Achieving better forecasting accuracy through error measurement

In Customer Operations, there are many benefits of effective resource planning, including CX & revenue maximisation, cost control and improved employee engagement. Yet resource planning is reliant on accurate forecasts of future customer worktypes such as emails, chats or calls. From these, reliable resource requirements can be determined and robust plans can subsequently be designed.

In this post, I discuss why the measurement of forecast error within an operation plays a crucial step in helping to reduce such error.

📉 Learn from historical errors
So much can be learnt from individual errors. Notable errors should be investigated to determine whether improvements can be made to the forecasting model, or whether there was insight within the business that might have helped accuracy. For example, if regular bias is spotted in historical errors, this can be rectified. Also, if the cause of error was poor communication with Marketing, then this can be addressed.

🔄 Track improvement
It is insightful to understand in the short, medium and longer term whether the error is increasing or decreasing. While enhancements to forecasting methods aim to reduce error, increased variation in customer volumes may cause errors to increase – despite the improvements. In such situations it may be useful to compare the deployed forecasting model to a naive method, to demonstrate the value from the current forecasting model.

🛠️ Select the best model
When selecting which forecasting model should be deployed, the performance of different candidate models can be tested using sensible error measures. The model with the lowest error (on test data not used to build the model) can be selected as the one to be used. This provides assurance that the selected model has performed well.

In addition to improving forecasting accuracy, an understanding of error alongside other metrics, such as AHT, Schedule Match and Schedule Adherence, can help diagnose the root causes of service level failure or costly overstaffing. This empowers operational leaders to drive the best actions to improve future performance.

Lastly, be aware that labelling something as a ‘forecast error’ shouldn’t automatically imply a blameworthy mistake. Instead, it signifies a variance between predicted outcomes and actual results – a natural occurrence in complex business operations. Viewing forecast errors through a lens of improvement rather than blame fosters a culture of learning and innovation, enabling teams to refine forecasting techniques and drive continuous improvement without fear of repercussion.