What does approval rate measure?
The share of a Digital FTE's proposals in one task class that a person approved as proposed, over one cycle. On ftes.ai it is computed in the Performance tab per class: the count of proposals approved divided by the count decided, where decided means approved or sent back, and items still waiting are excluded because they carry no decision yet. Approved with an edit counts as approved; the edit becomes a send-back reason only if the approver records one. The rate is shown per class and, for orientation only, per role.
It answers one question: are this class's proposals landing? It does not answer why the others did not.
How is it computed, with the inputs left blank?
Role: [role]. Class: [class]. Cycle: [date] to [date]. Proposals made: [N]. Approved as proposed or with edits: [A]. Sent back with a reason: [S]. Still waiting at cycle end: [W]. Decided: [A] plus [S]. Approval rate: [A] divided by ([A] plus [S]). Volume check: is [A] plus [S] large enough that one send-back moves the rate by less than the manager's tolerance? If not, the rate is noise this cycle and the reasons are the only signal.
Write the same lines for each class the role handles. Never add the classes together.
- Numerator: approved, including approved with edits
- Denominator: approved plus sent back; waiting excluded
- Scope: one class, one cycle
Why is it read per class and not per role?
Because a role handles classes with different difficulty and different consequences, and a blended rate hides the one that matters. A Customer Support FTE may have a clean rate on order-status replies and a poor one on returns-policy replies; the role-level rate sits between them and tells the manager nothing actionable, while the per-class rates say exactly where the knowledge base has a gap. The Promotion Standard is applied per class for the same reason, and a role-level rate can never justify promoting anything.
The workforce view shows a role-level rate for orientation across roles; the decision is always made one level down.
What patterns does it show at the review?
Rising through the first month: calibration is working, the memory and remit are absorbing the send-backs. Flat and high with tone-only send-backs: the class may be ready for the Promotion Standard. Flat and low: the class was given judgement work, or the approver disagrees with the remit; read the reasons and the Three-List Remit. Falling after a promotion: the step-back trigger, and the class returns a level. Falling with no remit change: drift, usually a source that changed, a knowledge base article edited, a fee schedule updated, and the detection log will show it. Rising suddenly with no send-backs at all: an approver may be rubber-stamping, which is a management conversation rather than a metric.
The trend across cycles is the signal; a single cycle's point is a number.
How does it mislead?
Three ways. Small volume: a class with a handful of decisions swings wildly on one send-back, and the rate should be read as noise until the volume is meaningful. Rubber-stamping: an approver who approves everything produces a perfect rate and a useless record, and the first substantive error at autonomy will reveal it; the guard is reading the reasons and occasionally sampling approved items. Blending: a role-level rate, or a rate across cycles of different length, hides the class or the period that changed. None of these is a flaw in the metric; each is a reason it is never read alone.
The send-back rate explainer covers the other half of the same record.
How is it applied?
Read it weekly per class in the Performance tab alongside the send-back reasons for the same class. At the monthly review, write the per-class rate and trend into the review note with the reasons summarised, and use it as the first of the Promotion Standard's five conditions, never as the only one. When it falls, open the reasons and the detection log before changing anything. When it is perfect, sample five approved items and check them; a perfect rate on a new class is a question, not an answer.
The monthly review note template has a line per class for exactly this.