Forecasting is a critical action for any business but particularly production operations such as contact centers. Performing this action is particularly challenging when there is a change in volumes due to seasonality, changes in the business, unexpected events, and automation of the work. As a result of the challenge, people responsible for forecasting often just use best guess and trends. Unfortunately, forecast missed of 5% or more does have a material impact on delivery of outcomes including service levels. This article will focus on contact centers and their service level goals.
When coaching staff on how to forecast both financially and for contact center staffing the first rule I like to share is – in forecasting you are never right, the goal is to be as little wrong as possible. The first step to successfully and consistently forecast is to build a model which considers;
- Historic volumes by month of the year for at least two years, preferably three years
- Historic volumes within each month relative to weekdays, weekends, and holidays. Some holidays fall on different days of the week and will influence volumes before and after the holiday differently depending on the day of week.
- Historic volumes by interval for each day of the week. Each weekday is most likely to have different intraday interval (interval may be measured in 15 minutes, 30 minute, or one hour increments) distributions, so it’s important to know the distribution by day of week. Advise using the previous 3 to 4 days for each day of week and average each interval to obtain a future plan. Using the previous weekdays would exclude a holiday or holiday influenced day, a day with a service impacting event (application crashed, for example), or other anomalous effect for previous days. You’ll want 3 to 4 “clean” days to use for averaging.
- Depending on the type of business being supported it is likely important to know the number of each weekday in a month or at least how many weekend days versus weekdays. If the business only supports service Monday through Friday, it is impactful to the monthly forecast when more weekend days fall in a month than the previous year, etc.
- Average Handle Time (AHT) is the time it takes to complete a contact. For calls AHT usually includes Average Talk Time(ATT), Average Hold Time (AHT), and Average After Call Work(ACW). Outbound calls supported by technology may also need to include Average Preview Time (APT) before the call is connected by the system. However, AHT is determined it’s important to use a similar approach as the contact volumes, such as knowing the AHT throughout a day, by day of week, etc.
- Once the forecasting model is created, include an indicator for days which have unplanned events impacting volumes for that day is it may be easily identified and excluded from future forecasting.
- It isn’t recommended to re-forecast every day for the next day. One particular operation I managed was using a daily approach which was resulting in over a 10% miss every day. Once we moved the process to a weekly forecast with the detail above, the results were plus or minus 3% with a 2% variance often achieved.
One note on a business whose volumes are heavily impacted by regular events. Credit card authorizations used to fund gambling cash advance may need to know ahead of time what sporting events are planned and even rate the level of impact for each event type. The NFL Super Bowl is likely is have a heavier volume impact than a local dog or horse race, for example. The same consideration should be made for marketing campaigns which would be expected to drive increased volumes.
The first effort in building the forecast model should be assumed to have missing elements, incorrect formulas, or incorrect assumptions. Remember, forecasting is never right, just try to be as little wrong as possible. Trial the model for a few cycles (days, weeks, months) tracking actual results and how those resulting volumes drive the assumptions set used to learn where actual results and ratios differ from the forecast assumption set. Do your best to understand why there was a difference and if there is an opportunity to tighten that assumption.
One very important point to share. It’s common for people performing the forecasting to assume that since the results aren’t likely to be accurate, they take shortcuts. Shortcuts harm the accuracy and should be avoided. One example is the assumption for the number of workdays in an average month when the operation is Monday through Friday. The forecaster may look at the calendar and decide the average number of weekdays or workdays in a month is 21. That approach and result is an error. It’s important to use math anywhere possible to do so. The more accurate way to determine average weekdays or workdays in a months is to calculate it – 5 weekdays in a week, 52 weeks in a year, and 12 months in a year results in a formula: (5 x 52) / 12 = 21.6667 or 21.7 rounded. Using the two different approaches can result in as much as a 4% variance in the average contacts per day which means staffing would be off.
Forecasting in a contact center operation is recommended as follows;
- Once per month forecast the months and weeks for the next three (3) months
- Once per month forecast the following month by day and week
- Once per week forecast the following week by day and by interval
- For all forecasts above include key assumptions such as holidays, expected changes in the business, planned automation which deflects manual volumes (you may want to forecast automated volumes for system capacity planning), or any other volume impacting event or information.
Always assume your forecast has a level of inaccuracy and use the actual results to closely compare and question the accuracy of the model and process being used. The idea is to continue tightening the model and process such that accuracy gets better and better. What’s a reasonable goal for accuracy? A good starting goal for accuracy is for actuals to be plus or minus 5% of the forecast. A world class accuracy may be achieved over time at plus or minus 2%.
Always seek information which will aid in improving forecast accuracy and investigate any forecast miss which is over or under your goal to learn why and how the future forecasts may be improved.