Most “ROI of AI” content is vibes. A confident multiplier, no method, no baseline, and no way for you to check it against your own business.

So here is the honest position. We don’t publish a portfolio-wide productivity multiplier, because the gain depends heavily on what a given role actually does, and a single number would mislead more readers than it helped. What we can give you is the measurement method, the real tool costs, one real pricing figure from our own billing records, and a clearly labelled worked example you can redo with your own numbers.

The headline numbers

The headline is the method. These are the numbers worth tracking, and where to get each one:

Metric	How to measure it	Baseline window
Time per support ticket	Handle time from Gorgias, Zendesk or your helpdesk reports	2-4 weeks before AI rollout
Time per recurring draft	VA time log per task (a Klaviyo email, a product description)	Same 2-4 weeks
Time per data task	Log a sample of CSV reshapes or report builds	Same 2-4 weeks
Tasks completed per hour	Time tracking against your task list	Same 2-4 weeks
Rework rate	Count of outputs you send back for correction	Same 2-4 weeks
Customer-facing quality	Your own 1-5 score on a weekly sample of 10 outputs	Same 2-4 weeks
Your admin hours	Your calendar or a simple weekly log	Before the VA starts

Measure the same tasks again at 30 and 60 days after the AI tools go live. The comparison that matters is your VA against their own baseline on your own work, not an agency’s average across businesses that look nothing like yours.

The tool cost vs the productivity gain

Per VA, per month, at approximate list prices (these move, so check current pricing):

Cost	$ AUD
Claude Pro	20
ChatGPT Plus	20
1Password Teams seat	12
Loom Starter seat	9
Total stack	61

Now the worked example. It is illustrative: the hourly rate is real, the hours saved are an assumption you should replace with your own measurement.

The rate: across DotVA’s billing records from June to September 2026, clients paid on average $17.47 AUD per hour actually worked (ex GST). Your own rate depends on tier: $12-17 for admin, $18-25 for specialist, $25-35 for bookkeeping. Monthly cost is hours x rate x 4.33, so a 20-hour week at $17.47 is about $1,513 a month.

Hours the AI saves the VA each week (assumed)	Monthly value at $17.47/hr	Multiple of the $61 stack
1	$76	1.2x
3	$227	3.7x
5	$378	6.2x

Break-even is about 0.8 of a saved hour a week. That is the useful takeaway: the tool stack only has to save your VA under an hour a week to pay for itself, so the real question is not whether the tools are worth buying but whether your VA is trained to use them.

For context on the placement itself rather than the tools: the same 20 hours at a modelled local admin rate of about $37.86/hr all-in (12% super, around 1.2% workers comp and around 13% paid-leave loading on a $30/hr base) comes to roughly $3,279 a month. That comparison is modelled, not measured; the method is on our benchmark methodology page.

Where the gains come from

In our experience the time savings cluster in three buckets. We describe the mechanism for each rather than quote a percentage, because the split depends on what your VA’s week is made of.

1. Customer support speed

The drafting step shrinks. The VA reads the customer email, runs a structured prompt, edits the AI draft for tone (usually two or three sentences), and ships. Time five tickets before the tools go live and five after, on similar ticket types, and you have your own number.

The quality can lift too: support replies tend to sound more measured because the AI defaults to a calm professional tone, and the VA’s edits add the specific human warmth on top.

2. Content draft acceleration

Klaviyo flow emails, product descriptions, blog drafts, social captions. The AI gets the VA past the blank page with a usable first draft, and the VA’s time moves from writing to editing for your voice and your facts.

Before AI, the VA was either staring at a blank page (slow start) or repurposing US template content (wrong voice).

3. Data and admin productivity

CSV merges, inbox summaries, meeting note extraction, weekly reports. One-off data reshapes are where the difference is most visible, because a manual VLOOKUP exercise becomes a prompt followed by a row-count check.

Inbox triage: the VA scans the morning’s emails, the AI suggests a tag for each, and the VA confirms and processes the batch. Log how long a triage cycle takes in the baseline weeks and again at day 30.

Where AI doesn’t move the needle

These tasks are unchanged or slightly slower with AI:

Phone calls (no AI involvement on the call itself, briefing prep maybe 1 min faster)
Sensitive customer complaints (we deliberately route around AI, see Why we don’t replace VAs with AI)
First-time client onboarding tasks (novel context AI can’t carry)
Bookkeeping and ASIC/TPB compliance (regulated work, human-only)

How much of a role sits in this category varies enormously. For a phone-heavy reception role it can be most of the week; for an ecommerce support role it is a much smaller slice. That is exactly why a single company-wide multiplier would mislead you, and why the only ROI figure worth trusting is one measured on your own role.

The training cost we don’t show in the table

To get any of the gains above, every VA goes through the 5-day AI training in week 1 (see Training a new VA on AI in their first week).

The day plan in that checklist adds up to just under four hours of VA time across the week, plus the onboarding reviewer’s time checking the practice work. At $17.47/hr, four hours of VA time is about $70.

Recoverable cost: if you then measure the illustrative 3 hours a week saved from the example above, the training time is recovered in well under a fortnight of use. If you measure less, it takes longer, and the measurement tells you so.

The 60-day retention insight

The insight is that retention is the benefit most likely to be overstated, so it is the one to be most careful with.

A well-trained VA who is visibly faster on the work you care about is plausibly more likely to still be in the role at day 60 and day 90. We don’t publish a retention percentage, because a comparison like that needs a cleaner before-and-after than any agency our size can honestly claim.

What you can price is the cost of a replacement to you. DotVA never charges a recruitment fee for a replacement VA, so the real cost is the ramp-up time: a new shadow week, SOPs re-walked, and a few weeks of slower output while the new person learns your systems.

Add retention into your ROI calculation only once you have your own data on it. Until then, leave it out and treat it as upside.

The placements where AI didn’t work

Honest disclosure: some VAs never adopt AI properly, and we don’t publish a percentage for the same reason as above. Where it happens, we retrain, and if that fails we replace the VA.

The pattern: the VA either fundamentally doesn’t trust AI output and refuses to ship anything without re-doing it manually, or fundamentally over-trusts AI and ships hallucinations. Both are training failures. Both need either re-training or replacement.

The day-5 competency check exists to make that failure visible in the first week, rather than two months into the placement when you have already built habits around a VA who is quietly avoiding the tools.

What this means for your placement

If you are evaluating a VA agency in 2026:

Ask whether they train on AI in week 1. If not, walk away.
Ask which AI tools they pay for per VA. If “none, we expect you to provide them”, walk away.
Ask for their training curriculum. If it’s verbal, walk away.
Ask whether there is a documented 30-day check-in for AI adoption specifically. If the answer is vague, walk away.
Ask any agency that quotes you a productivity multiplier how it was measured. If there is no baseline, treat it as marketing.

We publish all of this so you can ask the same questions of any agency you’re talking to, including us. The method above is the bar.

What’s next

For the stack details, see The AI-augmented VA stack.

For the training plan, see Training a new VA on AI in their first week.

For pricing context, see What does a VA actually cost in Australia in 2026.

If you want a placement that comes with the training and tool stack already running, book a discovery call.