Modelling and simulating retail management practices: a first approach

Peer-Olaf Siebers, Uwe Aickelin, Helen Celia, Chris W. Clegg

TL;DR

This paper presents a pioneering multi-agent simulation framework designed to quantify how "soft" people management practices—like staff empowerment and training—impact "hard" retail metrics. By modeling individual interactions between customers and staff, the researchers demonstrate that organizational performance is an emergent property of human behavior, offering a powerful "what-if" tool for retail managers.

What Problem Does This Paper Address?

Retailers often struggle to link management practices (how they treat and train staff) directly to productivity and sales. Traditional operations research usually focuses on "hard" logistics like supply chains or simple queuing theory, often neglecting the human element.

The authors address the "productivity gap" in the UK retail sector by investigating the complex, non-linear relationship between shop-floor management and store performance. They argue that because retail is inherently human-centric, we need modeling tools that can account for individual behaviors, varying levels of staff expertise, and the subjective nature of customer satisfaction.

Methods

The researchers employed Agent-Based Modeling and Simulation (ABMS) using the AnyLogic 5.5 platform. Unlike standard Discrete Event Simulation (DES), which focuses on the flow of entities through a process, ABMS treats customers, staff, and managers as autonomous "agents" with their own rules and states.

Key Methodological Features: * State Charts: Agents move through logical states (e.g., Browsing → Seeking Help → Queuing). Transitions are triggered by probabilities or environmental factors (like queue length). * The Service Level Index: A novel, dynamic performance measure. Instead of a static "satisfaction score," each agent accumulates "satisfaction weights" based on their experience. For example, receiving immediate help adds +4 points, while leaving a queue prematurely due to impatience subtracts -4 points. * Multi-Disciplinary Input: The model logic wasn't just theoretical; it was built using qualitative and quantitative data from a top-10 UK retailer, including participant observation and 40 staff interviews.

Datasets

The study is grounded in a deep case study of two distinct retail environments:

  • Audio & Television (A&TV): Characterized by high-priced items, longer service times, and a high requirement for staff expertise.
  • Womenswear (WW): Characterized by higher customer volume, higher conversion rates, and lower requirements for specialized advice.

The authors used internal company documentation and expert estimates to populate frequency distributions (specifically triangular distributions) for task durations and probabilities for decision-making (e.g., the probability a customer seeks help).

Key Insights and Ideas

  • The Staffing Trade-off: The model revealed that simply adding more cashiers isn't always beneficial. In the simulation, increasing cashier numbers eventually reduced the number of sales staff on the floor. In service-heavy departments like A&TV, this "squeezed" the availability of advice, leading to lower overall sales despite shorter checkout lines.
  • The Empowerment Paradox: Empowering cashiers to handle refunds autonomously (rather than referring to a manager) significantly boosted customer satisfaction for those seeking refunds. However, it unexpectedly decreased total transactions because less experienced staff took longer to process these complex tasks, creating new bottlenecks.
  • Learning vs. Doing: "Empowerment to learn" (allowing staff to shadow experts) creates a short-term hit to customer satisfaction because staff are occupied with training. However, the model suggests this is a necessary "investment" for long-term expertise gain.
  • Contextual Divergence: The same management intervention can have opposite effects depending on the department. WW is volume-driven, while A&TV is expertise-driven, meaning a "one size fits all" management strategy is likely to fail.

Key Results

  • Validation: The model successfully replicated the "peak" efficiency phenomenon, where sales increase with staffing up to a point before declining due to floor-coverage imbalances.
  • Empowerment Impact: In the task empowerment experiments, the effect size for customer satisfaction regarding refunds was a massive 1.00, though the number of transactions saw a slight, statistically significant decrease (Effect Size = 0.53).
  • Staff Utilization: Empowering staff to learn led to a significant increase in "normal" staff expertise (Effect Size = 0.99) and more efficient utilization of the total workforce (Effect Size = 0.83).

Limitations and Future Work

The authors acknowledge that this "first approach" has several abstractions:

  • Homogeneous Customers: Currently, all customer agents behave similarly. Future versions will introduce "customer personas" (e.g., the "browser" vs. the "mission shopper").
  • Lack of Memory: Agents currently don't remember their experiences. The authors plan to implement a "finite population" where a bad experience today reduces the probability of that agent returning tomorrow.
  • Simulation Duration: The 10-week runtime may be too short to see the full "break-even" point where the benefits of employee development outweigh the initial cost of training time.
  • Staff Morale: The current model doesn't account for how empowerment affects staff job satisfaction or retention, which are critical components of real-world retail success.

Summary updated

Abstract

Multi-agent systems offer a new and exciting way of understanding the world of work. We apply agent-based modeling and simulation to investigate a set of problems in a retail context. Specifically, we are working to understand the relationship between people management practices on the shop-floor and retail performance. Despite the fact we are working within a relatively novel and complex domain, it is clear that using an agent-based approach offers great potential for improving organizational capabilities in the future. Our multi-disciplinary research team has worked closely with one of the UK's top ten retailers to collect data and build an understanding of shop-floor operations and the key actors in a department (customers, staff, and managers). Based on this case study we have built and tested our first version of a retail branch agent-based simulation model where we have focused on how we can simulate the effects of people management practices on customer satisfaction and sales. In our experiments we have looked at employee development and cashier empowerment as two examples of shop floor management practices. In this paper we describe the underlying conceptual ideas and the features of our simulation model. We present a selection of experiments we have conducted in order to validate our simulation model and to show its potential for answering "what-if" questions in a retail context. We also introduce a novel performance measure which we have created to quantify customers' satisfaction with service, based on their individual shopping experiences.
BibTeX
@article{siebers2009modelling,
  author = {Peer-Olaf Siebers and Uwe Aickelin and Helen Celia and Chris W. Clegg},
  title = {{Modelling and simulating retail management practices: a first approach}},
  journal = {Int. J. Simul. Process. Model.},
  year = {2009},
  url = {https://openreview.net/forum?id=jL1IWyWvyE},
}

Loading related content...

Figure