Hi buddy,
How you doing today?

Yes, you are right.
I believe there are at least four different ways to figure it out.

1. Qualitative Forecasting Methods

Qualitative forecasting methods attempt to use actual data to determine a qualitative or actual market trend toward a certain position or function in the market. These methods involve looking at non-numerical data. For example, if you were attempting to forecast whether a new product would be successful, you could review customer surveys in which customers have described their views of the ideal product function. Qualitative forecasting methods are not as effective as quantitative methods, which are the most common methods and which come in a variety of different formats. They are often applied when a great deal of data are available on the target market.


2. Quantitative Methods

In general, quantitative methods use numbers -- sales numbers, web traffic numbers, the amount of new accounts or cancellations of existing accounts -- for a specific period depending on the breadth of the forecast being performed. Quantitative methods can be further broken down into explanatory methods and time-series methods.


3. Explanatory Methods

Explanatory forecasting methods use data to attempt to explain trends and to forecast future market direction based on existing data. These consider past performance and marketing trends, such as consumer spending reports and consumer confidence indexes, to attempt to determine the future path for a certain product, website or company. Explanatory methods involve looking at market activity to explain how and why trends occurred, not just to predict what will occur. Because the "how" is important here, this method is different than a time-series method which just considers what the future trends will be.


4. Time-Series Methods

Time-series methods are used only with historical data to predict future performance. For example, if $6 million in sales were made over the last year, a time-series method might predict that $6 million in sales could be achievable this year, with a slight increase allowed for additional business. If a website was hit 340 times on Sunday last week, a time-series method might predict a similar influx on a future Sunday.

Again, the accurate and reliable assumption is the key to get your job done.
Good luck. :)

自己顶一下吧，貌似一般都从时间序列的角度着手啊

您可以在图书馆借阅或者购买黄洁纲老师的《存贮论原理及其应用》，看第一章第七节中有对需求量的预测方法。书中写了有时间序列法（包括滑动平均法，指数滑动平均法，带趋势性的指数滑动平均法，简单季节预测法），指数函数法，回归分析法，专家调查法（也称德尔菲法）。个人觉得需求预测与销售预测差不多吧。当然一线销售人员的感觉、判断也很重要。

thx，貌似很多机器学习的算法也开始应用了。效果也不错。多谢推荐~~

Thanks for your guide. Maybe it is better to adopt the explanatory methods. However the real condition usually limits and hard to get a reasonable casue-efferct realtionship. I found that the machine learning technology is used to do the predicting job now.