Call forecasting is what that serves us a smooth driving of call centre. It gives us clear look into call traffic fluctuations regarding peak hours and off-peak timings; guiding us towards sensible scheduling and staffing. Call forecasting is indeed a hard nut to ground, yet on the flipside it renders our planners to calculate the perfect accuracy rate for upcoming predictions.

A wrong forecast means a whole wrong deployment of (Wait for Me) WFM team and really shamble handling of call traffic. Hence, right forecasting is the cornerstone of quality performance by WFM team. In short, we never overlook estimating our call volume, all contact volume, emails, workload, (Full-time equivalent) FTE or any other measure affecting our smooth traffic run.

For an always-on-go call responding, we keep measuring all sort of traffic in comparison with AHT. It will be an absolute wrong forecast if we just calculate the traffic volume but no analysis of how much time an agent utilizes for handling a single call and overall, during his working hours. Further are the metrics for maintaining a streamlined call centre dashboard:

Although we have got flexibility to adapt this metric according to your business and product demands, still our standard measure for it is a time span of 30 minutes. Measuring this metric on weekly or monthly basis will depict an average of all is well but that will surely not end well. Call centre working is a matter of seconds, so we don’t put it off to measure within just weeks or months.

Measuring an actual volume is a few and far between perfection. Why? Because in general incoming call data collectively comprises all calls i.e. repeat calls, abandoned calls and the calls that were answered or resolved. Who is going to give a wakeup call to the software to part the resolved from unresolved? So here we go for a thoughtful analysis by measuring calls offered, rather than the calls that are answered. An imperfect approach, but less bad!

Overhung refers to the calls that get connected in one interval and linger on till the next one. With an increased call volume the occupancy rate increases and with decreased, vice versa. So we avoid calculating the intervals that are less than double our AHT. This way measuring ends up on too many agents being engaged in calls from the previous interval. An accurate accuracy however enables us to induct the workforce according to the upcoming influx of inbound calls.

It is a very simple metric that is also called percent difference. Our analytics measure it by the difference between the actual volume and the forecast volume. Formula:

Percentage Error = Actual Volume – Forecast Volume/Actual Volume x 100

It’s a dependable calculation that our forecasters use by finding out average volume of different intervals of 30 minutes; at specific times, intraday.

Mean Absolute Percent Error working out simply goes with measuring percentage of average volume i.e. calls offered. MAPE is a useful tool for depicting large numbers in percentage that makes understanding quite easy. For it we just calculate the percentage error of each interval with this formula:

We then calculate the mean average of the percent errors for the data sample of the MAPE.

Using absolute error, we workout Mean Absolute Deviation, that is relatively simple measure. MAD measuring better suits to small businesses. It goes with calculating mean average of the errors (or deviations) for the set data.

Evaluating the pulse of accuracy over a long period really works while analyzing a wealth of data. For it we go for calculating standard deviation of the variation percentages. It is one of the most useful tools used by forecasters as it reveals even mean variations between the ebbs and flows of longer, average variations e.g. weekly. Its formula is:

Standard Deviation (s) = Standard Error * √n

Variance = s raise to power 2

This metric works while comparing the variations from one period to another. By it we measure the call frequency of two periods to find out if their patterns match or not. For instance, comparing call frequency of a specific Monday to that of another. It directs us towards productive measures of arrival figures of workloads and the variations in AHT over the intervals. Such analysis helps our forecasters to know what happened on that Monday is useful or not. Its formula is:

Measuring forecast accuracy and applying its effective outcomes needs certain measures. For an improved accuracy we adapt the following tactics:

Our analytics involve the workforce out of WFM team as they also affect the forecast accuracy. They highlight anomalies in the actual workload by tagging it with reasons (i.e. marketing campaigns, mailings, billing cycles). A collective working of all departments helps smoothen call frequency rate.

Our target is achieving accuracy for the customers who are contacting us, not an increased volume. So we go for an accuracy rate analysis that best suits to the industry or contact type. We make the forecast accuracy as simple as possible by going for a daily analysis and working on it as benchmark.

Once we get to know how many agents we need in a specific time, we come up with a higher or lower requirement scenario. For it we follow a good rule of thumb that is +/- 10% FTE. It helps us instantly add or remove staff to accommodate the difference between our forecast and what really happens.

We never ignore the importance of forecast accuracy on basis of shelf-life products (like fresh food and medicines) in comparison with that of slow-movers (the merchandise with less customer demand within a specific time interval). On a critical demand base forecast, we deploy the agents in accordance with the demand urgency criteria. Forecast Accuracy measurement is a very delicate analysis for the set data may give really good or damn bad scores, depending on the chosen metric and the way calculations are done. For it our analytics take really minute steps that better suit to our even call traffic.