Virtual Assistant KPIs: How to Measure Performance Without Rewarding Busywork

Most virtual assistant KPIs end up looking like this: the weekly update says 40 tasks completed.

Sounds great. Until you notice several were ticked off before review, one big job was split into a pile of tiny subtasks, and a customer problem is still unsolved despite five friendly replies.

The number is accurate. Forty boxes were ticked. It just doesn't tell you whether forty useful things happened.

That's the trap with most virtual assistant KPIs. They measure activity, because activity is easy to count. And whatever you count is what people learn to produce.

Good KPIs (key performance indicators) help you make a decision: Is the work getting done properly? Where are delays coming from? What needs fixing?

This guide shows you how to build a small, honest scorecard, with a worked example showing how "95%" can mean three completely different things.

Here, a virtual assistant (VA) is an independent freelancer or agency providing a service. You're measuring the service, not watching a person's presence.

Define one finished result before you count anything

Pick a unit that represents a meaningful result:

• Inbox admin: a correctly handled eligible enquiry.
• Research: an accepted brief.
• Product admin: a checked listing ready for the next step.

Don't mix units. A research brief and a renamed file might both be "tasks" in an app, but adding them together gives you a meaningless number.

Then define exactly where each case starts and ends.

If your VA only prepares drafts, "done" might be an accepted draft, not a customer reply you send. If they own resolving the issue within policy, "sent a reply" isn't done.

Write it down in plain words:

"One case is one eligible customer enquiry prepared for my review, with the correct source checked and all required fields complete."

Otherwise, two people can calculate two different results from the same week without either making a maths mistake.

Every percentage needs to say "out of what?"

"Completion rate" sounds clear. It's not. It could mean completed tasks out of:

• tasks assigned,
• tasks due this week, or
• tasks started this week.

Those answer completely different questions.

For on-time delivery, a good starting point is: work due this period, under the agreed scope. Decide how you'll handle cancelled, rescheduled and incomplete-input cases. Keep the history of changes, so a missed deadline can't just be quietly moved and disappear.

For quality, say whether you checked everything or a sample. "No errors found" means something very different in 5 checked cases than in 50.

Don't quietly leave out awkward cases. A task waiting for your approval still affects delivery to the customer. Keep it in the overall picture, and record the cause separately.

And if nothing was due? Show "not applicable," not 100%. A week with no work isn't a week of perfect performance.

Speed without quality rewards the wrong behaviour

A response-time target can encourage quick "got it!" replies while the actual problem stays unsolved.

A completion target can encourage closing things early.

Neither requires anyone to be dishonest. The measure just rewards the wrong finish line.

So always pair speed with quality. Ask:

• Did the work meet the requirements first time?
• Was "complete" actually justified?

A draft delivered fast but needing major rework shouldn't score the same as a correct draft delivered fast.

Separate serious errors from preferences. A wrong price and a heading you'd phrase differently aren't the same kind of "error." Define your categories before using them to judge anyone.

And add a guardrail: something that must stay acceptable while another number improves. For example: faster handling must not increase wrong recipients, unsupported claims or unauthorised actions.

You don't need targets on day one. Watch how the service behaves first, then agree realistic standards. A copied benchmark can be very precise without being right for your work.

Separate a slow VA from a slow approval

Some delays aren't caused by your VA at all.

A draft might sit ready for your review while the overall task stays "incomplete." The customer experiences a delay. But asking your VA to work faster won't fix your approval queue.

So track waiting states where they matter:

• waiting for client input,
• waiting for an outside response,
• needs VA correction.

The Kanban Guide defines a few useful ideas:

• Throughput: finished work per period.
• Work item age: how long unfinished work has been open.
• Cycle time: how long it took finished work to go from start to done.

It also stresses that start and finish points must be clearly defined.

Why this matters: a scorecard showing only finished tasks can look healthy while an old, difficult case quietly sits untouched. Always include a view of ageing open items and who needs to act next.

Use cause labels carefully, and back them up with evidence. Don't let a dropdown become an excuse, or a way to automatically blame someone.

Our time zone guide explains how one question can eat a whole working day across time zones. A good performance conversation asks how to remove that wait, not just who to blame for it.

