Improving trade promotion forecast accuracy at circle k vietnam

MINISTRY OF EDUCATION AND TRAINING HO CHI MINH CITY UNIVERSITY OF TECHNOLOGY AND EDUCATION FACULTY FOR HIGH QUALITY TRAINING LOGISTICS & SUPPLY CHAIN MANAGEMENT IMPROVING TRADE PROMOTION FORECAST ACCURACY AT CIRCLE K VIETNAM SUPERVISOR: MBA HO TH HONG XUYEN STUDENT: NGUYEN NHU QUYNH SKL 0 8 Ho Chi Minh City, April, 2021 MINISTRY OF EDUCATION AND TRAINING HCMC UNIVERSITY OF TECHNOLOGY AND EDUCATION FACULTY OF ECONOMICS BACHELOR THESIS TITLE: IMPROVING TRADE PROMOTION FORECAST ACCURACY AT CIRCLE K VIETNAM Student: Nguyễn Như Quỳnh Student ID: 18132055 Class: 181320B - K2018 Major: Logistics & Supply chain management Lecturer: Dr Hồ Thị Hồng Xuyên Ho Chi Minh City, April 2022 NHẬN XÉT CỦA GIẢNG VIÊN HƯỚNG DẪN Tp HCM, ngày tháng năm Giảng viên hướng dẫn i NHẬN XÉT CỦA GIẢNG VIÊN PHẢN BIỆN Tp HCM, ngày tháng năm Giảng viên phản biện ii ACKNOWLEDGEMENT I am grateful to have an opportunity to work at Circle K Vietnam where I learned and accumulated some lessons for myself, knowledge and invaluable experience from my mentor, Mr Duong Hoang Phuc, Planning & Replenishment Supervisor With his dedicated guidance, support and empathy, I have currently touched the piece of experience in supply planning in Retail I also would like to express my great gratitude to Ms Ho Thi Hong Xuyen, my lecturer in the final project before closing my university time and other lecturers who were with me during academic journey Although I tried my best with hope to make a valuable research, there are still some limitations in terms of knowledge, ability to apply theory to practice and practical experience, so mistakes cannot be avoided I hope to receive feedback from lecturers to make my report better, which can become helpful reference for the next student generation Thank you sincerely! Ho Chi Minh City, March 2022 Student Nguyễn Như Quỳnh iii GLOSSARY TERM STAND FOR CIC Category-in-charge CRM Customer relationship management Ctn Carton DC Distribution center FA Forecast accuracy KPI Key performance indicator MAPE Mean absolute percentage error Pcs Pieces PO Purchase order ROP Reorder point SKU Stock keeping unit SO Store order SOH Stock on hand SOQ Suggested order quantity WIP Work in progress iv LIST OF TABLES Table 2.1 Tradeoffs in retail industry……………………………………………….19 Table 2.2 Class in ABC analysis management…………………………………… 22 Table 2.3 Category features as combining ABC & XYZ analysis…………………24 Table 2.4 Data set for example…………………………………………………… 28 Table 2.5 Summary output………………………………………………………… 28 Table 2.6 Anova…………………………………………………………………… 29 Table 2.7 Residual output……………………………………………………………31 Table 3.1 Item list for instance………………………………………………………36 Table 3.2 Strongbow Promotion (Cider Dark & Cider Gold)……………………….38 Table 3.3 SKU Classification for Wine & Liquor……………………………………45 Table 3.4 Item list of CUNG DINH noodles…………………………………………46 Table 3.5 Replenishment analysis for CUNG DINH noodles……………………… 46 Table 4.1 Case statistics…………………………………………………………… 49 Table 4.2 Actual Demand and Forecast in ABC class with multiple categories…….50 Table 4.3 MAPE data set for each class…………………………………………… 51 Table 4.4 Difference between customer demand and SO for each classification……53 Table 4.5 List of variables’ effects sizes…………………………………………… 55 Table 4.6 Data set for multiple linear regression…………………………………….56 Table 4.7 Summary output………………………………………………………… 57 v Table 4.8 Multiple Regression Analysis…………………………………………… 57 LIST OF FIGURES Figure 1.1: CKVN’s digital loyalty programme and proposition …7 Figure 1.2 Asian convenience store market growth forecast by country (2017-2022) Figure 1.3 Number of outlets for convenience store brand in Vietnam (2021)……….9 Figure 1.4 Number of outlets, floor space and sales growth for convenience stores 10 Figure 1.5 Board of CKVN management ……………………………………………11 Figure 1.6 Merchandising management structure……………………………………11 Figure 2.1 Some FMCG departments