Why the benchmark needs a boundary
Abstract
Response time is a reasonable service measure and a poor excuse for continuous surveillance. A leadership team can learn whether requests are answered promptly without reading message content or tracking every click. The key is to define the service, sample the work, and report aggregates. This review sets out a benchmark method that respects the assistant and the people who write in. [1][2][3]
Methodology
The review draws on privacy and security guidance for data minimization and access control, and on occupational context for the support role. It defines response time from the requester's view and uses a consecutive sample plus a small audit of quality. The method keeps message bodies out of the dataset unless a specific case requires review. It reports medians and ranges rather than individual rankings. [1][2][4]
The question and its boundary
The question is how long a request waits before a meaningful response, not how quickly someone opens an email. A meaningful response may be an answer, a routing note, or a clear request for more information. The definition should be written before the sample is drawn. It should also say which channels count, because chat and phone calls behave differently from email. [2][3]
The boundary should exclude personal, legal, and medical messages unless a separate rule allows review. It should also exclude automated mail and newsletters that never required an answer. The benchmark covers work the desk can influence. Without that boundary, the measure drifts into surveillance of relationships rather than service. [1][3]
Why surveillance fails
Continuous monitoring changes behavior in ways that hurt the service. An assistant may answer quickly with a shallow response to protect a metric, or avoid complex requests that take longer. The measure improves while the quality falls. A benchmark should reward the right outcome, not the fastest keystroke. [1][2]
Surveillance also creates a privacy problem. Message subjects and sender names can reveal health, legal, or personal information. Collecting more data than the question needs increases the risk of a breach and erodes trust. The team should use the least data that can answer the service question. [2][4]
Define service time from the requester's view
Start the clock when the request reaches the desk, not when someone first opens it. Stop the clock when the requester receives a response that moves the work forward. An acknowledgment alone is not a response unless the process defines it that way. Record the definitions in a short service note. [2][3]
Use separate targets for different request types. A routine calendar confirmation is not the same as a board packet review. A single target across all work encourages the desk to prioritize easy messages. The benchmark should show the target and the volume for each category. [3][4]
Sampling and privacy
Use a consecutive sample from a defined window, such as two weeks. Add a small targeted sample of the oldest unresolved items so the tail is visible. Record the channel, the request type, the start time, the response time, and the outcome. Use categories instead of names and subjects. [2][4]
A second reviewer should check a portion of the cases for consistent coding. Access to the raw records should be limited to the reviewers and removed on a schedule. If a case requires reading content, use a documented exception with the executive's approval. Do not make content review the default. [1][3]
Reporting
Report the median, the spread, and the share of requests that met the target for each category. Add the volume so a small number of easy requests does not look like strong performance. Show the tail separately, because a few very old items matter more than a small change in the median. [3][4]
Do not publish a ranking of individual assistants. A team-level view supports staffing and process decisions without turning colleagues into competitors. If a person needs coaching, use a private conversation with examples. The public report should describe the service, not the person. [2][4]
Governance
Write down who can see the data, how long it is kept, and what the team will do with the result. The executive should approve the rules before the first sample. Review the rules when the tools or the team change. A benchmark without governance becomes an open-ended data collection project. [1][3]
Give the desk a way to challenge a result. A slow response may reflect a missing approval or a broken system rather than effort. The review should record the cause alongside the time. That context prevents the benchmark from becoming a blame exercise. [2][4]
A worked example
Consider a hypothetical leadership team that wants to know whether its support desk answers client requests promptly. The team samples two weeks of client threads and records the channel, the request type, and the time to a meaningful response. Routine scheduling requests have a median of four hours, while document reviews have a median of two business days. The difference is expected because the work is different. [2][4]
The report shows that most of the slow document reviews waited for an executive decision, not for the assistant. The team adds a named backup for approvals and keeps the same measurement for the next month. The benchmark becomes a way to fix a bottleneck rather than a way to rank the desk. That shift is the point of the exercise. [1][3]
What to test next
Test one service definition at a time. If the team adds a target for client scheduling, measure that category separately from internal requests. Do not change the target and the routing rule in the same week. Otherwise the team cannot tell which change produced the result. [2][4]
If the median improves but the tail does not, the fix addressed the easy cases. Look at the oldest unresolved items and the cause of each. A backup approver or a clearer escalation path may do more for the tail than a faster first response. The next test should target the tail directly. [1][3]
Communicating the benchmark
Explain the purpose before publishing the first result. The team should know that the measure is meant to improve routing and coverage, not to monitor individuals. Share the definitions, the sample window, and the privacy rules in the same note. A benchmark introduced without context will be treated as surveillance even if the data is aggregated. [1][4]
Publish a short summary and invite questions. If someone believes a result is unfair, ask for the case and record the cause. The goal is a measure that the desk trusts enough to use. A benchmark that people game or ignore is worse than no benchmark at all. [2][3]
Why the requester experience matters
A response-time measure can improve while the requester experience gets worse. A quick acknowledgment that does not answer the question may meet a target and still leave the sender waiting. The benchmark should include a small quality check, such as whether the response moved the request forward. That check keeps speed tied to usefulness. [1][4]
Ask a small sample of requesters whether they knew what happened next after the response. A clear routing note or a named owner is often more valuable than a faster but ambiguous reply. The service definition should reward a response that closes the loop. A benchmark that ignores the requester will optimize the wrong behavior. [2][3]
Keep the quality check small enough to repeat. A three-question review of ten sampled threads is enough to show whether fast responses were also clear. Do not turn the review into a second scoring system. The purpose is to catch a gap between speed and service before it becomes the norm. [1][4]
Limits and responsible use
This method cannot measure kindness, judgment, or the quality of a relationship. It also cannot capture work that moved to a call or a hallway conversation. The report should state those blind spots. A fast response to the wrong question is still a failure. [1][3]
Use the benchmark to improve routing, coverage, and service definitions, not to intensify monitoring. A leadership team that trusts its support staff will get better information than one that watches every click. For related reading, see our guide to <a href="/blog/build-a-vendor-mail-lane-that-keeps-invoices-out-of-the-executive-inbox">building a vendor mail lane</a> and our <a href="/services">executive support services</a>. [2][4]
Key findings
- Define response time from the requester's view and set separate targets for different request types. [2][3]
- Use a consecutive sample plus the oldest unresolved items, and keep message content out of the default dataset. [2][4]
- Report medians, spread, target attainment, and volume by category rather than individual rankings. [3][4]
- Approve access, retention, and use rules before the first sample, and review them when tools change. [1][3]
- Record the cause of a slow response so the benchmark improves routing instead of blaming a person. [2][4]
Sources
- Privacy Framework, NIST
- Protecting Personal Information: A Guide for Business, Federal Trade Commission
- Security and Privacy Controls for Information Systems and Organizations, NIST
- Executive Secretaries and Executive Administrative Assistants, U.S. Bureau of Labor Statistics
- Hours of Work, U.S. Department of Labor