A small scorecard beats a big dashboard

Start with three or four measures that answer different questions:

• Volume: how much eligible work was delivered.
• Timely correct completion: was it right and on time?
• Rework or exceptions: how much needed fixing?
• Ageing open work: what's stuck, and for how long?

For each measure, record:

• definition,
• data source,
• review period,
• denominator ("out of what"),
• who owns it,
• what you'll do if it changes.

If you can't say what you'd do with the result, it probably doesn't belong on the scorecard.

Keep definitions the same week to week. If you change one, note the date. Don't draw a trend line across two different definitions.

A simple tool helps. Google Sheets' dropdown lists let you standardise status labels and cause categories, making counting easier. But someone still needs to check each label actually matches the case.

Don't build a huge surveillance system for a modest service. A small, accurate register with a few reviewed examples beats any number of dashboards built on fuzzy activity data.

A worked example: why "95%" can be three different numbers

Here's a hypothetical example.

A VA prepares customer enquiry drafts for the owner. "Done" means a draft that follows the current response rules and is ready for approval. Sending and unusual compensation decisions stay with the owner.

This week, 40 eligible cases were due. All 40 were eventually submitted.

• 38 were submitted on time. But 2 of those needed significant correction.
• 2 were submitted late, but were correct first time.

The owner checked the first submitted version of every case. A later correction doesn't turn a failed first attempt into a first-time success.

Now the maths:

• On-time submission: 38 ÷ 40 = 95%
• First-time quality: 38 ÷ 40 = 95% (but a different 38 cases!)
• Correct first time AND on time: 36 ÷ 40 = 90%
• Significant rework: 2 ÷ 40 = 5%

Notice that two "95%" figures describe different sets of cases. That's exactly why every percentage needs its own definition.

These are example numbers, not recommended targets.

Now the owner looks at the two late cases:

• One waited for a missing client decision.
• One was delayed inside the VA's preparation.

Both stay counted as late, because the customer doesn't care whose fault it was. But the cause notes show who needs to improve what.

The review now has three specific actions:

1. Clarify the rule behind the two corrected drafts.
2. Fix the client decision delay.
3. Investigate the VA-side delay.

"40 tasks completed" would have revealed none of that.

Open the corrected cases before setting a new target

Look at the two drafts that needed fixing. Find the exact standard that was missed, and check: was that standard available when the work was done?

A missing rule and an overlooked rule are different problems.

Suppose both drafts used an old delivery promise from a previous template. Check how the old template was still accessible. The fix might be deleting it and practising the new rule, not demanding a higher score.

Look at the late VA-side case separately. Was it displaced by an urgent request? Underestimated? Forgotten? Blocked without being recorded? Don't guess from a timestamp.

Ask your VA for their view. They might spot a source conflict or a scope change the scorecard can't show.

And don't treat the first measured week as a permanent baseline. It might have unusual volume or a new type of work. Use early weeks to test your definitions before judging anyone.

Check quality with a deliberate sample when you can't check everything

If there's too much work to review it all, agree a sampling method. Pick cases in a way that doesn't let only the easy or best-looking ones be shown.

Keep two kinds of check separate:

• General sample: a representative mix of routine work.
• Targeted checks: every high-risk exception.

Don't merge them into one percentage. A targeted hunt for errors isn't representative of the whole queue.

Record the sample size, how you chose it and what you checked for. Say the result applies to reviewed cases. Don't claim unreviewed work is error-free.

If you find a serious problem, look wider. A wrong factual rule might have affected similar cases, even if you only spotted one.

Don't punish good questions, or reward fake independence

A target like "handle cases without help" can encourage healthy independence. It can also discourage your VA from escalating things that genuinely need you.

If a case is outside their authority, asking you is the right result. Counting every question as a failure would reward guessing.

Separate:

• Avoidable questions: about a clear, current rule they could have found.
• Legitimate questions: conflicting evidence, or decisions reserved for you.

Look at what each question was about, not just how many there were.

Same with other measures. A low complaint count is meaningless if complaints aren't recorded consistently. A high reply count can mean lots of unresolved conversations.

For every measure, ask: "How could this number improve while the service gets worse?" Then add a check against that.

And track serious exceptions separately. A tiny formatting issue and an unauthorised payment should never average out into a comfortable middle score.

