User:Rysc00

Problem definition

The supermarket faces inefficiencies in its process at checkouts, leading to customer dissatisfaction. Currently, there are no easily applicable rules for opening new checkouts, which results in problems with too long queues during peak hours and low customer satisfaction.

The goal of this project is to develop data-based rules for opening new checkouts, ensuring that customer satisfaction remains consistently above 70%. The rule for opening checkouts must be easily applicable during store operations - it has to be based on visible elements.

Method

The simulation of the customer checkout process at the Tesco Vodňany supermarket is based on an agent-based simulation (ABS) approach, implemented in NetLogo. This methodology was chosen for its ability to model individual customer behaviors, dynamic interactions, and emergent system properties, which are essential for addressing the complexity of checkout operations.

Several methods were considered:

Queue theory provides mathematical analysis but struggles with dynamic, heterogeneous behaviors like queue switching and satisfaction changes. It is also outside of scope of the course 4IT496. ABS, on the other hand, models individual customers as agents, allowing for the simulation of dynamic behaviors such as queue selection, switching, and satisfaction adjustments. ABS was chosen as the most suitable approach due to its flexibility and ability to capture customer variability.

ABS in NetLogo was selected for its ability to realistically model complex customer interactions and provide practical insights for maintaining customer satisfaction above 70%. Unlike queue theory, ABS captures the dynamic and decentralized nature of customer behavior, making it the optimal choice for this simulation. Since it is also benefitial for the simulation to visualise the checkout process and state of queues, the ABS in NetLogo was a suitable solution.

Model

Overview of the Model

The model is designed to replicate the customer checkout processes, focusing on interactions between customers and the store environment. It incorporates dynamic customer behaviors, such as queue selection, willingness to switch checkouts, and the impact of these decisions on their satisfaction.

The simulation integrates varying customer arrival rates throughout the day and models the performance differences between regular and self-service checkouts. By analyzing the effects of these factors, the model enables exploration of strategies for improving customer satisfaction in a supermarket setting.

Customers

The model represents customers as individual agents with the following key characteristics:

1. Number of Items in Purchase

• Each customer is assigned a number of items to purchase, following a lognormal distribution with parameters μ = 3 and σ = 0.3.

2. Checkout Type Preference

• Customers have a preference for either regular or self-service checkouts.
• This preference is determined by the number of items in their purchase:
  • Customers with more than 20 items strongly prefer regular checkouts.
  • Customers with 10–20 items have a variable preference based on a lognormal distribution.
  • Customers with fewer than 10 items tend to prefer self-service checkouts.
• Preferences are scaled from 1 (regular checkouts only) to 100 (self-service only).

3. Willingness to Switch Checkouts

• This attribute reflects how likely a customer is to switch queues if it could increase their satisfaction.
• The initial value is drawn from a lognormal distribution scaled to 0–100 with parameters μ = 2.9 and σ = 0.4.
• Each switch reduces the willingness to switch by half, modeling the decreasing likelihood of repeated queue changes.

4. Satisfaction

• Customers begin with an initial satisfaction value based on a normal distribution (mean = 95, standard deviation = 5).
• Satisfaction changes dynamically during the simulation based on:
  • Waiting Time: Satisfaction decreases superlinearly if waiting exceeds 3 minutes.
  • Queue Switching: Switching queues can decrease satisfaction, depending on the customer’s willingness to switch.
  • Checkout Type: Satisfaction changes depending on whether the customer uses their preferred type of checkout.
• Satisfaction cannot fall below zero.

These characteristics ensure a realistic representation of customer behaviors and preferences, enabling the model to capture complex dynamics within the supermarket environment.

Checkouts

The model represents checkouts as individual agents with the following key characteristics:

1. Checkout Type

• Regular Checkouts:
  • Managed individually, with a dedicated queue for each checkout.
  • Faster for larger purchases due to fewer user errors and quicker handling.
• Self-Service Checkouts:
  • Share a single queue for all self-service terminals.
  • Generally slower for individual items due to higher error rates and customer handling times.

2. Queue Length

• Represents the number of customers waiting at each checkout.

3. Status

• Indicates whether the checkout is currently open or closed.

4. Queue Start Position

• Defines the spatial location of the queue for the checkout.
• Self-service checkouts share a common queue, while regular checkouts have individual queues.

5. Currently Served Customer

• Tracks the customer currently being processed at the checkout.

