What does send-back rate measure?
The share of a Digital FTE's proposals in one task class that a person sent back with a reason rather than approving, over one cycle. On ftes.ai every send-back requires a reason, so the rate is always accompanied by a log of why, and the Performance tab shows both per class. The rate is the complement of the approval rate over the same decided items and adds nothing on its own; its value is that it is the doorway to the reasons, which are the most useful management information the workspace produces.
A manager who reads the rate and not the reasons has read the least useful half.
How is it computed, with the inputs left blank?
Role: [role]. Class: [class]. Cycle: [date] to [date]. Decided: [D]. Sent back: [S]. Send-back rate: [S] divided by [D]. Reasons, sorted: facts the memory lacked or had wrong, [F]; rules the remit did not state or stated unclearly, [R]; edge cases outside the class's bounds, [E]; tone, format, or preference, [T]. Shares: [F], [R], [E], [T] each divided by [S]. Trend: the same four shares for the previous cycle, [F], [R], [E], [T]. Reading: the mix should move from facts and rules toward tone; where it does not, the largest category names the fix.
The four categories are the manager's, applied when reading the reasons; the FTE records the reason text, and the manager sorts it.
- Rate: sent back over decided, per class
- Reasons: facts, rules, edge cases, tone
- Trend: the mix across cycles
What does each reason category tell the manager to change?
Facts: the Brand Brain or the approved sources have a gap or an error; correct the item, and confirm its provenance. Rules: the remit's Owns or Asks list is missing a rule or states one ambiguously; add it, and record the change at the review. Edge cases: the class's bounds are too wide, or the class contains two classes; narrow the bounds or split the class, and the Promotion Standard will apply to each separately. Tone: the voice in the Brand Brain needs a correction, or the approver has a preference the template can absorb; adjust, and expect the share to fall to a background level rather than zero. A send-back that fits none of the four is usually a judgement item mis-sorted onto the Owns list.
Each category maps to one tab: Memory, Setup, Setup, Memory. That is the design.
What does the curve look like over the first month?
Week one: facts and rules dominate, because the memory was pre-filled from the site and the remit was written before the work started. Week two: rules fall as the remit absorbs them; edge cases appear as the class meets its boundaries. Week three: tone dominates, which is the sign the substance has transferred. Week four: the rate is at its background level for the class, mostly tone, and the class is a candidate for the Promotion Standard's second condition. A class still dominated by facts in week four has a memory gap the manager has not filled, and the fix is specific and quick.
The myth piece on working perfectly from day one describes this curve from the manager's side.
How does it mislead?
When reasons are recorded lazily: an approver who writes the same word every time produces a log that cannot be sorted, and the manager should ask for reasons that name the fact, rule, or preference. When the rate is compared across classes: a difficult class has a higher background rate than a routine one, and the comparison says nothing. When a rising rate is read as failure: a rate rising after a source changed, a knowledge base article edited, a fee schedule updated, is the record doing its job, and the detection log will usually show the cause. And when a zero rate on a new class is read as success rather than as an approver who is not reading.
The approval rate explainer covers the mirror-image pitfalls.
How is it applied?
Read the reasons weekly per class, sort them into the four categories, and make the one change the largest category points at. Record the sort at the monthly review in the note, with the trend from the previous cycle. Use the tone share, not the rate, as the Promotion Standard's second condition. When the rate rises on a promoted class for a facts or rules reason, step the class back the same day and find the cause. When an approver's reasons are unsortable, have the conversation about reasons that name things; the record is only as good as its reasons.
The glossary entry on the send-back gives the short definition and the platform's reason rule.