Onboarding

A quality sampling plan for a virtual assistant's first month

A first-month quality sampling plan that sizes reviews by risk, scores with a short rubric, and turns repeat defects into process fixes.

By OnboardingEmployees editorial team | | 5 min read

A quality sampling plan for a virtual assistant's first month
On this page
  1. Sample by risk instead of volume
  2. Define the review unit and the evidence
  3. Set a weekly sample size that fits capacity
  4. Score with a short rubric
  5. Give feedback the same day
  6. Track repeat defects, not just scores
  7. Increase independence when the sample is stable
  8. Publish the sampling plan to the client
  9. Related reading

Sample by risk instead of volume

Not every task deserves the same review rate. A client-facing message, a payment change, or an access request carries more consequence than an internal note. Rank the assistant's tasks by what happens if the output is wrong. Sample the high-consequence work more often and let routine work run with a lighter check. Write the ranking down and review it with the client before the first sample.

Write the sampling rate beside each task so the reviewer does not decide from memory. A high-risk task might be reviewed in full during the first week, while a low-risk task might be sampled once a day. The rate should fit the reviewer's available time. A plan that assumes unlimited review capacity will be abandoned by the second week. Adjust the rate when the task mix changes rather than keeping last month's plan.

Define the review unit and the evidence

Decide what counts as one review unit before sampling starts. It might be a completed request, a sent message, or a closed ticket. Name the evidence the reviewer will inspect, such as the source request, the output, and the system record. Without a shared unit, two reviewers can report different volumes from the same day. Put the unit and the evidence in the review template.

Keep the evidence in one place so the sample can be reopened later. A screenshot of the final message is not enough if the source request is missing. The reviewer should be able to see what the assistant received and what the client saw. That record is what makes a quality finding reproducible. Store the sample in a folder with the date and the reviewer's name.

Set a weekly sample size that fits capacity

Calculate the sample size from the reviewer's available hours, not from a target percentage. If a review takes ten minutes and the reviewer has three hours, the realistic sample is eighteen units. Write that number down and protect the time on the calendar. A smaller sample that actually happens is more useful than a large sample that is skipped. Put the calculation in the plan so the number is not mistaken for a quota.

Increase the sample only when the reviewer has capacity and the work is stable enough to learn from more cases. Decrease it when the queue is unstable or the reviewer is also covering another role. Review the sample size at the end of each week. The plan should reflect the real operating constraint. Record the reason for each change in sample size.

Score with a short rubric

Use four or five criteria that the reviewer can observe directly. For a client message, those might be accuracy, completeness, tone, next step, and timeliness. Define each criterion with an example of passing and failing work. A short rubric produces a conversation about the work instead of a debate about taste. Give the reviewer a one-page rubric with the examples attached.

Do not combine the criteria into one score too early. A single number hides whether the assistant needs a fact check, a writing fix, or a process change. Keep the individual results and use the pattern to decide what to coach. The total score is useful for tracking, not for explaining. Review the pattern across criteria before changing the rubric.

Give feedback the same day

Return the review on the same day the work was sampled. A correction delivered two days later is harder to connect to the original decision. Attach the evidence and name the specific criterion that failed. The assistant should be able to see exactly what to change on the next attempt. Ask the assistant whether the correction arrived in time to use it.

Separate a correction from a new instruction. If the rubric did not cover the issue, update the rubric before treating the output as a failure. Ask the assistant what information was missing. A review that only points out errors teaches people to hide uncertainty instead of raising it. Record the instruction gap separately from the performance result. Review the gap at the next weekly check-in.

Track repeat defects, not just scores

Count how often the same defect appears across the sample. A single tone miss may be a one-off, while the same missing next step three times points to an instruction gap. Group the defects by criterion and by task. The pattern tells the reviewer where to spend the next coaching session. Review the defect count by task and by week.

Do not use the repeat count as a performance rating during the first month. Use it to decide whether to rewrite a task guide, add an example, or change the review step. Record the fix and the date it takes effect. The next sample should show whether the fix traveled to the work. Share the pattern with the client only after the process fix is agreed.

Increase independence when the sample is stable

Move a task to a lighter review only when the sample shows a stable result across several cycles. Stability means the same criteria pass without correction, not that the assistant has been in the role for a set number of days. Keep the evidence for the decision so the client can see why the review rate changed. Write the evidence for stability beside the task, not in a separate report.

If the defect rate rises after the change, return to the previous rate without treating it as a failure of the assistant. The purpose of sampling is to calibrate oversight, not to prove trust. Tell the assistant what triggered the change and what would restore the lighter review. A visible rule is easier to accept than a sudden loss of autonomy. Explain the change to the assistant before the next shift.

Publish the sampling plan to the client

Share the sampling plan with the client at the start of the engagement. Explain which tasks are reviewed in full and which are sampled. Give the client a route to request a deeper review for a specific concern. The plan sets an expectation for quality without promising perfection. Name the person who will handle a deeper review request. Tell the client how to reach that person.

Send a short monthly summary of the sample size, the criteria, and the process fixes made. Do not include individual scores that the client could mistake for a performance ranking. The summary should show that review happened and that the findings changed something. A sampling plan is credible when its results are visible and its limits are honest. Keep the summary short enough for the client to read in one pass.

Continue building the workflow

Related support: onboarding operations support, contact us. The U.S. Office of Personnel Management — Training and Development guidance on employee development and performance support.