An Agent-Based Simulation of In-Store Customer Experience

Peer‐Olaf Siebers, Uwe Aickelin, Helen Celia, Chris Clegg

TL;DR

This paper introduces ManPraSim, an agent-based simulation model designed to understand how human resource management practices influence retail productivity. By modeling individual customer and staff behaviors, the research aims to provide insights into optimizing in-store experiences and organizational capabilities.

What Problem Does This Paper Address?

The retail sector faces a significant productivity gap, and while management practices are known to influence performance, the specific impact of Human Resource (HR) management practices remains under-explored. This is partly due to the inherent unpredictability of human behavior. Traditional Operations Research (OR) methods often focus on static snapshots or post-implementation analysis, making them less effective for understanding dynamic systems where emergent behavior arises from individual interactions. The authors aim to bridge this gap by using agent-based simulation (ABS) to model the complex, dynamic interactions within a retail environment, focusing on how HR practices affect customer satisfaction and ultimately, productivity.

Methods

The core of the paper is the development and refinement of the ManPraSim (Management Practice Simulation) model, built using AnyLogic™ software. The model employs an Agent-Based Simulation (ABS) approach, representing customers, sales staff, and managers as autonomous agents with defined states and behavioral rules.

Key methodological aspects include:

  • Agent Design: Agents are conceptualized using state charts, which visually depict their states and the transitions between them triggered by events. This approach facilitates understanding and validation by domain experts.
  • Customer Evolution Modeling: The model incorporates mechanisms for customer evolution, including:
  • Word of Mouth (WOM): Satisfied customers influence others to visit, and dissatisfied customers advise against it. This is modeled by adjusting the customer pool size based on daily satisfaction levels.
  • Internal Stimulation (in progress): Customer behavior is influenced by memory of past experiences, with higher satisfaction leading to more frequent visits.
  • Performance Measures: Beyond standard metrics, a novel service level index is introduced, incorporating satisfaction weights for different events to capture the nuanced customer experience.
  • Noise Reduction Mode: A specific mode is implemented to reduce stochasticity, allowing for sensitivity analysis by isolating the impact of individual variables.
  • Empirical Data Integration: Frequency and probability distributions, derived from case studies and expert estimates, are used to model state change delays and decision-making processes, creating a population of agents with individual differences.

Datasets

The research is grounded in case studies conducted in four departments across two branches of a leading UK retailer. Data collection involved participant observation, semi-structured interviews, survey questionnaires, and analysis of company data. Specific data points mentioned include:

  • Sales transaction data for realistic footfall modeling.
  • Company data on work team numbers, composition, opening hours, and peak times.
  • Customer satisfaction data (though specific datasets are not detailed, the paper discusses how they are used to inform model parameters).

The authors note that operational case study data are confidential and have only been compiled for internal reports.

Key Insights and Ideas

  • ABS as a Natural Fit for Human-Centric Systems: The paper argues that ABS is particularly well-suited for modeling complex human-centered systems due to its ability to represent agent autonomy and emergent behavior, offering a structural correspondence to real-world systems.
  • The Nuance of Customer Satisfaction Measurement: The distinction between "accumulated historical data" (CSM-AHD) and "experience per visit" (CSM-EPV) satisfaction measures is crucial. CSM-AHD shows more pronounced satisfied/dissatisfied outcomes, while CSM-EPV is more stable, especially with larger customer pools.
  • The Importance of Modeling Customer Evolution: Incorporating factors like WOM is essential for realistic simulation, as it directly impacts customer flow and can create bottlenecks if not managed.
  • Noise Reduction for Sensitivity Analysis: The noise reduction mode is a valuable tool for isolating the impact of specific variables, but it's important to recognize its limitations when analyzing day-to-day variations or rare events.
  • Staffing Flexibility is Key: The initial validation highlighted that oversimplifying staff roles (e.g., assuming a single cashier role) can lead to unrealistic queue lengths and customer dissatisfaction, underscoring the need for more nuanced modeling of staff task allocation.

Key Results

  • Customer Pool Size Impact: Varying customer pool sizes did not significantly affect standard performance measures but did influence CSM-AHD, showing greater convergence to CSM-EPV as the pool size increased.
  • Mode Comparison: Runtime performance measures were largely similar between "normal" and "noise reduction" modes. However, "normal" mode revealed significant day-to-day variations (e.g., in customer traffic and transactions), which were absent in noise reduction mode. Notably, WW departments showed a higher proportion of customers leaving while waiting to pay in normal mode.
  • WOM Implementation: T-tests revealed significant differences in performance measures with varying WOM adoption fractions. In A&TV, increased WOM led to a small increase in customer figures. In WW, a much greater effect size was observed, with increased customer visits and a surprising increase in "happy" customers, suggesting WW can better handle increased demand. Transaction numbers remained relatively stable, indicating potential bottlenecks in service provision.

Limitations and Future Work

  • Abstraction Level of Staff Roles: The authors acknowledge that their current abstraction level for staff roles is too high. They plan to model staff task allocation more granularly to better reflect real-world flexibility and address issues like cashier bottlenecks.
  • WOM Implementation: While the current WOM implementation shows interesting effects, further investigation is needed. Future work will explore a second WOM strategy where the customer population itself expands or shrinks based on WOM influence, rather than just re-using existing customers.
  • Internal Stimuli: The modeling of internal customer stimuli (memory) is still in progress and requires further development.
  • Staff Flexibility: Enhancing staff flexibility (empowerment) to respond to customer demand is a planned future development to address staffing issues.
  • Teamwork and Advanced Agent Skills: Long-term goals include extending the model to simulate teamwork-related management practices and equipping agents with skills like reasoning, negotiation, and cooperation.
  • Validation Challenges: Initial validation experiments showed discrepancies with real-world data, primarily due to the simplified staff role modeling. The authors emphasize that the model is currently used for insights rather than precise numerical predictions.

Summary updated

Abstract

Agent-based modelling and simulation offers a new and exciting way of understanding the world of work. In this paper we describe the development of an agent-based simulation model, designed to help to understand the relationship between human resource management practices
\nand retail productivity. We report on the current development of our simulation model which includes new features concerning the evolution of customers over time. To test some of these features we have conducted a series of experiments dealing with customer pool sizes, standard and noise reduction modes, and the spread of the word of mouth. Our multidisciplinary research team draws upon expertise from work psychologists and computer scientists. Despite the fact we are working within a relatively novel and complex domain, it is clear that intelligent agents offer potential for fostering sustainable organisational capabilities in the future.
BibTeX
@inproceedings{siebers2008agent,
  author = {Peer‐Olaf Siebers and Uwe Aickelin and Helen Celia and Chris Clegg},
  title = {{An Agent-Based Simulation of In-Store Customer Experience}},
  booktitle = {Nottingham ePrints (University of Nottingham)},
  year = {2008},
}
Figure