TL;DR
This paper introduces a novel Hierarchical Bayesian model to disentangle the complex, simultaneous influences of customer and employee satisfaction on firm financial performance. By modeling both supply and demand sides of a firm's operations, the authors reveal how satisfaction levels strategically alter the effectiveness of tactical marketing inputs, offering a more nuanced understanding of how to optimize resource allocation for profitability.
What Problem Does This Paper Address?
Marketing managers constantly face pressure to demonstrate the financial impact of their actions. However, linking marketing efforts to firm performance is challenging due to the multifaceted nature of marketing interventions, which can operate on different timescales and involve numerous intervening factors. A key difficulty lies in accounting for "endogeneity," where managers' decisions about marketing inputs (X) are influenced by their expectations of the outcomes (y), violating standard statistical assumptions and leading to misallocated resources. Existing models often treat these relationships in isolation or as simple linear chains, failing to capture the strategic, moderating role of factors like customer and employee satisfaction on the effectiveness of tactical marketing activities. This paper aims to address this by developing a model that simultaneously considers supply and demand, explicitly models endogeneity, and quantifies the strategic impact of satisfaction on firm technology.
Methods
The core of this paper is a Hierarchical Bayesian model designed to integrate simultaneous supply and demand analysis.
- Demand Model: A multiplicative production function is used to model unit-level revenue as a function of managerially controllable inputs (e.g., FTE, base salary, incentive compensation). This model allows for diminishing returns to scale and interprets coefficients as elasticities.
- Strategic Influence: Latent levels of customer and employee satisfaction are incorporated by constructing a hierarchy on the response coefficients of the demand model. This allows satisfaction to moderate the productivity of tactical inputs, acting as a strategic influence.
- Supply Model: To address endogeneity, a supply-side model is derived from a constrained profit-maximization problem. This explicitly models how managers choose input levels based on their knowledge of the demand function and budget constraints.
- Estimation: A full-information Bayesian approach is employed, utilizing a hybrid sampler (Metropolis-Hastings algorithm) due to the non-conjugate nature of the model. This allows for the estimation of complex, interdependent relationships.
Datasets
The study utilizes data from a national financial services firm, specifically:
- Unit-level income statements for approximately 13 months from 898 retail banking branches.
- Employee satisfaction study data.
- Customer satisfaction study data.
These datasets were collected independently but cover roughly the same time period, allowing for the integration of financial performance and satisfaction measures at the branch level.
Key Insights and Ideas
- Satisfaction as a Strategic Moderator: The most significant insight is that customer and employee satisfaction don't just directly impact revenue; they strategically alter the effectiveness of tactical marketing inputs (like base salary and incentive pay). This means the same marketing dollar can yield different results depending on satisfaction levels.
- Endogeneity is Crucial: The paper strongly supports the premise that managers set marketing inputs with an expectation of their impact on outcomes. Models that ignore this endogeneity (like the exogeneity model M₁) perform poorly.
- Decentralized Decision-Making is Optimal (in this context): Model M5, representing decentralized decision-making with unit-level budget constraints, outperforms other models, suggesting that for this specific bank, balancing resources within branches is more effective than centralized control.
- Employee Satisfaction's Stronger Link: Employee satisfaction shows a more robust and statistically significant relationship with firm technology (i.e., the productivity of inputs) than customer satisfaction.
- Customer Satisfaction's Nuanced Role: While customer satisfaction has a positive posterior mean, it's not statistically significant in the main model. Interestingly, a reduced-form model shows a negative impact, highlighting how ignoring strategic effects can lead to misleading conclusions about customer satisfaction.
Key Results
- Model M5 (Decentralized Supply & Demand) is the Best Fit: This model, which assumes managers make input allocation decisions at the branch level with a single budget constraint, provides the best fit for the data, as indicated by Bayes factors.
- Employee Satisfaction Boosts Revenue Productivity: Employee satisfaction is positively correlated with the multiplicative intercept (β₀), implying more satisfied employees lead to higher baseline revenue potential. It also significantly influences coefficients for base salary and incentive pay.
- Customer Satisfaction's Inverse Link with Incentive Pay: Customer satisfaction is negatively correlated with the incentive pay coefficient (β₃), meaning incentive compensation is less effective in driving revenue at branches with more satisfied customers.
- Reasonable Parameter Estimates: The proposed model (M5) yields parameter estimates that adhere to economic production function requirements (e.g., diminishing returns to scale), unlike simpler models (M₁) which produced counterintuitive results.
Limitations and Future Work
- Competitive Effects Not Modeled: The current model does not account for competitive effects, which could be important in other contexts and would require adjustments to the demand and supply-side specifications.
- Cross-Unit Dependencies Ignored: The assumption of independence across branches might not hold in all retail banking settings (e.g., proximity effects), suggesting a need to incorporate cross-unit dependencies.
- Tactical Actions' Long-Term Impact: The model assumes tactical actions (like base compensation changes) do not have long-term strategic impacts on variables like employee satisfaction. This limits causal interpretation and suggests future work with repeated cross-sectional satisfaction measures.
- Normality Assumption: The model relies on the assumption of multivariate normality for satisfaction distributions, which may be stringent. Exploring models that relax this assumption is suggested.
- Exploring Other Satisfaction Distribution Portions: The current work focuses on the latent mean of satisfaction. Investigating other portions or percentiles of satisfaction distributions could yield different insights.
- Eliciting Managerial Input for Supply Models: Future supply-side models could benefit from directly incorporating managerial insights into decision-making processes, rather than relying solely on presumed optimal behavior.
- Satisfaction and Revenue Variance: Exploring the relationship between satisfaction and the variance of revenue generation, not just the mean, is a promising avenue.
Summary updated