and categories ……………………………… 21 Figure 2.2 Scatter Plot for Multiple Linear Regression…………………………… 30 Figure 2.3 Forecast chart result………………………………………………………31 Figure 3.1 The flow of supply chain at CKVN………………………………………35 Figure 3.2 Sales by month (2019-2021)…………………………………………… 36 Figure 3.3 STRONGBOW sales (10/2021 – 2/2022)……………………………… 38 Figure 3.4 Variability of store order and actual demand for STRONGBOW Cider Dark (2021-2022)…………………………………………………………………… 39 Figure 4.1 Comparison of MAPE value over ABC classification……………………52 Figure 4.2 MAPE values SO for each classification…………………………………53 Figure 4.3 Forecast after model applied…………………………………………… 58 vi CONTENTS GLOSSARY iv LIST OF TABLES v LIST OF FIGURES vi CONTENTS vii PREFACE Background of the study Research methodology CHAPTER 1: AN INTRODUCTION TO CIRCLE K STORES INC 1.1 About Circle K Stores Inc 1.2 Circle K Vietnam 1.3 Market share 1.4 Management structure at CKVN 10 CHAPTER 2: LITERATURE REVIEW 13 2.1 The concept of demand planning 13 2.2 The characteristics of demand planning 14 2.3 Forecasting principles and process 16 2.4 Retail industry 18 2.4.1 Definition and characteristics of retail 18 2.4.2 Retail Operations 19 2.5 Apply multiple linear regression to demand forecast in retail 25 CHAPTER 3: REPLENISHMENT PLANNING ACTIVITIES AT CKVN 32 3.1 General replenishment planning at CKVN 32 3.2 Demand forecasting at CKVN 33 3.3 Lead time forecasting 40 3.4 Order cycle and service level goal analysis 42 3.5 Replenishment 44 CHAPTER 4: ANALYZE AND IMPROVE PROMOTIONAL FORECASTING ACCURACY 49 4.1 Reason for low promotion forecast accuracy 49 4.2 Analysis of forecast accuracy based on ABC classification 50 4.3 Difference between SO and actual demand 53 4.4 Multiple Regression for promotional dummy variables 55 vii 4.4.1 Independent variables 55 4.4.2 Implement Multiple Linear Regression with variables 56 4.5 Existing problems and future suggestions 60 CONCLUSION 62 REFERENCE 63 viii 40.00% 30.00% 20.00% 10.00% 0.00% A B C MAPE (Actual) All MAPE (CK) Figure 4.1 Comparison of MAPE value over ABC classification The figure 4.1 shows the effect of deviation of the promotional situation on the performance Class C has the worst performance, class B in the middle and class A gets the best Although class A is the best one, it is expected MAPE values of class A can decrease because A SKU’s is account for the higher sales, lower MAPE means better forecast accuracy In addition, C class contains more variance than A or B class, normally, this can be caused by the fact that C class are more often severely promoted and forecasting model can not be suitable in such heavy promotions The other problem is also remarkable that MAPE (actual) value is higher than MAPE (CKVN) value for A class The reason might derive from calculation of MAPE (actual) overforecast and under-forecast for MAPE (CKVN) This means merchandise in A class tends to be over-forecasted while B, C class tend to be under-forecast 4.3 Difference between SO and actual demand To get the insight in the difference between SO and actual demand, I will calculate the difference percentage between both SO connected to the promotion are delivered daily to stores managed by Operation team First, promotion items are prepared to have stock at stores about a week in advance of promotion to make sure for display on the shelf SO is usually requested following the demand and stock at every store, however, the phenomenon of quantity inflation happens as high stock at store with expectation to sell more Second, DC stock levels and store stock levels have a certain influence on SO When there is a lot of stock available in the stores and/or DC of CKVN, they will order fewer quantity for an upcoming promotion 53 When the promotion intensity is high, stock levels can be high as a result of earlier promotions (1 week before promotion) In case supplier disruption, DC with low stock and have the sign of stockout in a short time, stores will