With the matrix in place, Hora handles the routine on her own and sends Patrícia whatever is not routine. That creates a new job that no manager's job description had: the exception queue. In this lesson you work that queue in a simulation and leave with the daily and weekly routine of someone who supervises agents.
4Build the supervision routine and measure its load
Goal: work the exception queue: approve, send back or stop each case with a reason, an owner and a deadline, in a daily and weekly routine.
01 · The path of an exception
No exception goes without an owner. Tap each step.
Every decision becomes learning: a new test case, an adjustment to the matrix or to the constitution
Click each block in the diagram to see what it does.
Normal, doubtful and critical cases; exception package and supervision rituals: Gestoria method (INEMA).
02 · Simulation: Patrícia's queue at 5 p.m.
Simulated conversation for practice. Hora brings the day's exceptions; you decide each one. The comment appears right after.
Hora · sales agent (simulation)Horizonte Distribuidora, fictional study company
Choose your answer. The comment appears right after, and the conversation goes on according to your choice.
03 · The routine of whoever supervises
The simulation is one late afternoon. These are the four pieces that make the queue work every day.
1. Who resolves each type
Insufficient or conflicting data goes back to whoever can complete it (the agent, with the customer, or Juliana). A request outside the policy goes to whoever has the authority limit (Patrícia, or Carlos above her). A critical case goes to a defined person and can halt the operation.
The destination of each type is written in the constitution. An exception that needs a meeting to find out whose it is has already arrived late.
At HorizonteMissing data → send back to the customer through Hora, with a deadline. Outside the policy → Patrícia (up to 10%) or Carlos. Manipulation or data leaking out → stop, Patrícia and Rafael.
2. Demand the package
An approval request without evidence shifts the work of investigating onto you. The agent brings the package: case, pending action with the exact content, reason, evidence, what it already tried, the rule involved and a recommendation.
An incomplete package is sent back to the agent, not completed by you. Within two weeks, the packages arrive whole.
Quick quizCan you decide by reading only the package, without opening the whole conversation? If not, the package is incomplete.
3. Stopping is a button
Stopping is different from sending back: there is risk, so the case (or the whole agent, on the off switch) leaves normal operation until someone understands what happened.
The script fits in four lines: stop, contain (confirm through a known channel, undo what you can), log the incident and look for similar cases. Afterward, the cause becomes a test case.
The hotel caseStop the order. Call the buyer at the phone number on file. Log the incident. Scan this week's conversations for similar text.
4. Measure the supervision load
The queue has a cost too: how long each case waits and how many minutes a day you spend deciding. If the queue grows, either the matrix is too tight or the agent is escalating without a package.
These numbers feed the cost per result in lesson 2 and the decision to promote in lesson 5. Supervision that nobody measures becomes the manager's invisible overtime.
At Horizonte (fictional numbers)First week: 9 exceptions a day, 40 minutes of Patrícia's time, average wait of 2 h. Third week: 3 a day, 15 minutes, wait of 40 minutes.
04 · Templates to copy
The package the agent brings, the record of your decision, and the supervision routine.
Exception package + decision
EXCEPTION PACKAGE · case <id> · <date and time>
Agent: <name> · version: <v> · process: <...>
TYPE: [ ] doubtful (data) [ ] outside policy [ ] critical
PENDING ACTION (exact content): <what would be done, with values>
WHY IT IS IN THE QUEUE: <rule or limit that blocked it>
EVIDENCE: <conversation, record, document, log>
ALREADY TRIED: <what the agent did before escalating>
AGENT RECOMMENDATION (within its limits): <...>
DEADLINE TO DECIDE: <...> · OWNER: <name>
HUMAN DECISION
[ ] APPROVE this content, valid until <deadline>
[ ] SEND BACK: missing <what> · to <whom> · by <deadline>
[ ] STOP: risk <which> · containment <what was done> · notified <whom>
JUSTIFICATION (rule or reason): <...>
Decided by: <name> · <date and time>
PROPOSED LEARNING: <new test case / line in the matrix / sentence in the constitution>
Supervision routine
SUPERVISION ROUTINE · <process> · owner: <name> · backup: <name>
DAILY (15 min, fixed time: <hh:mm>)
[ ] Exception queue: does everything have an owner and a deadline? anything critical stuck?
[ ] Critical errors today: were there any? already logged as an incident?
[ ] Stopped operations: any waiting for a decision?
[ ] Budget: today's spend within what you expected?
