營收之預測機制研究-以A公司為例

Abstract

在這薄利時代，產業競爭激烈的環境趨勢下，如何能保有企業優勢，再創事業的另一個高峰，使得公司不在這股銷售寒流中被三振出局，是許多公司關注的議題。預測目前在各產業已經被大量的運用，顯示出多數高階企業經營者逐漸瞭解營收預測對於企業經營的重要性。準確的營收預測不僅能減少庫存成本、增加獲利並有利於提升客戶的滿意度。 根據分析歷史資料結果顯示，每年產品線的營收曲線似乎是有規律的、有跡可循的。所以過去的歷史資料應該可以提供我們一些參考資訊，例如：公司的營收目標可能達成嗎？各產品線未來的發展趨勢為何？透過統計方法與人工智慧建立預測模型，藉由電腦系統進行資料分析與判斷來預測營收，避免人為判斷的主觀性，同時可以節省時間、更專注於改善的活動上。 本研究以資料探勘的方法論，應用類神經網路來建立一套預測模型，以預測營收之變化，並與傳統預測方法的績效做比較。利用平均絕對誤差(MAE)、平均絕對誤差百分比(MAPE)及誤差均方根(RMSE)來衡量預測誤差，以衡量預測方法的準確性。實驗結果顯示本研究建構的類神經網路預測模式其預測能力比時間序列等模式為佳。

Nowadays, the company faces challenges to keep the advantage of competition. In the small profit era, how to make better forecasting becomes an important issue for a company to keep the lead-edge of competition and to create more profile. Forecasting now is largely applied in several areas. It reveals that managers of company realize the importance of revenue forecasting in business management. Accurate forecasting can reduce the inventories, increase profit, and also enhance satisfaction of the customers.

According to the analysis of past revenue data, the yearly revenue curve for each product line seems to follow some patterns. Therefore, we can dig out useful data from the past data and use the extracted information to answer the questions, such as "Is it possible to reach the target of the revenue?", or "How is the potential for each product line?”

Utilizing the statistical method and the artificial intelligence, we can build up the forecasting models. With the help from current IT technology, those models can predict the revenue based on the past data. The proposed model can pre event human judgment failure, save the time for the managers, and help them more focus on process improvement activities.

The research uses the methodologies of data mining and the artificial neural network to create forecasting model, which is used to predict the curve of future revenue. We use the performance measures of MAE, MAPE and RMSE to evaluate the forecasting accuracy. Comparing other traditional and benchmark forecasting models, our proposed model gets better forecasting accuracy.