compete to request SO with a higher demand for this time to get the sales at their store Third, the promotional displays are a lot larger than the normal displays and need to be full to the end of the promotion period Therefore, more stock are needed on the shelf than normal and this stock needs to be ordered extra above on the expected consumer demand It can be called that bullwhip with retailer, though this variance is much lower than other parts in the supply chain, it also affects negatively to stock level at both stores and DC SO are mostly larger than the consumer demand and difference between the stores is quite large To present these, I calculated the absolute difference to compare among each classification Absolute difference = | SO − actual sales | actual sales The result is shown in the table 4.4 Table 4.4 Difference between customer demand and SO for each classification MAPE (Actual) 47.85% 52.50% 64.43% Class A B C MAPE (CK) 19.95% 26.82% 29.32% 70.00% 60.00% 50.00% 40.00% 30.00% 20.00% 10.00% 0.00% A B MAPE (Actual) C MAPE (CK) Figure 4.2 MAPE values SO for each classification 54 A class clearly performs better than B and C class B class also performs better than C class A law of large quantity can be explained for this The sales of A class are larger than that of B and C class, so variance in SO is less for A class However, it also recommends to the condition that SO was placed for promotional orders which deviated from the forecast when the sales volume was lower expected The chart shows unsatisfactory result when C class’s performance declines substantially when looking at the MAPE (actual) values Therefore, directly forecasting SO is concluded not to be an appropriate approach, it should adapt the consumer demand in another way, not based on SO shortly trend and it should connect with Operation to coordinate supply chain 4.4 Multiple Regression for promotional dummy variables 4.4.1 Independent variables Data set recommended is recorded in December 2021, when high demand from customers for holiday, especially in Christmas and New Year Eve Many categories perform the high sales, it is also the time to see the most fluctuation of demand Promotion in this period also means pushing stock to consumer after the year Following historical data is a basis to forecast for the next period, however, I also consider peak season for demand to make sure enough stock to satisfy customer’s demand There are many elements impact significantly to forecasting accuracy of promotional items High or low sales depend on promotional proposal contracted with suppliers I call these are independent variables and the performance of data set extracted reflected its effects to purchasing consumer behavior The deeper discount, the higher demand for those items It is noted that the sales after this promotion are possible to decrease when we stimulate the demand at the specific time Table 4.5 reflects the impact of variables to the forecasting accuracy The number of plus signs indicate the expected size of the effect on the promotional sales 55 Table 4.5 List of variables’ effects sizes No Variables Effect Promotional length + All-unit discount + Incremental discount ++ Upselling +++ Holiday category ++ Weekend + Explanation The longer promotional length, the higher promotional sales A common discounted price is applied for all units in the order, higher discount, higher sales Discount price is applied for the number of units purchased above the threshold Discount for one product and discount much more with the two ones plus Discount for product in peak season results in higher promotional sales High demand happens in the weekend 4.4.2 Implement Multiple Linear Regression with variables In this part, I choose three variables (Weekend, Upselling, Holiday) rules over forecast accuracy throughout promotional period for A class which makes the largest revenue for CKVN for research The goal of this part is to find more effective