log in to your account

welcome back! please log in to continue.

Altivon

Avoiding Analysis Paralysis in Monthly Forecasting

April 3, 2014


ForecastChart

In our previous blog on volume forecasting, we discussed the key concepts in understanding your customers’ needs (“Know Thy Workload” – March 20). The next step is to apply those concepts to annual and monthly forecasting.

There is both science and art to forecasting. Mastering both is essential to success.

The science of forecasting

Many volume forecasters rely on models to produce the most exact number possible for their contact center. There are several tried and true models, with a range of reliability:

  • Intuitive—as the name implies, this model relies on the modeler’s understanding of the center’s operations, number of interactions per channel and time estimations per type of interaction. Generally, this model is applied when data is insufficient to take a more analytical approach; it is often used by individuals who are new to volume forecast modeling, or by experts given little time to prepare.
  • Time series—this volume forecast model requires historical data to enable predictions of future outcomes. This model depends on the existence of significant data in uniform time intervals, so analysis can be conducted on a day-over-day, year-over-year basis. It also assumes that there is a correlation between the compared time frames. For example, historic volumes from the same month in recent years (Jan. 2011, Jan. 2012, Jan. 2013) may be utilized to predict the same month in the future (Jan. 2015). However, caution must be exercised in conclusions drawn from time series analysis. The modeler must have a valid basis to compare. In businesses where the day of the month is most critical (e.g., bills are due on the 15th of every month), then call patterns may be strongly influenced by that cause or event, regardless of the day of week.
  • Causal—this volume forecast model requires the modeler to look at an event’s cause-and-effect relationships and predict whether that event will happen again. Holidays are typically recurring events that are predictable. Of course, most holidays are on a fixed date (e.g., Christmas, December 25), and the day of the week will impact customers’ need and desire to call; this is a reason to utilize Causal modeling. In contrast, those holidays that float (e.g., Easter) may be even harder to run a comparison. Other events like inclement weather and catalog drops can be expected, yet their timing is not so certain. For example, hurricanes, tornadoes and blizzards arrive in seasons, which will interrupt Utilities’ field work, however exact arrival patterns remain unknown. Extremes like Superstorm Sandy, major product recalls and errors in marketing campaigns are likely (and hopefully!) not repeatable events; they should be documented and marked as exclusions for future forecasts.
  • Simulation—this model is applied when a rich set of historical data and predictable recurring events are known. Simulation can give the most robust range of forecasted outcomes.

Each modeling technique, and various methods for all, are learned through practice and exercised with discipline. Just like the weatherman, a forecast will be missed, but the accolades are great when the forecast is right.

Always bear this goal in mind: make the volume forecast as accurate as possible. Each center and model is also affected by outside factors, which is why business acumen must add the finishing touches—and, where the art of forecasting is applied.

The art of forecasting

External factors influence all contact centers. It is crucial that the contact center be in regular communication with the rest of the organization to ensure awareness of these forecast-mitigating factors.

In the “Know Thy Workload” blog, we advocated for tighter integration and communication with internal groups like Marketing and IT. Both entities can create disruptions for the center, if caught unaware.

Moving forward

There can be a tendency toward ‘analysis paralysis’ when considering all the models to choose from, the people to talk to, the data to analyze. There certainly is no simple checklist, but it is manageable. Move forward with what you can—Enhance, Elevate and Evolve over time.

Here are some goals to strive for:

  1. Work your contacts. Know what your colleagues throughout the organization are doing and how this may affect you.
  2. Plan for the unpredictable. There are factors that affect your center. Know what they are and when they might occur. Have a plan for dealing with prolonged IVR outages, extreme weather and massive reprints of billing statements should they occur.
  3. Use models to make your forecast more scientifically sound and document the assumptions, communications and other information utilized.
  4. Revise your annual and monthly forecasts when departments share information, when business processes change, or flex your mental muscle and challenge stale assumptions. Remember, the past can provide insight and guidance; the goal is to render an accurate forecast, so update forecasts as often as necessary to achieve this goal.
  5. At the end, know your limits, too. Margin of error should reduce as you learn more and re-forecast closer to the actual period, and in smaller periods (e.g., respectively, Week 50’s margin should decrease as the actual week approaches, and weekly forecasts should improve upon monthly forecasts’ margin of error). The margin of error is also dependent on the center’s volume of calls. A center with 500 calls per week will have more variability and volatility than one that receives 2 million calls per week.

Forecasting can and must be done to give your contact center the chance to deliver great customer experience and stay sane in the process. Be sure to combine science and art:

Science + Art = Volume Forecast

Stay tuned for the next blog, which will get into the details of weekly forecasting.

You might also want to review previous blogs on Workforce Optimization.