List the exceptions that reach a manager
Start with the exceptions that already interrupt a manager's day. Review two weeks of chat messages, emails, and meeting notes for requests that needed a decision above the assistant's level. Group them by the decision being asked for, not by the person who asked. The list should describe real work rather than a theoretical risk register. Include the requests that arrived through a phone call, because those are easy to miss in a message review.
Ask the manager which items could have waited and which needed an immediate call. That distinction is the beginning of severity. Do not copy an escalation list from another company, because its thresholds reflect a different client, contract, and tolerance for delay. Build the first draft from this team's actual interruptions. Write the date range of the review beside the list so the evidence is clear.
Assign severity by consequence
Define severity by what happens if the decision waits. A missed client deadline, a payment error, a security concern, and a legal notice belong near the top. A formatting question or a routine approval can wait for the next review block. Consequence gives the assistant a reason to act instead of a label to memorize. Keep the definitions short enough to fit on one screen.
Use three levels at most. More levels invite debate about whether an issue is a two or a three while the clock runs. Name each level with the outcome it protects, such as client commitment, financial accuracy, or access safety. Print the consequence beside the level so the rule survives a change in staff. Test the levels against a real event from the past month.
Name one owner and one backup per row
Every row needs a primary owner and a backup who can make the same decision. Use a role name as well as a person so the matrix survives a vacation or a departure. A shared inbox is not an owner, because nobody is accountable for the next action. The assistant should know exactly who receives the escalation. Put the owner's name and role in the same cell to avoid ambiguity.
Give the backup the same authority as the primary, not just the task of forwarding the message. If the backup cannot approve the decision, the escalation will stall at the worst moment. Review the backup list monthly and after any team change. An outdated name is worse than a clear gap because it creates false confidence. Tell the assistant what to do when both people are unavailable.
Set response windows that fit the business
Write a response window for each severity level. A client commitment may need an answer within thirty minutes, while a reporting correction may allow one business day. The window should describe when the owner responds, not when the assistant sends the first message. Both times matter and should be recorded separately. Record the response time from the moment the escalation is sent.
Check the windows against the client's service agreement and the team's actual coverage. A rule that requires a response before the owner starts work is not usable. Add a named fallback for hours when the primary and backup are offline. An escalation matrix that ignores the clock is only a list of names. Review the windows when the client's hours change.
Write triggers in observable terms
Replace words like urgent and sensitive with something a person can see. A trigger might be a request to change a bank account, a message from a regulator, or a client asking to cancel a signed service. Observable triggers reduce the chance that two assistants classify the same event differently. Give two examples for each trigger so the boundary is visible.
Include a catch-all for anything the assistant cannot resolve with the current instructions. The catch-all should say where the item goes and what information must travel with it. Do not punish the assistant for using the catch-all when the matrix is silent. Those uses are evidence of a missing row, not a performance problem. Tell the assistant what information must accompany a catch-all escalation.
Test the matrix with tabletop scenarios
Run five short scenarios before the assistant starts taking live work. Read each scenario aloud and ask the assistant to name the severity, owner, and next message. Compare the answer with the matrix and correct the gaps on the spot. The exercise takes less than an hour and exposes ambiguous language quickly. Keep a written record of the changes made during the test.
Add one scenario where the primary owner is unavailable and one where the client disagrees with the severity. These cases test the backup path and the judgment behind the labels. Record the changes made during the test. A matrix that has never been rehearsed is a document, not an operating rule. Ask the assistant to explain the decision in their own words.
Keep the matrix from becoming a bottleneck
Review how often each row is used after the first two weeks. A row that fires every day may be a normal process that needs a delegated decision instead of an escalation. A row that never fires may be too vague or may describe a risk the team no longer has. Use the pattern to simplify the matrix. Compare the escalation count with the volume of work in the same period.
Give the assistant authority to resolve low-consequence exceptions within a written limit. State the limit in observable terms, such as a credit amount or a schedule change. Keep the higher levels untouched until the record shows the lower delegation is stable. Escalation should protect the business, not train everyone to wait for a manager. Write the delegated limit in the matrix, not in a separate message.
Review escalations at the end of the week
Hold a short weekly review of every escalation. Record what triggered it, who responded, how long the response took, and what changed as a result. Look for repeated triggers that point to a missing instruction or an unrealistic client expectation. The review should produce one or two process changes, not a longer meeting. Invite the owner who received the escalation to the review.
Share the pattern with the client when it affects their work. Ask whether the severity levels match their priorities and whether any response window is unrealistic. Update the matrix and tell the assistant what changed. A matrix that is reviewed becomes a learning record; one that is filed away becomes another document nobody trusts. Put the updated matrix in the same place as the task instructions.
Continue building the workflow
Related support: onboarding operations support, contact us. The U.S. Small Business Administration — Hire and manage employees guidance on setting expectations and managing employee performance.
Related reading
A client tone guide for a new virtual assistant
A practical tone guide built from real client messages, with examples for delays, errors, and routine updates a new virtual assistant can follow.
A recurring task calendar that keeps weekly work visible
A recurring task calendar with clear cadence, start dates, backups, and a monthly review that keeps weekly work from disappearing.
A shadowing schedule that protects the manager's calendar
A short shadowing plan that gives a new virtual assistant real observation time without consuming the manager's entire week.