Common VA KPIs, and the trap hidden in each

Here's a quick reference of popular measures, and what to watch for:

KPI	Useful for	Watch out for
Tasks completed	Rough volume	Tasks split up or closed early
Response time	Customer-facing work	Fast "got it" replies with no resolution
On-time delivery	Deadline work	Deadlines quietly moved after being missed
First-time-right rate	Quality	Corrections counted as first-time success
Rework rate	Process problems	Your preference changes counted as errors
Age of open cases	Spotting stuck work	Cases parked in "waiting" with no owner
Questions asked	Training gaps	Punishing correct escalations
Hours logged	Billing	Mistaking time spent for value delivered

Starter KPIs for different kinds of VA work

If you're not sure where to begin, here are sensible starting points by role. Pick two or three, define them properly, and adjust after a few weeks.

Inbox and admin: eligible messages handled correctly within the agreed window; age of the oldest unhandled message; items wrongly archived (from a sample).

Calendar management: bookings that needed rescheduling because of a preventable clash; invitations with incorrect details; requests waiting for a decision longer than agreed.

Customer support: cases resolved (not just replied to); reopened cases; escalations that arrived with enough information to decide.

Bookkeeping admin: documents filed with the correct source; unmatched transactions older than the period; invoices issued with missing required fields.

Research and lead generation: records accepted on review; records with a verifiable source for each key fact; duplicates created in the CRM.

Social media: posts published exactly as approved; corrections needed after publishing; posts held correctly because evidence was missing.

Notice that several of these count good pauses as success. That's deliberate. A measure that only rewards finishing will quietly punish your VA for doing the right thing.

Use each review to choose one change

At every review, look at the scorecard and a few real cases together. Pick the most important recurring issue, agree one change, and say what evidence would show improvement.

Don't change five rules at once. You won't know which one worked.

In the enquiry example, the first change might be removing the old template and confirming the new delivery rule. Next review: check similar cases for that specific error, rather than celebrating a higher score caused by easier work.

Record your own actions too. If you agreed to answer decisions faster, check whether you did. Otherwise the scorecard becomes one-sided.

Our performance issues guide covers what to do if problems persist. Metrics show you where to look. They don't replace a fair conversation about evidence and causes.

Good performance isn't the same as good value

Correct work matters. But it isn't automatically money in the bank.

A perfectly maintained report nobody uses doesn't justify its cost. Check whether the service still supports a real decision or customer outcome.

Our VA ROI guide separates time saved, costs and real results. Don't multiply minutes saved by an imagined hourly rate and call it revenue.

Keep enough history to spot patterns, but don't collect unnecessary personal data. Your VA should know what's measured and why. Open measures are much easier to discuss than hidden monitoring discovered later.

Your VA scorecard template

VA SCORECARD

Service and definition of one "done" case:

Review period:

Measure 1: Volume
  Definition / out of what / source / owner / action if it changes:

Measure 2: Correct first time and on time
  Definition / out of what / source / owner / action if it changes:

Measure 3: Rework
  Definition / out of what / source / owner / action if it changes:

Measure 4: Age of open cases
  Definition / source / owner / action if it changes:

Guardrail (must not get worse):

Sample method for quality checks:

Cases reviewed this period (links):

Main issue found:

One change agreed, and who owns it:

My actions as the client:

Evidence to check next review:

A useful scorecard lets both of you explain the same week: what was due, what was delivered correctly, what waited, what needed fixing, and what will change.

If your numbers can't support that conversation, improve the definitions before asking anyone to improve the score.

For the day-to-day side, read how to manage a virtual assistant.

Frequently asked questions

What KPIs should I use for a virtual assistant?

Start with a few tied to the service: correct first-time completion, on-time delivery of eligible work, rework rate and age of unfinished cases. Define each carefully, including what counts as done and what the percentage is out of.

Should I measure how many tasks my VA completes?

Task counts alone are misleading. Tasks can be split, closed early or vary in difficulty. Count meaningful results of the same type, and pair them with a quality check.

How do I measure quality if I can't check everything?

Use a deliberate sample. Record how many you checked and how you chose them, and be clear the result applies to reviewed cases.

Should I penalise my VA for asking questions?

No. Some questions are exactly the right thing to do, like escalating a decision outside their authority. Look at what each question was about rather than counting them.

What is a good on-time rate for a virtual assistant?

There's no universal benchmark. Set targets based on your actual commitments, consequences and the inputs you provide, after watching how the service behaves.