TL;DR
This paper tackles the question of how to optimize in-store product sampling for maximum sales impact. By developing a sophisticated sales-response model, the authors demonstrate that sampling provides both immediate and sustained sales lifts, outperforming traditional in-store displays, especially when conducted multiple times. This research offers actionable insights for manufacturers and retailers to strategically deploy sampling campaigns for greater profitability.
What Problem Does This Paper Address?
In-store product sampling is a widely used promotional tactic in the retail industry, particularly in grocery stores, aimed at driving sales and enhancing the customer experience. While its immediate impact on sales is generally accepted, there's a significant knowledge gap regarding its long-term effects, how these effects vary across different conditions, and how they compare to other in-store promotions like product displays. Manufacturers and retailers need to understand not just if sampling works, but when, where, and under what conditions it is most effective to justify the investment and maximize returns. This includes understanding the carryover effects, the impact of repeated sampling, and the influence of store characteristics and competitive environments.
Methods
The authors propose a flexible sales-response model that accounts for several key factors:
- Carryover Effects: They model the decay of sampling effects over time using a gamma distribution, allowing for various decay shapes beyond simple exponential decay.
- Endogenous Store Selection: The model explicitly addresses the potential bias that stores with higher expected sales lifts might be preferentially chosen for sampling events. This is handled by modeling the probability of sampling occurring in a store as a function of the expected sales lift.
- Multiple vs. Single Events: The model differentiates between the impact of a single sampling event and repeated events for the same product.
- Heterogeneity: It incorporates store-specific parameters and allows them to vary based on store characteristics (like ACV and number of competitors) to understand moderating effects.
- Inter-temporal and Cross-brand Dependencies: The model accounts for serial autocorrelation in sales and potential spillover effects of sampling on competitor brands.
The estimation is performed using hierarchical Bayesian methods, specifically a Markov Chain Monte Carlo (MCMC) algorithm, which is detailed in the appendix.
Datasets
The study utilizes six different scanner data sets:
- Five Manufacturer (Univariate) Data Sets: These datasets each contain a single in-store sampling event for a specific product. They cover various product categories (Diet/Health snacks, frozen snacks, snacks in a coffee shop chain) and provide data for different lengths of time post-sampling (9 to 31 weeks). The number of stores involved ranges from 23 to 223.
- One Retailer (Multivariate) Data Set: This dataset comes from a large grocery chain and includes multiple in-store sampling events for a single SKU over two years. This allows for a more in-depth analysis of repeated sampling effects.
The authors note that obtaining this type of fused data (scanner data combined with sampling event information) is challenging, as manufacturers often lack competitor data, and retailers are hesitant to share detailed information.
Key Insights and Ideas
- Sampling is a Quality Promotion: Unlike price promotions that can lead to brand switching and stockpiling, sampling is presented as a "quality event" that adds value and can lead to category expansion rather than pure substitution.
- Sustained Impact: In-store sampling isn't just a short-term sales boost; it has a significant and lasting carryover effect on sales, especially when conducted multiple times.
- Multiple Events Amplify Impact: While the immediate lift from a second sampling event might be slightly lower than the first, the overall duration and sustained impact of repeated sampling are considerably greater than single events.
- Localized Impact: The effect of in-store sampling appears to be localized, with the number of competitors not significantly moderating its impact, suggesting it primarily influences customers already within a specific store.
- Smaller Stores Benefit More: The interaction effect of store ACV (Average Category Volume) with sampling suggests that smaller stores with less diverse assortments tend to benefit more from sampling than larger stores.
Key Results
- Immediate Sales Lift: In-store sampling consistently shows a positive and significant immediate impact on sales across all data sets.
- Carryover Effect: The effect of sampling dissipates over time, lasting anywhere from 2 to 8 weeks for single events and significantly longer (around 45 weeks) for multiple events.
- Sampling vs. Display: Sampling events provide a much larger immediate sales lift (close to 2x) compared to in-store displays. The total incremental lift from sampling is also substantially larger than from displays.
- Multiple Events vs. Single: For multiple sampling events, the decay curve is sustained for much longer, indicating a more enduring impact.
- Category Expansion: The retailer data shows that sampling by one brand can lead to a positive spillover effect on competitor brands within the category, suggesting a category-wide sales increase.
- Optimal Store Selection: The study demonstrates that by using their model, manufacturers can identify optimal store locations for sampling to maximize incremental profit and determine cost thresholds beyond which sampling becomes unprofitable. They can also identify the minimum number of stores needed to achieve a target incremental profit.
Limitations and Future Work
- Generalizability of Multiple Events: While the study provides insights into multiple sampling events using the retailer data, the authors acknowledge that more data would be needed to generalize these findings broadly.
- Experiential Events Beyond Tasting: The current study focuses on tasting/sampling. Future research could explore how this translates to other types of experiential events, such as product demonstrations for durable goods.
- Endogeneity of Baseline Sales: The current model addresses endogenous store selection based on expected sales lift. Future work could investigate another form of endogeneity related to the selection of stores with higher or lower baseline sales.
- Spatial Modeling: Incorporating more detailed spatial information and competitive landscape data could further enhance store selection models.
- ROI Comparison: The authors suggest a need for further research to contrast the Return on Investment (ROI) of sampling versus display promotions, given cost information.
Summary updated