SALES FORECASTING

Nikita Naik.

 

Griva, Butorina, Et All (2023)

A sales funnel is a marketing model that outlines the journey of a potential customer from initial contact with a product or offer to making a purchase. It comprises various stages where clients interact with a company and its brand. Analyzing the sales funnel is crucial for understanding where bottlenecks occur and where customers drop off. The goal is to identify constraints in the process, correct mistakes, and optimize the use of resources. The sales funnel typically has two main stages: opening and closing a deal. To conduct sales funnel analysis effectively, it’s essential to digitize and automate the tracking of each stage. The most effective sales funnels consist of at least four stages, with customers continually evaluating the importance of the company’s offer to them. The customer journey can be represented as an inverted pyramid, starting with initial interest and narrowing down to actual purchases. Customers move through the funnel by answering questions about the company’s offer, product liking, purchasing ability, and the buying process.

Hwang, Et All. (2023)

The study introduces an integrated model for sales forecasting of products with short lifecycles, such as mobile phones. By analyzing 38 mobile phone sales data in the Korean market between 2020 and 2021, the research identifies key variables affecting sales, including factors related to both the product and sales. The Random Forest model outperforms other machine learning models, highlighting the importance of selecting the right forecasting model to avoid significant errors that can impact the entire supply chain. The study also suggests that companies producing short-term products can optimize their supply chain strategies by implementing the Random Forest model or a similar analysis process. Additionally, the variation in predictive performance by brand underscores the need to consider brand-specific marketing strategies in sales forecasting and gather additional data to enhance accuracy. Future research directions include exploring outlier processing, advanced model implementations, and analyzing the model’s effectiveness in different market conditions and product lifecycles.

Chen, Et All (2022).

In the fashion retail industry, accurate sales forecasting plays a pivotal role in various aspects, including pricing, inventory management, marketing, production, and financial planning. However, the unique challenges of the industry, such as short product life cycles and low inventory strategies, make forecasting particularly complex. This study addresses these challenges by introducing a two-layer (TLs) model for total sales forecasting of new fashion products. The model considers both actual and censored demand in brick-and-mortar stores, incorporating parameters related to conversion rates of inventory to sales and customer preferences. Using real-world data from a multinational fashion retailer, the TLs model outperforms other forecasting methods, providing valuable insights for fashion industry managers. The study also defines two indicators, the average conversion rate, and the marginal conversion rate, to aid in product competitiveness and inventory decision-making. This research contributes to enhancing sales forecasting accuracy and decision support in the fashion retail sector.

Kurniawan,Et All (2023)

This study focused on sales forecasting in Indonesia’s consumer goods companies with active warehouses, employing Artificial Neural Network techniques to process large and dynamic datasets. Traditional methods were used, involving historical sales and remaining stock data, along with multiple warehouse-related variables for forecasting. Qualitative methods were employed to assess data quality. The results demonstrated an impressive accuracy rate of 111% using the Extreme Learning Machine, with a Mean Square Error score of 0.02716 for sales forecasting. This suggests significant potential for sales and profit growth, enabling the company to make informed decisions, expand production lines, and underscores the practical value of the study.

Dai, Et All (2023) 

China’s automobile industry is undergoing significant diversification and facing considerable uncertainty. The evolving landscape of automobile consumption reform poses challenges for both the country, its enterprises, and consumers. To assess the development potential of mainstream power vehicles, various statistical and machine learning models were compared, with the Prophet model and BP neural network model demonstrating high prediction accuracy. A VAR multivariable BP neural network model was developed to consider lag effects on automobile sales comprehensively. Using these models, sales of traditional fuel vehicles, battery electric vehicles, and plug-in hybrid vehicles in China are forecasted for April 2023–December 2025. The univariate forecast suggests that by 2025, these vehicles will account for 43.8%, 44.4%, and 11.8% of the market, respectively. Meanwhile, the multivariate prediction indicates that by 2025, their market shares will be 51.0%, 37.9%, and 11.1%, respectively.

Ramos, Et All (2023)

This study addresses the complex challenge of demand forecasting at the store and SKU level, particularly in the context of large retailers. The research proposes novel approaches to incorporate various drivers affecting SKU demand, including promotions, into statistical models like ARIMA and ES. The findings demonstrate that models with covariates outperform those without across different forecast horizons, with RidgeX being the most accurate. However, for promotional periods, PCA-based models excel. The research highlights the importance of transparent forecasting models in retail, which can automatically account for various effects and provide accurate predictions while supporting promotional planning. The ease of implementation is a significant advantage of these models, making them practical for retail forecasting support systems. Future research may focus on software interface design and further diversification of promotional information.

Dana-Mihaela Petroșanu, Et All(2022)

E-commerce has become a global phenomenon, offering numerous advantages to businesses, including the flexibility of not needing a physical location, 24/7 accessibility, and reduced operating costs. Automated inventory management tools and data-driven customer targeting through email campaigns are key benefits of e-commerce businesses. Accurate sales revenue forecasting is crucial for resource allocation, investment planning, and overall business success. This paper introduces a novel sales forecasting method using a Directed Acyclic Graph Neural Network (DAGNN) for Deep Learning, tailored to provide fine-grained daily sales revenue predictions at the product category level. The study is based on the requirements of an e-commerce store owner in Romania and offers insights into the methodology and results achieved, contributing to the field of e-commerce sales forecasting.

