Genesys Cloud - Main

I'd like to start a discussion about Forecasting

Imagine the following scenario:

A customer service queue has been very stable for several months, with consistent call volumes from Monday through Friday. However, over the past four weeks, every Tuesday has experienced a 40% increase in call volume due to marketing campaigns. These spikes don't occur on other weekdays, and they only happen when a campaign is running.

My question is:

When generating a new forecast, would you keep the entire historical dataset so the forecasting model can learn this pattern, or would you exclude or adjust those days because they represent atypical events?

What criteria do you use to determine when an event stops being an outlier and becomes part of the normal pattern that should be included in the forecast?

I'm interested in understanding how you approach this  and what impact you would expect your decision to have on forecast accuracy in the following weeks.

#WEM-Quality,WFM,Gamification,etc

------------------------------
---------------------
Leticia Roque Oliveira
Senior Full Stack Developer
---------------------
------------------------------

Hello, Leticia.

In the specific case you described, I would recommend keeping the historical data from the marketing campaigns in the dataset used for forecasting.
In this scenario, we know that the 40% increase on Tuesdays is directly linked to marketing campaigns and has occurred over the last four weeks. This indicates that it is not a random event, but rather a behavior that may be becoming a recurring pattern.

My decision would depend primarily on a few criteria:

• Does the company intend to continue running these campaigns?
• Is it possible to know in advance when the campaign will take place?

If the answer to these questions is yes, I would treat these Tuesdays as a business pattern and keep the data in the forecast, or use system features such as forecast adjustments.
The expected impact of this decision is improved forecast accuracy. If I ignore a pattern that will continue to occur, the forecast for Tuesdays will likely be underestimated, resulting in understaffing and a drop in service levels. Conversely, if I incorporate an event that will not recur, the forecast will be overestimated, leading to excess capacity and unnecessary costs.

That is why I believe a forecast should combine statistical analysis with business insight. Historical data is important, but understanding the operational context is what allows us to decide whether an event should be treated as an exception or as a new pattern.

------------------------------
Lilian Masselli
-
------------------------------