Forecasting: Adding Value to Web Analytics Part 2

In Part 1 of this series, we discussed the application of forecasting models to estimate missing data points. In this part, we will discuss identifying future trends and anomalies.

Forecasting Application: Identify significant traffic patterns

Almost all trended graphs include peaks and valleys, which capture our attention. These anomalies stand out to us and raise the questions “What happened?” and “What should we do next?” Before diving into the details, pause and determine if the change from the trend is expected, or if it is something of significance. If the change in the trend is unexpected, we can employ statistical modeling to determine if the change is significant, meaning do we need to raise an alarm and do something about it. Time-series forecasting models will not only allow us to view expected outcomes, but the models provide an upper and lower confidence bound. If a data point lies outside one of those bounds, then that is a good data point to start analyzing.


While preparing for the end of year holiday sales, an online retailer wanted to get an estimate of traffic and sales volumes to their various product pages. With the data available in Adobe Analytics, we used a time-series forecasting model (actually two to validate, an ARIMA and a Holt-Winters decomposition) to forecast expected traffic through the holiday season. To everyone’s surprise, one of the product lines was estimated to have a significant drop in traffic. Other product lines had either increasing or decreasing trends, but with limited resources, the team could not devote time to each one. Through the time series model, the product line with the significant decrease was identified and action was taking to improve traffic and conversion into that area. Here is what we did:

  1. Identified relevant traffic sources and variables for modeling.
  2. Identified a good time-series model for each source (part of this series identifies potential forecasting methodologies).
  3. Added the models to Excel dashboards.
  4. Identified the data points below the trend lines and confidence boundaries.
  5. Identified the data points and variables highly correlated with this product line – This can be done through a simple correlation in Excel, but be aware that just because two items are correlated, does not mean that one event caused the other.
  6. Took action on the correlated activities

To help with the action step, we only looked at areas that we could influence such as paid marketing channels. If we saw a relationship with social referring traffic, that’s great, but may be more difficult to act on as opposed to email or paid search.

Forecasting is a great tool to add to your analytics tool belt, and it is easier to apply than you may think. Remember to use a good representation of your population as your sample data for forecasting. Also try to have a minimum of 18 months of data – statistical models tend to work better with more data, and sometimes the math will not work if you have less than 12 months.

As a final thought, remember the context. Outside variables may be influencing your data. The idea is to not let modeling and statistics replace your current thought process and intuition, but supplement it to strengthen your analysis.

For additional information, or to speak to a member of the Predictive Marketing team about other benefits of data mining and applied statistics in Digital Marketing, contact your Adobe representative.