Research question and scope

This protocol asks what share of escalations reach an owner without a clear decision question. The unit is one escalation event, identified by the case, the sender, the receiver, and the time it was sent. An escalation is counted as complete only when it states the decision needed, the options, the evidence, and the deadline. The study measures escalation quality, not whether the owner made a good decision.

Set the scope to one queue, one owner group, and one escalation channel for a fixed window. Record the channel, the template, and the service expectation at the start. If the channel or template changes during the window, keep the earlier definition or analyze the change separately. A new form can change the measured rate without changing the underlying behavior.

Population and unit of analysis

Define the population as every escalation sent during the window, including ones that were resolved by chat and never reached a ticket. The unit is the event, not the case, because one case can be escalated several times as facts change. A re-escalation is a new event with its own quality score. Link the events so the analysis can show a chain.

Exclude routine status updates that do not request a decision and record the exclusion count. Include escalations that were withdrawn, because withdrawal may indicate that the question was unclear or that the sender solved it alone. Test the unit definition on several ordinary escalations before collection begins. Keep the test cases with the protocol so a later reviewer can check the definition.

Definitions that make the measure reproducible

Define a decision question as a sentence that asks for a specific choice, approval, or priority call. Define a complete escalation as one that names the case, the question, at least two options or a recommended option, the evidence, and the response deadline. Define a silent escalation as one that forwards a message without a question. A request to take a look is not a decision question.

Do not count a question that the sender could answer from the written policy. That is a training gap, not necessarily an escalation-quality problem, and it should be coded separately. Each definition needs one positive and one boundary example. Publish the definitions with the sample so a reviewer can apply them consistently.

Evidence standard and data dictionary

For each event, capture the sender, the receiver, the case reference, the message text, the attachments, the time sent, and the time of the first substantive reply. Add whether the owner asked a follow-up question before deciding. Record the decision and the time it was made. Keep the packet with the case so a later reviewer can reconstruct the exchange.

Also capture whether the sender had access to the policy or the customer record at the time. That field helps separate a communication problem from an access problem. If the escalation was verbal, record the note that was written afterward and mark it as reconstructed. A reviewer should be able to score the event without interviewing the sender.

Sampling and collection

Draw the sample from the escalation channel and the ticket system together, because some events appear in only one. Record the export date, the filters, the row count, and the selection rule. Include every sender and every day of the week so the result is not dominated by one person or one shift. A fixed interval is acceptable when the population is ordered by time.

Collect the message as it was sent, not as it appears after edits or thread cleanup. If the sender added a question after the first reply, record the edit time and analyze the original separately. Do not infer a decision question from the eventual answer. Note the collection date so the reader can see how much of the record has aged.

Analysis and reporting

Report the number of escalations, the share with a complete decision question, and the median time to a substantive reply. Break the rate down by sender, queue, and receiver. Show the most common missing element, such as the deadline or the options. Use the same denominator for every rate.

Compare complete and incomplete escalations on the time to decision and the number of follow-up messages. That comparison shows whether the quality gap has an operational cost. Report cases with missing evidence separately. If the sample is small, show counts rather than percentages.

Limitations and sources of error

The record cannot show whether the owner understood the question but chose not to answer. It also cannot capture an escalation that was resolved in a hallway conversation with no written trace. A complete escalation is not proof that the decision was correct. The study measures the quality of the request, not the quality of the outcome.

Messages can be edited or deleted, and a verbal handoff may never be recorded. If the sender and receiver are the same person, the event should be excluded or analyzed separately. Record these limitations with the sample. Do not use the result to rank individuals without accounting for task difficulty.

Review cadence and next action

Give the result to the queue owner and the sender group and ask for one change. The change might be a required decision-question field, a worked example, or a rule that no escalation is sent without a deadline. Record the owner, the effective date, and the measure that will show whether the change worked. Change one part of the template at a time.

Repeat the protocol after the change and compare the same fields. If completeness rose but decision time did not fall, the owner may be receiving better questions but still lacks capacity. If decision time fell only for one sender, the result may reflect training rather than the template. Close the review when the owner accepts the finding and the next sample is scheduled.

Sender training and template design

Escalation quality depends on what the sender was taught, not only on the form. Train senders with two examples: one complete escalation and one forwarded thread with no question. Ask them to rewrite the weak example. A short practice set teaches the pattern better than a policy paragraph. Keep the practice material with the template.

Test whether the template asks for the decision, the options, the evidence, and the deadline in separate fields. A single free-text box invites a narrative and hides the question. The form should make the missing element obvious before the sender submits. If the form allows a blank decision field, the rate will reflect the form as much as the training.

Cost of incomplete escalations

Measure the follow-up messages and the elapsed time that an incomplete escalation adds. Count the owner's time to reconstruct the case, not just the sender's time. Compare complete and incomplete events on the same case types. The difference is the operational cost of a missing question.

Do not assume every follow-up is waste. A clarifying question can be the right response to a genuinely complex case. Separate follow-ups that ask for missing basics from follow-ups that explore a real trade-off. Only the first group is a quality signal. Report both so the owner does not over-correct.

Reporting without ranking people

Report escalation quality by queue, case type, and missing element rather than by sender name. A sender with difficult cases may look worse than one with routine work. If individual coaching is needed, use the examples privately and pair them with the case mix. The report should improve the system before it evaluates the person.

Include the share of escalations that the sender resolved without escalation after training. That measure shows whether the queue is teaching judgment or creating dependence. A falling escalation rate with stable quality is a good sign. A falling rate with rising customer errors is not.

Replication and version control

Store the definitions, the template version, and the sample query with the report. Repeat the measurement after the template change using the same fields. If the channel changes, publish a new baseline. A second analyst should be able to reproduce the classification from the packet.

Set the next review date before closing the current one. Assign the owner and the sample size. Keep the earlier report available for comparison. A quality measure that is not repeated cannot show whether the change worked.