[ ] Note: number of exceptions, minutes spent, longest wait
WEEKLY (45 min, <day>)
1. Did the result improve against a comparable baseline?
2. Did quality, critical errors and exceptions change? Which causes keep repeating?
3. How much human work was needed and where (supervision minutes)?
4. What was the full cost, including retries and rework?
5. Which action needs less autonomy, or a pause?
6. Which action has evidence to be promoted (checklist from lesson 5)?
7. What learning becomes a test case, a line in the matrix or a sentence in the constitution?
8. Who owns each action and when will it be reviewed?
PER CHANGE (model, tool, constitution, matrix)
[ ] Test cases run before it goes in
[ ] Affected actions go back to yellow until there is new evidence
[ ] Change log: what, who approved, when, how to reverse it
05 · Practice
Train on Patrícia's queue; then build your own routine and measure it for a week.
1 · The simulated queue with the top score
10 min
Run this lesson's simulation. If you don't reach the top score, read the comments, start over and try again.
Done when: you finished the simulation with the top score and can say, for each of the three cases, which verb you used and why.
2 · Three justified decisions
20 min
Fill in an exception package with the decision for each case in the simulation: the hotel's order with the hidden instruction, Seu Antônio's 15% discount, and the customer without a CNPJ. In the justification, cite the rule from the matrix or the constitution that decided it.
Done when: you have three packages, each with a different verb (stop, approve with a reason, send back), the rule cited, the next step with a deadline and a proposed learning.
3 · Your routine, measured
15 min + 1 week
Copy the supervision routine and adapt it to your process: fixed time, backup and the items that make sense. Run the daily routine for a week, noting the number of exceptions, minutes spent and longest wait. Close the week with the 8 questions.
Done when: the routine is written with an owner and a backup, 5 days are logged with exceptions, minutes and wait, and the weekly review answered the 8 questions with at least one decision.
06 · Quick quiz
5 questions. The explanation appears after your attempt.
Question 1 of 5
What is the difference between sending back and stopping?
Correct
They are distinct decisions and leave distinct records. The customer without a CNPJ is a send-back; the order with the hidden instruction is a stop. Mixing the two lets an incident pass as an ordinary pending item.
Not that one
They are distinct decisions and leave distinct records. The customer without a CNPJ is a send-back; the order with the hidden instruction is a stop. Mixing the two lets an incident pass as an ordinary pending item.Review: The path of an exception
Question 2 of 5
Hora only asks, "approve order 4471?" What do you do?
Correct
Without a package, you are the one investigating, every day. Sending back the incomplete request teaches the pattern; within a few weeks the packages arrive complete.
Not that one
Without a package, you are the one investigating, every day. Sending back the incomplete request teaches the pattern; within a few weeks the packages arrive complete.Review: The routine of whoever supervises
Question 3 of 5
Seu Antônio asks for 15%; Hora's authority limit is 5% and yours is 10%. Which decision is best?
Correct
Within your authority limit, deciding is your job, with the reason in writing. Giving 15% goes beyond your authority limit and sets a precedent; escalating to the owner what is yours clogs their queue.
Not that one
Within your authority limit, deciding is your job, with the reason in writing. Giving 15% goes beyond your authority limit and sets a precedent; escalating to the owner what is yours clogs their queue.Review: Simulation: Patrícia's queue at 5 p.m.
Question 4 of 5
Why sort the queue by risk and deadline, and not by order of arrival?
Correct
An order with a manipulation attempt and R$ 28 mil at stake needs a decision before a missing record does. With first come, first served, priority is left to luck.
Not that one
An order with a manipulation attempt and R$ 28 mil at stake needs a decision before a missing record does. With first come, first served, priority is left to luck.Review: The path of an exception
Question 5 of 5
Patrícia's queue went from 3 to 12 exceptions a day. What might this indicate?
Correct
Supervision load is an indicator. Before you turn the agent off or approve in bulk, look at the causes that repeat in the weekly review; often the fix is one line in the matrix or in the constitution.
Not that one
Supervision load is an indicator. Before you turn the agent off or approve in bulk, look at the causes that repeat in the weekly review; often the fix is one line in the matrix or in the constitution.Review: The routine of whoever supervises
Quiz complete
You got {a} of {n} right. Get them all right and the lesson is marked as complete.
Pick an answer. The explanation shows up after you try.
07 · Summary
Three types of caseNormal follows the matrix; doubtful goes to the queue; critical goes first and can stop everything.
Package before the decisionExact content, reason, evidence and recommendation. No package, send it back.
Three verbsApprove with a reason, send back with a deadline, stop with containment and a record.
Measured routineA short daily one, a weekly one with 8 questions, and supervision load as a number.