capture the perceived influence
behind a purchase decision. Combined with platform data, they form a powerful
triangulation layer in your measurement stack.
Dark Funnel Visibility
Captures channels invisible to digital tracking -- podcasts, word-of-mouth, offline media, and organic social.
Customer Voice
Reflects actual perceived influence rather than algorithmic guesswork. Customers tell you what mattered.
Platform Calibration
Cross-reference self-reported data against platform-reported conversions to find over- and under-counted channels.
Low Implementation Cost
A single post-purchase question can be deployed in days. No pixel infrastructure or data warehouse required.
Important limitation: Survey attribution relies on customer memory and honesty.
Response bias, recency bias, and social desirability effects can distort results.
Always calibrate survey data against at least one other measurement method before making budget decisions.
Step-by-Step Process
Deploy post-purchase survey -- add a "How did you hear about us?" question to your order confirmation or thank-you page.
Collect responses -- accumulate data over a sufficient period (typically 2-4 weeks minimum) to reach statistical significance.
Calculate survey-attributed revenue -- multiply each channel's response share by total revenue from respondents.
Compute implied revenue -- adjust for response rate to estimate total channel revenue across all customers.
Calculate implied ROAS -- divide implied revenue by ad spend to get the survey-based return on ad spend.
Key Concept
Survey ROAS captures the customer-perceived return on your spend. When this diverges
significantly from platform-reported ROAS, it signals either platform over-counting or a channel
whose influence extends beyond trackable clicks.
Interactive Calculator
Results
Channel
Ad Spend
Survey Revenue
Implied Revenue
Implied ROAS
Calibration bridges the gap between survey-attributed revenue and platform-reported revenue.
The normalization multiplier tells you how much each platform over- or under-reports relative to customer perception.
A multiplier > 1.0 means the platform under-reports its true influence.
A multiplier < 1.0 means the platform over-reports (takes credit for organic conversions).
Platform Revenue Inputs
Enter each channel's platform-reported revenue. Default values assume a 1.4x platform ROAS on spend.
Calibration Results
Channel
Survey ROAS
Platform ROAS
Multiplier (%)
Calibrated ROAS
Insights
Survey Quality Metrics
Before trusting survey results, assess the quality of your response data. These metrics help
identify systematic biases that could distort attribution.
38%
Response Rate
Percentage of customers who complete the survey. Below 20% risks non-response bias.
2.4s
Median Response Time
Time to answer the attribution question. Under 1 second suggests satisficing behavior.
12%
Multi-Select Rate
Share of respondents who select multiple channels. High rates indicate genuine multi-touch journeys.
4.2%
Other / Write-In Rate
Percentage choosing "Other." Above 10% means your option list has coverage gaps.
Response Bias Detection
Three common response styles can systematically distort survey attribution data.
Detecting these patterns allows you to apply corrections before drawing conclusions.
Extreme Response Style (ERS)
Tendency to always pick the first or most prominent option. Check if one channel consistently dominates regardless of actual exposure.
Moderacy Bias
Tendency to avoid extreme answers. In multi-select surveys, respondents may hedge by selecting several "safe" middle-ground channels.
Acquiescence Bias
Tendency to agree with all presented options. Manifests as unusually high multi-select rates where respondents check every channel shown.
Factorial Experiment Analytics
Run controlled experiments on survey design to isolate the effect of question wording,
option ordering, and response format on attribution outcomes.
Question Wording
"How did you first hear about us?" vs. "What most influenced your purchase?" can shift attribution by 15-25%.
Option Order Effects
Primacy bias inflates the first listed option by 8-12%. Randomize option order to control for this.
Single vs. Multi-Select
Single-select forces a "winner take all" model. Multi-select better captures multi-touch journeys but inflates total credit.
Open vs. Closed Format
Open-ended responses yield richer data but require NLP processing. Closed formats are easier to analyze at scale.
Bias Correction Methods
Survey data is inherently biased. The respondent pool differs from the full customer base
along observable (and unobservable) dimensions. These methods help recover unbiased estimates.
Non-Response Adjustment
Respondents differ from non-respondents. Weight survey data to match the full customer population on known demographics.
Recency Correction
Customers over-weight recent touchpoints. Apply temporal decay to normalize credit across the full journey window.
Salience Adjustment
Memorable channels (TV, influencer) get over-reported. Calibrate against click-stream data to deflate salience-inflated channels.
Social Desirability
Respondents under-report ad influence and over-report organic discovery. Use indirect questioning techniques to reduce this bias.
Weighting Methods
Cell Weighting
Divide respondents into demographic cells (age x gender x channel). Weight each cell so its
share matches the known population distribution. Simple and transparent, but breaks down
with many cells and sparse data.
Inverse Probability Weighting (IPW)
Model the probability of responding given observed covariates. Weight each respondent by
1 / P(respond). More flexible than cell weighting but sensitive to extreme weights.
Trim or winsorize weights above the 95th percentile.
Raking (Iterative Proportional Fitting)
Iteratively adjust weights so marginal distributions match known population totals.
Does not require cross-tabulated population data -- only marginals. The standard method
for survey weighting in market research.
Model-Based Approaches
When weighting alone is insufficient, model-based methods can account for unobserved confounders.
Multilevel Regression with Post-Stratification (MRP)
Fit a multilevel model predicting channel attribution from respondent characteristics,
then post-stratify predictions to the full population. Handles sparse cells well and
produces stable estimates even with low response rates.
Imputation Methods
For non-respondents, impute likely attribution based on observable behavior.
Hot-Deck Imputation
Match each non-respondent to a similar respondent and assign their attribution. Preserves the empirical distribution.
Multiple Imputation
Generate several plausible imputations to capture uncertainty. Combine estimates using Rubin's rules for valid inference.
Predictive Mean Matching
Use a regression model to predict attribution, then match to the closest observed value. Avoids impossible imputed values.
Survey Design Guide
Keep it short -- a single "How did you hear about us?" question with 6-10 options maximizes completion rate.
Randomize option order -- prevent primacy bias from inflating the first listed channel.
Include "Other" with write-in -- capture channels you haven't anticipated. Review write-ins monthly.