Teppei Syokudo - Improving Store Performance: ESK Findings

Home

Product Portfolio Analysis

Improving Store Performance

Project Management

Documentation

The Team

Introduction

Data Analysis Methodology

Hypotheses & Findings

References

Hypothesis 1

Hypothesis 1: We can increase store productivity by hiring good cashiers who can upsell (increase sales dollar per customer) and serve customers faster (increase customer number).

From the sales process, we know that the cashiers are the most customer facing staff. They are also the most likely to influence customers’ purchase decisions, based on their ability to upsell and cross-sell. This hypothesis looks at identifying good cashiers who are able to consistently increase sales dollars per customer through upselling and/or cross-selling. It also looks at the speed at which the cashier serves the customers, as more customers served within an hour means higher sales, which directly relates to higher store productivity.

For this analysis, we use simple linear regression and apply it to each cashier to see if a particular cashier is able to affect the number of customers, and sales dollar per customer. At the same time, we account for the time of day and day of week effects.

Fit Model. To do so, we use JMP’s Fit Model function. We first go to the Menu bar and click on Analyze, then on Fit Model.

A pop up should appear, prompting for a JMP Data Table. Select the data table for running the regression model. The Fit Model Dialog will pop up.

Next, select the Role Variables. Y refers to the dependent variable(s) that we want to analyse. For Hypothesis 1, Y would be Sales/Customer and CustNo. By performs a separate analysis for each level of Y. This means that we can use By to account for the time of day and day of week effects. We add Day and Peak/Non-peak to the By option. Construct Model Effects is our X variables, the independent variables. We add a single cashier (eg. Staff 31) to this box, so as to identify the effect that this single cashier has on our dependent variables. Finally, we click on Run to run the analysis.

The resulting Fit Model Dialog before running the analysis should be like this:

Analysis Results.

The results show that on Saturdays during Lunch Peak, Staff 31 does not make a significant impact on sales/customer as it has t-value greater than 0.05. However, Staff 31 does make a significant impact on CustNo with F-value and t-value of 0.0233. The results show that when Staff 31 works on Saturdays during Lunch Peak hours, the number of customers will increase by 5.96.

Implications. Teppei Syokudo should assign Staff 31 to work on Saturdays during Lunch Peak hours so as to increase number of customers for that period.

Hypothesis 2

Hypothesis 2: The presence of managers can positively impact store productivity by increasing the staff’s ability to upsell and serve more customers.

We have identified three shop managers present in Teppei Syokudo and we want to test whether the shop would perform better when these managers are present as the staff are more motivated to upsell and serve customers faster.

Fit Model. We use the Fit Model with the dependent variables Y as Sales/Customer and CustNo, as well as independent variables Manager 1, Manager 2 and Manager 3. Manager-variables have a value from 0 to 1, indicating their presence for every hour. Since this hypothesis involves a multi-linear regression with three independent variables, we included two-way interactions between the managers to account for the scenarios where there are two managers in the shop. As there is no scenario where there are three managers in the shop, we excluded three-way interactions. The resulting Fit Model Dialog should look like this:

We ran the Fit Model with the independent variables so as to identify the effect that each manager has on the dependent variables.

Analysis Results.

The results show that on Mondays during Lunch Peak hours, all three managers do not have significant impacts on average sales per customer (Response = Sales/Customer No) and customer number (Response = CustNo), as the Prob>|t| values are greater than 0.05.

Because there are multiple independent variables that might be correlated with each other, a multi-collinearity test should be done. The VIF function in JMP tests for multi-collinearity in a model. VIF values above 10 signal high multi-collinearity.

Right click on the Parameter Estimates of the Fit Model and go to Columns and then Click on VIF.

A VIF column should appear beside the Prob>|t| column. It seems that there is low multi-collinearity for this particular model because the VIF values are small.

Implications. Managers have no significant impact on the staff’s ability to upsell or to serve more customers on Monday lunch peaks. Teppei Syokudo can explore the feasibility of having managers to check into the store periodically since their continued presence do not motivate the staff to work harder.

Hypothesis 3

Hypothesis 3: We can increase store productivity by increasing the number of full-time staff present, as they are better at upselling and they serve customers faster as compared to part-time staff. Fit Model. We use the Fit Model with the Sales/Customer and CustNo as dependent variables Y. We used the Total number of Full-time Labour Hours and Total number of Part-time Labour Hours as independent variables X. Since this hypothesis involves a multi-linear regression with two independent variables, we included two-way interactions between both independent variables to account for the scenarios where there are both full time staff and part-time staff working in the shop. The resulting Fit Model Dialog should look like this:

We ran the Fit Model with the independent variables so as to identify the effect that the number of full-time labour hours, as well as the number of part-time labour hours have on shop performance.

