Returns: The Hidden Cost of Retail
Product returns are one of the most consequential topics in retail — and one of the least studied. The National Retail Federation estimates that US consumers returned over $740 billion in merchandise in 2023, and return policies have become a decisive factor in purchase decisions. Yet most retailers rely on internal transaction data to understand returns, missing the attitudinal and behavioral dimensions that drive consumer choices.
The Happy Returns (UPS) and National Retail Federation 2024 Consumer Returns in Retail study is one of the few large-scale consumer surveys dedicated to return behavior. Fielded in August–September 2024 with n=2,007 US consumers who had returned at least one online purchase in the past year, it covers holiday return planning, free return sensitivity, printerless return preferences, and bracketing behavior. For Simsurveys, it provides a valuable benchmark for our Consumer model in a specific, commercially important domain.
Study Design
We generated n=2,007 synthetic respondents using the Simsurveys Consumer model, matching the sample size and screening criteria of the original study. The simulation covered all 14 survey items on holiday return behavior, return policy sensitivity, free returns, printerless options, and bracketing.
For single-select questions with full response distributions, we measured alignment using KL Divergence. For attitudinal agreement items, we also compared top-2-box scores (the percentage of respondents selecting the top two agreement options) to provide a metric that is directly actionable for retail research teams.
Results: Strong Directional Alignment
The Consumer model demonstrated strong directional alignment with the live study across all 14 items. On the question of whether holiday return policies influence purchase decisions (Q1), the model achieved a KL Divergence of 0.036 — with 60% of simulated respondents saying they are more likely to buy from retailers with easy return policies, compared to 59% in the live data.
Return timing (Q2) showed good overall alignment with a KL Divergence of 0.086, though with an interesting distributional shift. The live data showed 46% of consumers returning within 2 weeks to 1 month, while the simulation estimated 40%. Conversely, 33% of live respondents returned within 1 week versus 51% in the simulation. The model correctly identified that most returns happen quickly, but overestimated the speed of the fastest returners.
Holiday return policy impact: KL Divergence of 0.036. Live data: 59% more likely to purchase with easy returns. Simulated: 60%. Near-perfect alignment on the headline metric.
The Stated Preference Amplification Effect
The most consistent pattern in this validation is what we call stated preference amplification. On attitudinal items — particularly around free returns — the simulated respondents show somewhat stronger stated preferences than live respondents. The top-2-box agreement on free returns was 76% in the live data versus 88% in the simulation, a gap of 12 percentage points.
This pattern is interpretable and consistent with what we observe across consumer validations. Synthetic respondents tend to express clearer, more decisive preferences on topics where consumer sentiment is already strong. When 76% of real consumers already agree that free returns matter, the model amplifies this to 88% — it captures the direction and strength of sentiment correctly but overstates the intensity.
For retail research teams, this means the Consumer model is highly reliable for identifying which return features matter most to consumers and how they rank relative to each other. The rank ordering of consumer preferences — which features drive the most purchase consideration, which return methods are preferred — is well-preserved. Absolute percentage estimates on attitudinal items should be interpreted as directionally accurate with a known upward bias on strong-sentiment topics.
Behavioral vs. Attitudinal Questions
An interesting secondary finding is the distinction between behavioral and attitudinal question performance. Behavioral questions — such as return timing and policy consideration — showed tighter distributional alignment. Attitudinal questions — such as agreement with the importance of free returns — showed the preference amplification effect described above.
This distinction matters for research design. When using synthetic consumer data for returns research, behavioral questions about what consumers actually do will tend to produce tighter estimates than attitudinal questions about how consumers feel. Both are useful, but the confidence intervals differ.
Implications for Retail Research
The returns category is growing in strategic importance as retailers tighten return policies and consumers increasingly factor returns into purchase decisions. The Simsurveys Consumer model provides a way to rapidly test return policy scenarios, evaluate consumer sensitivity to different return features, and benchmark return experience metrics — all without the cost and timeline of fielding a dedicated consumer panel study.
The full validation report, including item-level distributions and metric summaries, is available for download. For more on the Consumer model and its applications, visit the Consumer model page.