Many times as managers, marketers, and advertisers are asked to forecast future behavior. Without formal education in time series analysis, this can be a pretty daunting and scary task. However, with the proper tools and education, you can create a model that fits the data and creates a good forecast of your future website traffic.
Your website data has many fluctuations based on campaigns. Campaigns can be timely, constant, or pulsating, making traffic to your website change drastically over time. These variances in traffic data make it difficult for the model to predict future forecasts.
The right software package will allow you to choose the model that has the best fit for the data. Knowledge of Moving Average Models, Autoregressive Response Models, ARIMA Models and other smoothing models will help you decide on the right model fit for your data. The software package I am using also allows you to rank your models based on AIC (model quality measure), or MAPE (Maximum Absolute Percentage Error). The ability to have the software rank your model by the conditions allows you to easily choose between several models and pick the one with the best fit. Expanding the image to the left you will see that several models were used and ranked by the AIC information criterion.
In data modeling, we use a process called cross-validation to double check that the model we chose has a good fit for the data. In my example, I specifically left out the last 5 data points of weekly web traffic to see how close the model came to actual outcomes. The graph shows the plot of actual and predicted data points with lines of the range of how close the prediction came to actual. If you expand the image, you can see the orange highlighted area where the model was almost exact in predicting future outcomes for the last 5 data points.
Over time, and with a significant sample of data, you can create solid fitting time series models. These models can be helpful in allocating budget, and projecting future traffic to your website. The right software package, with the tools to evaluate different models, is very helpful in getting the right fit for your model. It is almost impossible to get the exact projection, however you can get very close with knowledge of modeling software.