Typically this is abbreviated to “Offered calls”. An offered call is counted at the moment the caller is transferred to the queue. So the caller dials the phone, hears some messages, pushes some buttons to indicate who he or she wants to talk to, hears a message announcing that the call may be recorded and possibly hears an estimated wait time in queue. After all this, the call is offered to the queue at which point the call is counted. The count is associated with a particular 15 minute interval like 9:00 to 9:15.
WFM software has never used “calls offered to the queue” for forecasting purposes. Instead, WFM always used the sum of answered plus abandoned calls. Most manual forecasting is practiced in exactly the same manner in accordance with the widely taught principals of interval based planning.
The formula Offered Calls = Answered plus abandoned calls is advocated almost universally by organizations that have a stake in the WFM industry. This includes:
Call Center management books
Call Center Management associations that offer training and consulting
In fact, “offered calls” is very different from “answered plus abandoned”. One is a function of demand. The other is roughly the limit of what the forecast prepared you to answer.
Calls are offered to the queue at the rate they wish to be serviced. Answered Plus Abandoned Calls is roughly the opposite. It’s the rate that callers exit the queue.
Answered Plus Abandoned is the rate you can either service callers or exhaust their patience. It actually has nothing to do with demand.
Calls offered to queue can never be used for forecasting purposes because it will cause the forecast to alternate between outrageously high and outrageously low staffing requirements. It is for this reason that answered plus abandoned calls are so widely used by not only WFM vendors but also by the vast majority of planners who forecast using spreadsheets.
Call center management schools universally teach planners to prepare interval forecasts using answered plus abandoned call counts. The upside is that the forecasts don’t look ridiculous. The downside is that the numbers are completely made up to resemble each interval’s historical answering capacity.
If agents are adhering to the schedule, then counting calls using answered plus abandoned call counts will produce an illusion of high forecast accuracy because calls will only be counted at the rate that the forecast prepared the call centre to answer calls.
An answered call is a call that has arrived some time in the past, waited in queue for any length of time and at the moment the agent answers the call, the telecommunications equipment counts it. Hence the total of calls answered in any 15 minute period says nothing about whether the calls arrived in the same interval. This call count also says nothing about how long calls may have waited prior to being counted.
An abandoned call is a call that arrived at some point, waited in queue but the caller hung up prior to the call being answered. Some callers wait a long time, before abandoning. Since abandons are counted at the moment the caller abandons, these calls are typically counted in an interval that is later than the interval in which the call arrived.
At the end of the day, the total of answered plus abandoned calls are usually equal to the total of calls offered to the queue for the entire day. Small differences can occur in 24-7 call centers because some calls may be offered just before midnight. Hence they are offered one day and answer or abandon the next.
Each interval the total of answered and abandoned calls is virtually never equal to the total of calls offered to queue. By ignoring this fact, WFM software is able to tabulate call counts that are artificially close to the number of calls that they forecasted for each interval. For example, if the forecast is for 300 calls but 500 are actually offered to queue, then the call center will only be able to answer about 300.
The number of callers who abandon in the same interval as they were offered is typically very low. Hence, each interval the total of answered and abandoned calls is never far off from the number of calls that could be answered.
This creates a closed loop in which:
a) The forecast determines the call answering capacity of each interval
b) The call answering capacity of each interval determines the number of calls that will count towards both historical forecast accuracy and agent requirements for future schedules.
Forecast always look accurate in hind sight. Future plans are isolated from real changes in demand.
Many call center ACD’s and switches tabulate offered calls at the time the call ends. They call them offered call counts but the calls are counted at the moment of call termination vs. call offer. This is very confusing to some planners because they believe that they are working with true offered calls but what they are actually working with is the sum of handled and abandoned calls. This is only marginally different from answered plus abandoned calls.
While the vast majority of WFM solution only import answered and abandoned calls, there is one that imports offered calls. However, that WFM solution is bundled with an ACD that pegs its offered calls to end times. Hence what might seem like an exception really is not. WFM vendors never attempt to forecast using true offered call counts.
The average daily abandon rate for most call centers is about 2%. Call centers on the verge of trouble have a 4% abandon rate. Any higher, and the level of agitation in the customer base tends to bring the call center to its knees. Agents face stressful angry conversations. Cancelations are high. Costs increase and revenues drop.
Some callers abandon quickly. This is often the case if they hear a message announcing that the wait times are long. Most WFM vendors ask customers to configure their switched to ignore short abandons. The definition of short can be anywhere from 10 to 30 seconds. This form of undercounting ties “answered Plus abandoned calls” even closer to the call answering rate. This reinforces the illusion of high forecast accuracy while further insulating planners from the realities of underserviced or lost customers.
If a caller hears an undesirable wait time message they may hang up before they can be offered to the queue in which they do not get counted at all (for forecasting purposes).
The most widely written definition of the call counts used by WFM is:
“Answered plus Abandoned”
However, in practice, the call count is more accurately described as follows:
Answered Plus Abandons…
excluding short abandons
excluding abandons that wait into the next interval
excluding calls that wait into the next interval
excluding deflected calls who are never counted at all.
If 4000 calls are forecasted for the day and 4400 are offered, the forecast accuracy is 90% calculated as follows:
=(4400 – 4000) / 4000
Three Forecast Accuracy Scenarios:
a) If hundreds of callers wait for 10 – 20 minutes, the forecast accuracy is still 90%
b) If every caller waits for 20 minutes the forecast accuracy is still 90%
c) If every caller waits 2 hours, the forecast accuracy is still 90%.
While Scenario b and c represent some extremes, they are effective at illustrating the point that daily forecast accuracy is impervious to how long customers wait.