forecasting model to improve forecast accuracy As I mentioned above, method based on SO is not effective due to the limitation of short trend, existing unstable demand The forecast should follow actual demand of customers and variables appeared in promotional proposal Table 4.5 expresses the data set of demand for A class in the promotional period from 7th December 2021 to 28th December 2021 The idea is to use predictors of model to capture trend and or maybe seasonality and the linear part of the regression will quantify the uncertainty in this forecast The model is based on customer demand and corrected to SO The smaller the deviation between the two, the less variation is added by SO process and the better the forecast accuracy for SO will be Directly forecasting 56 SO did not lead to satisfactory results Therefore, the consumer demand is taken as the starting point here and adapted to SO  Weekend: in 8th, 15th, 21st December  Upselling: 16th to 23rd December  Holiday: 24th, 25th (Christmas) Table 4.6 Data set for multiple linear regression (forecast is calculated after multiple linear regression) Date Demand Forecast Period Weekend Upselling Holiday 7-Dec 40,392 37,112 0 8-Dec 42,537 43,166 0 9-Dec 37,972 36,699 0 10-Dec 36,212 36,493 0 11-Dec 36,080 36,287 0 12-Dec 32,890 36,081 0 13-Dec 34,474 35,875 0 14-Dec 36,355 35,669 0 15-Dec 41,404 41,723 0 16-Dec 43,454 41,984 10 17-Dec 42,544 41,778 11 18-Dec 39,818 41,572 12 19-Dec 40,300 41,366 13 20-Dec 39,072 41,160 14 21-Dec 44,972 40,954 15 22-Dec 47,956 47,008 16 1 23-Dec 40,247 40,541 17 24-Dec 53,911 50,841 18 0 25-Dec 47,564 50,634 19 0 26-Dec 31,570 33,196 20 0 27-Dec 33,308 32,990 21 0 28-Dec 34,881 32,784 22 0 The result is shown in the table 4.6 where independent variables (x) are period, weekend, upselling and holiday effect, demand is identified as dependent variable (y) 57 Table 4.7 Summary output Regression Statistics Multiple R 0.935476 R Square 0.875115 Adjusted R Square 0.84573 Standard Error 2162.904 Observations 22 Multiple R is 0.93 indicating a strong linear relationship among variables, which means it is completely possible to use these variables to make a forecast and containing these variables helps to increase forecast accuracy R2 = 0.87 indicates that in 100% of demand uncertainty, we have 87% of uncertainty is due to impacts of period, weekend, upselling and holiday 13% is due to random factors and other factors not mentioned in the model The adjusted R square indicates about 84% of demand can be explained by independent variables, which is not bad Table 4.8 Multiple Regression Analysis ANOVA df Regression Residual Total Intercept Period Weekend Upselling Holiday SS MS F 557283532.3 1.39E+08 29.78118 17 79528585.23 4678152 21 636812117.6 Coefficients 37381.64 -211.21 6152.80 6995.84 17263.17 Std Error 1020.77 83.40 1364.69 1055.50 1835.60 t Stat 36.62 -2.53 4.51 6.63 9.40 P-value 0.0000 0.0215 0.0003 0.0000 0.0000 Following the above output, regression line is: y = 37382 -211x1 + 6152x2 + 6996x3 + 17263x4 From the result in the table 4.7, we can infer some conclusions: 58 Significance F 1.7645E-07 Lower 95% Upper 95% 35228.01 39535.27 -387.17 -35.24 3273.55 9032.05 4768.92 9222.76 13390.39 21135.96  If promotional items not fall into the weekend x2, no upselling proposal x3, not on holiday x4; the next period forecast quantity will decrease 211 pieces  If promotional items happen to the next period x1 without upselling proposal x3, not on holiday x4; the weekend x2 will make forecast quantity increase 6,152 pieces  If promotional items happen to the next period x1 but not on the weekend x2 and it is not on holiday x4; upselling proposal x3 will make forecast quantity increase 6,996 pieces  If promotional items happen to the next period x1 but not on the weekend x2 and it does not have upselling proposal x3; on holiday x4 will make forecast quantity increase 17,263 pieces As recommended in Chapter 2, all p-value after analyzing