Luo, Et All(2022)

Sales forecasting in the apparel retail sector is a complex task due to the vast number of stock keeping units (SKU) and various influencing factors like style, color, size, and external variables. Traditional methods include time series forecasting and machine learning models like Support Vector Regression and Neural Networks. Recently, decomposition models based on matrix factorization, such as Factorization Machine (FM), have gained popularity for their feature engineering capabilities. Residual correction is also crucial in improving forecast accuracy, with models like Autoregressive Integrated Moving Average (ARIMA) and grey forecasting being used. Deep learning techniques, particularly Long Short-Term Memory (LSTM), have shown promise in handling complex residual sequences. This paper introduces a novel xDeepFM-LSTM forecasting model for apparel sales, combining xDeepFM for initial predictions and LSTM for residual correction. The results demonstrate enhanced forecasting accuracy, outperforming other machine learning and deep learning models, making it a valuable tool for apparel retailers.

Tudor (2022)

The COVID-19 pandemic has had a profound impact on the e-commerce sector, driving accelerated growth and changes in consumer behavior. This study develops an integrated forecasting framework to assess the pandemic’s effects on total e-commerce retail sales and its share in total retail sales. By utilizing statistical and machine-learning forecasting methods, the research finds that the pandemic accelerated the e-commerce sector’s expansion by at least five years, resulting in an additional $227.820 billion in e-commerce sales and a 10.61% increase in its share of total retail sales. The study also suggests that the pandemic will have a long-lasting influence on consumer preferences, with e-commerce sales expected to continue growing through 2025, reaching $378.691 billion and representing 16.72% of the US retail sales sector by December 2025. This research provides valuable insights into the pandemic’s lasting impact on the e-commerce industry and its implications for policy and business strategies.

García Sánchez, Et All (2022)

This research addresses the challenge of demand forecasting in the high-implication purchases sector, focusing on the car market. Automotive original equipment manufacturers (OEMs) are unique in their ability to capture customer demand through their car configurator (CC) webpages, where potential customers can customize and configure their desired cars. The study aims to measure the reliability of CC data for demand forecasting. By comparing real weekly color mix sales with forecasts generated using machine learning algorithms and statistical procedures, the research demonstrates that CC data significantly improves forecasting accuracy, particularly in predicting car model and color mix sales. The correlation between CC data and actual sales occurs in the period from 1 to 6 months before purchase, making color customization a key factor in demand forecasting for the automotive industry. This study offers valuable insights and a methodology that can be extended to other automotive OEMs with similar online configurator services.

Conclusion:

In conclusion, the realm of sales forecasting and demand analysis is a dynamic and multifaceted field that caters to diverse industries and their unique challenges. From the intricacies of analyzing sales funnels and digitizing tracking processes to the adoption of advanced machine learning models like Random Forest, Neural Networks, and ARIMA, it’s evident that businesses are continuously evolving their approaches to achieve more accurate predictions. These studies emphasize the importance of aligning forecasting methods with the specific needs of different sectors, such as the fashion retail industry, automobile sales, and e-commerce, where factors like short product lifecycles, dynamic consumer behaviors, and external variables play pivotal roles. The COVID-19 pandemic’s profound impact on e-commerce and the automotive sector highlights the importance of adaptability and the use of data-driven insights to stay ahead in an ever-changing business landscape. These research findings underscore the significance of embracing novel techniques and transparent forecasting models, ultimately enhancing decision support and driving success in a competitive market environment.

References:

 

Reference:

Griva, Butorina, Sidorov, & Senchenko (2023). Analysis and forecasting of sales funnels. Mathematics, 11(1), 105.

 Reference:

Hwang, Yoon, Baek & Byoung-Ki Jeon. (2023). A sales forecasting model for new-released and short-term product: A case study of mobile phones. Electronics, 12(15), 3256.

Reference:

Chen, Liang, Zhou, & Liu (2022). Sales forecasting for fashion products considering lost sales. Applied Sciences, 12(14), 7081.

Reference:

Kurniawan, Triloka, & Ardhan (2023). Analysis of the artificial neural network approach in the extreme learning machine method for mining sales forecasting development. International Journal of Advanced Computer Science and Applications, 14(1)

Reference:

Dai, Yu, Wang, & Zhao (2023). Prediction of china automobile market evolution based on univariate and multivariate perspectives. Systems, 11(8), 431.

Reference:

Ramos, Oliveira, Kourentzes, & Fildes (2023). Forecasting seasonal sales with many drivers: Shrinkage or dimensionality reduction? Applied System Innovation, 6(1), 3.

Reference:

Dana-Mihaela Petroșanu, Pîrjan, Căruţaşu, Tăbușcă, Daniela-Lenuța Zirra, & Perju-Mitran, (2022). E-commerce sales revenues forecasting by means of dynamically designing, developing and validating a directed acyclic graph (DAG) network for deep learning. Electronics, 11(18), 2940.

Reference:

Luo, Chang & Xu (2022). Research on apparel retail sales forecasting based on xDeepFM-LSTM combined forecasting model. Information, 13(10), 497.

Reference:

Tudor (2022). Integrated framework to assess the extent of the pandemic impact on the size and structure of the E-commerce retail sales sector and forecast retail trade E-commerce. Electronics, 11(19), 3194.

Reference:

García Sánchez, Cardona, & Alexandre Lerma Martín. (2022). Influence of car configurator webpage data from automotive manufacturers on car sales by means of correlation and forecasting. Forecasting, 4(3), 634.