Processes in the Model

Start of simulation

• All 8 self-service checkouts and 1 of 6 regular checkouts are opened.
• The average satisfaction level is set for 0.

Customer Arrival

• Customers are generated based on predefined arrival rates:
  • Peak Hours (8:00–10:00, 16:00–18:00): Arrivals follow a normal distribution with a mean of 4 customers per minute.
  • Off-Peak Hours (10:00–16:00, 18:00–20:00): Arrivals follow a normal distribution with a mean of 2 customers per minute.
• Each customer is initialized with attributes number of items, checkout preference, willingness to switch queues, and initial satisfaction.

Queue Selection

• Upon arrival, customers evaluate all opened checkouts and choose the one expected to maximize their satisfaction.
• Satisfaction evaluation considers:
  • Waiting time: Estimated service time based on the queue length and checkout type.
  • Checkout type: The impact of using a regular or self-service checkout based on the customer's satisfaction.
• Customers are assigned to the chosen checkout queue and positioned at the end of the queue.

Queue Dynamics

• Customers move forward in queues as the person ahead is served and leaves or moves forward in the line.
• In regular checkouts, queues are managed individually, while self-service checkouts share a single queue.

Service at Checkouts

• Checkouts process customers at the front of their queue.
  • Regular Checkouts: Service time depends on the number of items and includes the possibility of errors (0.1% chance).
  • Self-Service Checkouts: Service time is generally slower for individual items and includes a higher error probability (1%).

Queue Switching

• Customers already in queues may decide to switch to another checkout if it offers higher potential satisfaction.
• The likelihood of switching depends on the customer's willingness to switch, which decreases with each switch.
• Switching comes with a penalty to satisfaction based on the effort required.

Customer Departure

• After completing service, customers leave the system, and their satisfaction is updated.
  • Factors affecting satisfaction:
    • Waiting time: Longer waiting times decrease satisfaction logarithmically.
    • Checkout type: Using a non-preferred type reduces satisfaction, while using a preferred type may increase it.
• Departing customers contribute to the overall average satisfaction, which is monitored throughout the simulation.

Checkout Management

• Checkouts are dynamically opened and closed based on customer satisfaction and queue lengths:
  • Opening New Checkouts: If the average satisfaction of the last 20 customers falls below the threshold (70%), a closed checkout is opened (if available).
  • Closing Checkouts: Regular checkouts with no customers are closed if more than one regular checkout remains open. Self-service checkouts remain open at all times.

End of Simulation

• The simulation runs for a total of 780 ticks, representing 13 hours of store operations.
• After the store closes, remaining customers in queues are processed, and all checkouts are closed.

Assumptions and Limitations

Customer Behavior

• Each customer corresponds to a single purchase, abstracting the possibility of group purchases or customers shopping for others.
• Customer satisfaction is influenced only by checkout-related factors, such as waiting time, checkout type, and queue switching. External factors (e.g., overall store experience) are not considered.
• No demographic data (such as age, gender etc.) are included in the simulation. Therefore, the variables such as checkout preference are not influenced by those factors even though they are in reality.

Checkout Types

• Regular checkouts are assumed to be faster for larger purchases due to fewer user errors and quicker handling of items.

Arrival Rates

• Customer arrivals follow predefined normal distributions based on peak and off-peak hours. Variability in arrival patterns due to external factors (e.g., weather or promotions) is not modeled.

Service Times

• Service times at checkouts depend solely on the number of items, the type of checkout, and random errors. Variations in cashier performance or customer familiarity with self-service systems are not accounted for.

Dynamic Checkout Management

• New checkouts are opened if average satisfaction of the last 20 customers falls below a threshold. This assumes that such a metric is sufficient to represent overall satisfaction trends in real-time.

Simplified Environment

• The simulation focuses exclusively on checkout processes, omitting other areas of the store (e.g., shopping aisles or interactions with staff).
• The simulation does not consider staff availability when opening new checkouts.
• The simulation does not consider that self-service checkouts may closed too.

Static Rules for Opening and Closing Checkouts

• The rules for opening and closing checkouts are fixed and may not adapt to unforeseen customer behavior patterns.

Data Sources

• Several parameters such as customer arrivals and satisfaction are based on general assumptions or personal observation and may be biased.
• Information on data sources and how the statistical distributions were derived are available here: https://www.simulace.info/images/Supermarket_checkout_simulation_-_data_sources.docx

Results

The simulation results indicate that dynamically opening new checkouts successfully maintains customer satisfaction above the target level. Specifically, the average customer satisfaction achieved during the simulation was 75.3, with a standard deviation of 0.3. This demonstrates a consistently sufficient level of customer satisfaction.

