Have you ever wondered about the real impact of an accurate (or inaccurate) contact center forecast? A forecast's most important output is the number of FTEs (full-time equivalents) required for a given time period. Every other sub-forecast (contacts, handle time, shrinkage) plays an important contributing role to the headcount forecast.
The forecasting function can be very technical or quite basic, yet neither extreme is helpful in practice. It can be challenging for management to make decisions when forecasters take a purely statistical approach. In contrast, taking a basic approach using simple calculations is typically a symptom of a lack of resources or tools that could help generate better forecasts.
Ultimately, the best approach falls between these two extremes. The base of your forecast should leverage historical data and incorporate the use of statistics. Your outputs should then be modified by applying business intelligence to improve their accuracy, as well as simplified so management can understand the scenarios and conclusions presented.
With that said, here are some common mistakes that can negatively impact a forecast's accuracy if you're not careful.
Using regression analysis to project forward assumes that what has happened in the past will continue into the future. This is valid for some types of forecasting, but not for contact centers.
Historical data will only get you so far in a contact center because the nature of the industry is characterized by many human decisions and fluid behaviors. That's why it's important to proactively get the business intelligence that will provide insight into the future activity impacting your statistical forecast.
Consider a marketing campaign. This would be an active approach by your business to generate more revenue from your customers, and the impact may be an increase in the customer contact rate.
Additional types of information that can help to improve contact center forecast accuracy are:
A workload is determined by multiplying the expected volume by the average handle time (AHT), and AHT is generally calculated by taking the historical AHT and projecting it forward. But this isn't the full picture. Your forecast should also take the expected learning curves for new hires into account.
Consider your average handle time: does it have a tight distribution or a wide variation? A tight distribution (small standard deviation) would result in a more reliable number. But what if you have a wide variation? That could mean your AHT is driven by a small number of agents, and if those agents leave or their performance changes, AHT could change significantly.
Take the time to look at the spread of handle time across the population. If you see outliers, I recommend tracking those agents separately to follow how they trend.
Bonus tip: this is a great opportunity to understand from leadership if there are any process changes to be expected. Handle times are heavily impacted by how simple or complex an agent's processes are, and if a step gets added or removed, you should expect to see a change in handle time.
A reality of staff planning is the movement of people across various queues, as well as in and out of the organization. With all the work that goes into determining a required headcount, it’s important to make sure you have a good idea of exactly how many agents will be available.
The human resources (HR) department may be able to provide insights into voluntary attrition (agents leaving on their own) in your organization, such as employee satisfaction data and/or risks from local job market competition.
Similarly, the operations department should be able to provide data on involuntary attrition rates and trends (agents terminated by the organization), in order to better understand actual availability. More aggressive performance management can result in higher termination rates, for example.
Any successful forecaster in the job for a significant amount of time has been asked to forecast to a budgeted number instead of more relevant data. This is often the result of business leaders aiming to compare team performance against a target. When this happens, however, it's easy to lose sight of what's really needed to achieve service level.
Consider a scenario where you're required to reflect a nine-minute handle time target in your forecast, yet you know that the actual trend is 11 minutes. If you use nine minutes, your forecast will show that you need fewer agents than you know to be true.
If you’re not in a position to change the methodology of having budget in the forecast, then you should show both the budget and the trend at the same time. This is to ensure you always have a clear view of what's needed for service level and that your forecast isn't missing any critical factors.
Occupancy is often considered an output of the forecasting process, however, it's also an input to the process when you're planning staff. For any staff group, there will be a correlation between the occupancy and the service level, but simply using an Erlang formula when doing the monthly staff requirements won’t give you what you need.
Your occupancy input has a direct impact on the staff you need to achieve service level, and this relationship changes as other factors change.
Let's say a group has historically achieved an 85% occupancy when they hit an 80% service level. If the schedule efficiency changes because agents now get two consecutive days off, then your occupancy will go down.
At this point, you may now only get an 82% occupancy at an 80% service level. The drop in occupancy means you now need more staff to achieve your service level.
Other factors that change the relationship between occupancy and service level are:
Contact center forecasting takes a lot of time and effort, so make sure to keep these five mistakes in mind before creating your next one!