Note that with a service level target of 80% in 120 seconds, the service level target could still be met — no matter how long the last 800 callers waited.
WFM vendors typically promise improvements to Service Levels as a benefit of switching from a manual method to a work force automation solution. If the customer has been performing interval based forecasting manually, prior to purchasing interval based WFM, there is no rational reason to expect improvements since the math behind forecasting method had not changed.
When service levels do improve numerically, this can sometimes be a byproduct of how service levels are tabulated. For example, some work force management systems calculate daily service levels using what is called an average of an average. Another way to express this is that the service levels in each interval are not properly weighted according to the number of calls that arrived in the interval. This is a statistically incorrect method that produces erroneously high daily service level values.
To illustrate this, consider the best and worst interval of your day for service levels. The best answer only 5 calls within target of 120 seconds. This produces a 100% service level for the interval. The poorest performing interval answers only 50 out of 100 calls within the 120 second target. Hence the service level for this interval is 50%. The real service level across both intervals is only 52% (calculated as 55/105). However the unweight average of the interval service levels is 75% (calculated as the average of 100% and 50%). Hence, using averages of averages, many WFM solutions will automatically produce an artificially high service level at the moment they become responsible for reporting service levels. When the same method is used across the 96 best and worst intervals of a day, the effect is the same. Service levels are artificially overstated. The distortion is very strong. Even if real service levels drop, improperly weighted service levels can appear to have risen substantially compared to historical that were tabulated correctly.
Another form of reporting distortion stems from the undercounting of abandoned calls. Most WFM solutions are implemented according to a best practice of ignoring abandoned calls that waited less than a time threshold. The time threshold may be anywhere from 10 to 30 seconds, sometimes higher. Hence, even if the real abandon rates increases, the trimmed abandon rates may appear much improved compared to historical abandon rates that counted all calls.
WFM vendors also endorse a best practice of playing a wait time in queue message to callers. This practice tends to increase the total number of abandons and deflections but decrease the number of callers that abandon after the short abandon threshold. The effect is strongest during any period when the call center is underprepared to answer the flow of calls. Thus, the implementation of the wait time in queue message chases callers out of the system during the periods most likely to produce lower service levels, long wait times and high abandons. The callers who flee in response to the wait time in queue message tend to go unreported because many will disconnect either during the message (deflection) or before the abandon threshold (short abandon).
As part of an initial implementation, WFM vendors often propose something called a forecast calibration. This consists entirely of shrinking the forecast which in turn shrinks the schedule. Hence it’s not really a forecast calibration. Instead it’s a capacity shrink.
The moment you are persistently understaffed across the entire day, you get two “phantom improvements”. The first is High Labor utilization. Being understaffed across the day means very long wait times for many customers but it also means a constant queue. Hence agents never run out of calls to answer.
While the agents are never idle, the high labor utilization is a very poor measure of productivity. Callers who wait too long, tend start their conversation with the agent in a frustrated state. The frustration makes the entire conversation longer and less productive. Talk time increases. First call resolution and/or sales conversions drop. It’s not unusual for long wait times to cause total productivity to drop by 20% or more. Agent stress and absenteeism can drag productivity even lower.
The second phantom improvement is high forecast accuracy. When a call center is persistently understaffed, interval forecast accuracy can be nearly perfect. This is because answered and abandoned calls can only recognize calls at the rate they are answered (give or take a nominal abandon rate). The longer customers wait, the higher the forecast accuracy each interval.
The forecast accuracy can fall apart if the abandon rate gets too high. To combat this, WFM vendors ask call centers to do three things to:
a) Configure their switch statistics to ignore short abandons
b) Set the short abandon threshold as high as possible
c) Insert a wait time in queue message prior to offering callers to the queue.
All three of these measures increase the number of people who “voluntarily” leave the system while counting as few as possible towards offered calls.
Anyone who hangs up when they hear the wait time in queue message has not been offered to the queue yet so they will not be counted at all. Those who hesitate long enough to be offered to the queue and abandon shortly after can also be ignored as short abandons.
The effect is to dynamically “Flush” excess demand. The longer the announced wait times, the higher the rate at which excess demand is flushed. Thus, a persistently understaffed call center can actually score pretty well on wait times, service levels, forecast accuracy and labor utilization. Superficially, one might call this an optimized call center.
Interval based forecasting and the associated interval metrics seem simple and informative on the surface. Unfortunately, they are geared less towards informing planners and more towards subterfuge. Interval planning metrics do nothing to help any call center keep pace with demand — because the metrics are entirely based on the rate that demand leaves the system. These are not demand indicators, they are measurements of the prevailing capacity limits. When a business plans its future capacity in a manner that is directly tied to past capacity it culminates in cascading failures. Angry customers, stressed agents, shrinking revenues, increasing costs, lower productivity and so forth.
Interval based forecasting has never worked.
For this reason SCO has zero reliance on any interval statistic. Instead, everything is grounded in the second by second truth that unfolds in each call center every day.
For decades, WFM software and manual forecasting practices have shared the limitations of interval based forecasts. When the forecast is flawed, everything else that follows will be sub-standard. Bad forecasts mean bad schedules. Adhering to bad schedules not only cements the problems, it creates more stress and less flexibility.
Whatever the weaknesses of the past, any WFM software or manual forecasting practice can be SCO enabled overnight. The results have always been phenomenal and far beyond the expectations of each customer. These are not inflated claims. Its exactly what any call center should expect when they make the leap from a forecasting method that could never tell truth to one that forecasts based on the absolute second by second truths of your historical call records.