It was verified that dynamically opening new checkouts based on the average satisfaction of the last 20 customers is an effective approach. However, this method relies on monitoring customer satisfaction in real-time, which is not feasible in a typical supermarket setting, as cashiers and staff cannot directly observe or measure customer satisfaction.

To address this limitation, the simulation results were used to derive an alternative, observable rule:

If there are 16 or more customers at the checkouts for 3 consecutive minutes, a new checkout should be opened.

This observable rule was found to correlate well with maintaining sufficient customer satisfaction levels and is practical for real-world application. It ensures that staff can respond effectively to changes in customer flow without relying on unmeasurable metrics like satisfaction.

The results also highlight the importance of adequate staffing during peak hours. During these periods (8:00–10:00 and 16:00–18:00), the simulation confirmed that one regular checkout and the self-service terminals are insufficient to handle the increased customer flow efficiently. To maintain smooth operations and high satisfaction levels during peak times, additional staff should be available to manage extra checkouts when needed. Conversely, during off-peak hours, the standard configuration of one regular checkout and self-service terminals suffices.

To operationalize the proposed observable rule, leveraging camera systems and AI technologies is recommended. By monitoring the number of customers at checkouts in real-time, the system can:

• Detect situations where the number of customers exceeds 16 for 3 consecutive minutes.
• Automatically trigger notifications for staff to open a new checkout.

Such an AI-driven solution would ensure quick reactions to changing conditions and reduce the reliance on manual monitoring. This technological integration would provide a scalable and efficient method to uphold customer satisfaction consistently.

Conclusion

The simulation results demonstrate how a structured approach can improve checkout management in supermarkets. By analyzing customer behavior and satisfaction, the study showed that dynamically opening new checkouts based on observable metrics—such as a sustained queue of 16 or more customers for 3 consecutive minutes—is an effective way to address operational inefficiencies. While real-time satisfaction monitoring is impractical, this alternative rule corelates with the satisfaction-based decision making and offers a realistic solution.

Beyond reaching the goal of the simulation, the model proved useful for broader operational insights. It can be applied to calculate staffing requirements for maintaining higher customer satisfaction than the current target, test the effects of adding additional checkouts, or simulate scenarios with increased customer flow during peak hours or special events.

Implementing the proposed rule requires careful planning, especially to ensure sufficient staffing during busy periods. Additionally, integrating AI and camera systems for real-time queue monitoring could automate the process, enabling quick and effective responses to changing conditions.

Code

https://www.simulace.info/images/Checkouts.nlogo

Resources

Bachárová, G. (2023, November 20). Pavol Halász, Tesco Stores ČR: Pro zákazníka je důležitá cena. Zboží a Prodej – zprávy z retailu. https://www.zboziaprodej.cz/2023/11/20/pavol-halasz-produktovy-reditel-tesco-stores-cr-pro-zakaznika-je-dulezita-cena-2/

Martin, J., Nenycz-Thiel, M., Dawes, J., Tanusondjaja, A., Cohen, J., McColl, B., & Trinh, G. (2020). Fundamental basket size patterns and their relation to retailer performance. Journal of Retailing and Consumer Services, 54, 102032. https://doi.org/10.1016/j.jretconser.2020.102032

Pallikkara, V., Pinto, P., Iqbal, T. H., & Commerce, B. (2024). Navigating waiting situations at retail checkouts: Associated emotional discomfort and its impact on shopping satisfaction. Management & Marketing, 12(2), 256–274. https://doi.org/10.2478/mmcks-2024-0012

Rajic, T., & Dado, J. (2013). Modelling the relationships among retail atmospherics, service quality, satisfaction and customer behavioural intentions in an emerging economy context. Total Quality Management & Business Excellence, 24(9–10), 1096–1110. https://doi.org/10.1080/14783363.2013.776759

STEM/MARK. (2024, February 13). Češi se k samoobslužným pokladnám staví pozitivně, reálně ale stále vítězí klasické pokladny s obsluhou. STEM/MARK. https://stemmark.cz/cesi-se-k-samoobsluznym-pokladnam-stavi-pozitivne-realne-ale-stale-vitezi-klasicke-pokladny-s-obsluhou/