Sales Forecasting in Industrial Services: Integrating Market Analysis and Historical Data for Sustainable Business Growth – Case from a Norwegian B2B Service Provider

• Main Author: Zaid, Syed Ali Hussain Mehdi
• Format: Dissertation
• Language: English

The current era is all about challenges due to competitiveness in the business-to business (B2B) domain, as there are numerous complexities in terms of market dynamics, longer sales cycle, and multiple macroeconomic factors which might change the entire sales of a business. Hence, forecasting the potential sales has become a major challenge in today’s world. The sectors where volatility in business is a mandatory aspect to be considered, like oil and gas companies, it is essential to provide accurate sales forecasts so that appropriate strategic planning and decision making, according to the internal and external factors which might affect the sales, can be made. The main objective of this study is to perform a comparative analysis between the traditional and the novel sales forecasting methods, and analyze the accuracy of both methods. In order to achieve the target, two main forecasting models have been deployed in this study which are Long-Short-Term-Memory networks and Random Forest models. The data which has been acquired and used for this study is from the Norway’s oil and gas industry, between the years 2015 to 2023 from a B2B service provider. In addition to this, some macroeconomic variables have also been deployed and considered in this study which are Interest Rates, Employment Rates, GDP, and Oil Prices, with the help of which accuracy of our model applied can be examined. With respect to the methodology, a multistep method has been employed in this study, which involves data collection, preprocessing, normalization of data, implementation of LSTM and RG models, while considering both with and without macroeconomic variables. Further evaluation has been performed using the performance metrics like R-square, Mean Squared square, Mean Squared Error (MSE), Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). Comparison of the models has been performed on the basis of their predictive accuracy. The implementation of the models, presented the findings which showed that LSTM can be considered as the model which provided highest predictive accuracy without macroeconomic variables. It also demonstrated the ability of this model to predict nonlinear patterns and long-term dependencies in the sales data. On the contrary, it has also been analyzed that by including macroeconomic variables, the accuracy and performance of both the models, including LSTM and RF were reduced, while also increasing the complexity of these models. A detailed literary