Analysis Results.

The results show that on Mondays during Lunch Peak hours, seven more customers usually accompany an hourly increase in the total number of full-time labour hours whereas six more customers usually accompany an hourly increase in the total number of part-time labour hours. The results are significant as the Prob>|t| values are smaller than 0.05.

In the same period of time, both the number of full-time and part-time labour hours do not have significant impacts on average sales per customer (Response = Sales/Customer No), as the Prob>|t| values are larger than 0.05.

It seems that there is low multi-collinearity for this particular model because the VIF values are small too.

Implications. Teppei Syokudo should take further steps to design a controlled experiment with the number of full-time labour and part-time labour as independent variables, to validate the results of this model. If it is proven that increasing the number of full-time labour allows the shop to serve more customers relative to increasing the number of part-time labour, then Teppei Syokudo can evaluate whether the extra revenue generated justifies the additional cost needed to hire more full-time labour.

Hypothesis 4

Hypothesis 4: We can increase store productivity by staffing full-timers or managers as cashiers as they are better at upselling or serving more customers than part-timers.

From the sales process, we know that the cashiers are the most customer facing staff. They are also the most likely to influence customers’ purchase decisions, based on their ability to upsell and cross-sell. This hypothesis looks at identifying good cashiers who are able to consistently increase sales dollars per customer through upselling and/or cross-selling. It also looks at the speed at which the cashier serves the customers, as more customers served within an hour means higher sales, which directly relates to higher labour productivity.

Fit Model. We use the Fit Model with the dependent variables Y as Sales/Customer and CustNo, and independent variables Manager, Full-time and Part-time. We run the Fit Model with each individual independent variable so as to identify the effect that each type of staff has on the dependent variables. The resulting Fit Model Dialog should look like this:

Analysis Results.

The results show that on Wednesdays during Dinner Peak hours, Full-time staff make a significant impact on Sales/Customer as its F-value and t-value are greater than 0.05. It tells us that on Wednesdays during Dinner Peak hours, having Full-time staff as cashiers will increase the average sales dollar per customer by $2.38. However, Full-time staff do not make a significant impact on CustNo, with F-value and t-value greater than 0.05.

Implications. Teppei Syokudo should assign a Full-time staff as cashier on Wednesdays during Dinner Peak hours so as to increase the average sales dollars per customer.

Hypothesis 5

Hypothesis 5: We can increase store productivity by decreasing the number of staff hours on time periods where there is excess capacity.

If the hypothesis is true and there were time periods where there is excess capacity, we would be able to identify time periods where the customers served, a proxy of how busy the shop is, per labour hour is significantly lower than other time periods. If the hypothesis is false, the shop should be at its optimal level of customers served per labour hour and there should be no significant difference between time periods.

As the independent X variable in this case is Day, which is a nominal variable, and dependent variable Y is the Number of customers served per labour hour (CustNo/Total number of Labour Hours), which is a continuous variable, we will be using the one-way ANOVA analysis to test whether there is a significant difference among the average CustNo/Total number of Labour Hours for each day of the week. We will repeat the analysis for the three time periods in a day to account for the time of the day effect.

Fit Y by X. To do so, we used JMP’s Fit Y by X function. We first go to the Menu bar and click on Analyze, then on Fit Y by X.

We then set CustNo/Total number of Labour Hours as the dependent variable Y and Day as the independent variable X. The resulting Fit Y by X Dialog should look like this:

Analysis Results.

The results show the different CustNo/Total number of Labour Hours for each Day and the line in the middle of the data points equals the mean of all the data points. To conduct the ANOVA analysis, click on the red triangle and the Means/Anova function.

After clicking on the Means/Anova function, the one-way ANOVA analysis will be conducted. Mean diamonds representing confidence intervals will appear, with the line near the center of the diamond representing each group’s mean and the vertical span of the diamond representing the 95% confidence interval for the mean of each group.

The results show that during Lunch Peak hours, the average CustNo/Total number of Labour Hours significantly differs across the days. Saturdays, Sundays and Public Holidays have significantly lower average CustNo/Total number of Labour Hours than the rest of the week as Prob > F value is less than 0.05.

However, the results for the Idle time period shows that the average CustNo/Total number of Labour Hours are significantly higher during the weekends and Public Holidays than the rest of the days as Prob > F value is less than 0.05.

Implications. Teppei Syokudo should explore the feasibility of decreasing the number of labour hours during lunch peak on weekends, as well as during Idle periods on weekdays as there is spare capacity during those periods.

Conclusion and Future Work

Our paper seeks to explore the use of regression analysis to identify opportunities to improve shop productivity. We tested five different hypotheses and our results, as well as the implications as stated below.