What each number is
Every accuracy measure is arithmetic on the forecast and actual contacts you enter; they are historical facts about those intervals. The tracking signal limit is a textbook convention, labelled as a threshold. Workload hours and agents busy are conditional on the AHT you enter and ignore service level and shrinkage.
Formulas
Sum over intervals of |actual − forecast| ÷ sum of actual contacts
The volume-weighted error: every contact missed counts once, above or below. Staffing is planned on this.
1 − WAPE
The headline. It falls below zero only when contacts missed exceed the contacts that arrived.
Mean over intervals with actual contacts of |actual − forecast| ÷ actual
Shown beside WAPE. A 5-contact interval weighs as much as a 100-contact one, so quiet intervals inflate it.
1 − |total actual − total forecast| ÷ total forecast
The day's total. Errors in opposite directions cancel, so it can read high while every interval misses.
(total actual − total forecast) ÷ total forecast
A fact about the day's direction.
Sum of (actual − forecast) ÷ mean absolute error
Whether the errors lean one way. Read against the limit below.
Contacts × AHT ÷ 3,600; per interval, contacts × AHT ÷ (30 × 60)
The workload the miss represents, before service level and shrinkage, which the Staffing Calculator adds.
Rules and bands
Actual ran above forecast more consistently than random error would.
Any bias is not distinguished from random error.
Actual ran below forecast more consistently than random error would.
The tool grades no accuracy figure. It publishes no excellent or acceptable line for WAPE or MAPE, because no published benchmark sets one for your channel and interval length.
Every constant and where it comes from
Outside plus or minus 4 the errors lean one way more than random error would, so the forecast is read as biased. Inside it, bias is not distinguished from noise.
Worked example
Computed by the tool's own engine on two intervals chosen to show why the tool leads with WAPE: they miss by 20 contacts in opposite directions, so the day's total is exact.
9:00: forecast 100, actual 80
9:30: forecast 100, actual 120
AHT: 360 seconds
200 actual against 200 forecast = 100.0%
(20 + 20) ÷ 200 = 20.0%, interval accuracy 80.0%
(20 ÷ 80 + 20 ÷ 120) ÷ 2 = 20.8%
(−20 + 20) ÷ 20 = 0.0, no consistent lean
20 contacts × 360 ÷ 3,600 = 2.0 hours unplanned at 9:30 and 2.0 planned with no contacts at 9:00; 4.0 agents busy for the 9:30 interval
Checked against
Cases whose answer is known outside this site. The test suite computes each one with this tool's own engine on every change.
Two intervals forecast at 100 each; actuals 120 and 80. Misses 20 + 20 = 40 on 200 actual contacts. Result: interval accuracy 80.0%, total-volume accuracy 100.0%. Weighted absolute percentage error: the absolute misses summed over the actual volume.
What this tool cannot tell you
- Accuracy here is measured on the intervals entered. One day is a small sample; the tracking signal is more telling across weeks.
- WAPE and MAPE depend on the interval length. A daily forecast always scores better than the same forecast read by half hour.
- Agents busy ignore service level and shrinkage, so the agents needed to answer the missed contacts on time are more.
- The sample volumes are illustrative. Enter your own forecast and actual contacts for a result about your operation.