Should we hire a VA or automate the work?

"We're a team of six and the admin is eating us. Bookings come in through a form, someone copies them into the CRM, sends the confirmation, updates the schedule sheet, then chases the invoice. My cofounder wants to automate all of it. I have VA quotes from $8 to $35 an hour and that feels simpler. Which way do we go?"

Most admin jobs are two jobs wearing one name

The question assumes the job is one thing. It almost never is. Take the booking job above and write out the steps. Copying the form into the CRM is rules. Sending the confirmation is rules. Updating the sheet is rules. Noticing that the customer booked the downtown location but wrote "hoping to come to the Mississauga one!" in the notes field, and calling them to sort it out, is judgment.

We have built more than 600 workflows, and the pattern holds across almost every "should we hire or automate" call: somewhere between 60 and 80 percent of the steps are rule-following, and the rest needs a person. So the real question is not VA or automation. It is: which steps are which, and what does each one cost to cover?

The four variables that decide it

Variable	Points to automation	Points to a VA
Volume	Daily or hourly, and growing	A few times a week, flat
Variability	The last 20 runs followed one path	Every instance needs a different response
Judgment	Rules cover it: if X, then Y	Tone, exceptions, trade-offs, apologies
Error cost	Typos are expensive and humans get tired	One wrong message is expensive and silent failure is worse

Two of these deserve a closer look, because they are the ones people get wrong.

Judgment is narrower than you think

"This needs a human" usually means "nobody has written the rules down." We built a shift-allocation system for a client where staff sent unstructured requests in plain English. It looked like pure judgment work. An AI step parsed the requests into structured bookings, ownership locks killed the double bookings, and allocation went from hours to near-instant. The genuine judgment, deciding what to do when two reasonable requests truly conflicted, stayed with a person. It was a much smaller pile than anyone expected.

Error cost cuts both ways

For pure data entry, software wins on accuracy. A tired VA at hour six pastes the customer's email address into the phone field, and nobody notices until the confirmation bounces. Software never gets tired. But when software fails, it fails at scale and in silence: a renamed form field, a run history full of "required field 'email' is empty," and 80 bookings that never reached the CRM before anyone looked. A VA makes one mistake at a time and usually notices. High error cost does not pick a side. It demands alerting if you automate, and checklists if you hire.

The math, done honestly

These numbers are round and invented to show the mechanism. They are not from a client. Say the booking admin takes 10 hours a week, and 7 of those hours are rule-following.

We will use $35 an hour for the VA: a vetted, managed VA working your time zone. Yes, you can find $8 to $12 an hour offshore. That moves the break-even point, not the logic, and we cover it in the FAQ below.

Option A: VA only

10 hours times $35 is $350 a week, about $18,000 a year, every year. Add the part nobody budgets: 20 to 40 hours of your own time in the first two months writing instructions and reviewing work, and the cost of retraining when the VA moves on. And the cost scales with volume. If bookings double, the bill doubles.

Option B: automate only

The 7 stable hours are a build: form to CRM, confirmation email, sheet update, invoice nudges. A build of that shape typically runs 15 to 25 hours of work. At our flat $150 an hour, call it $3,000 one time. Software on top: a mid-tier Zapier or Make plan, tens of dollars a month, not hundreds. Maintenance: a few hours a year when an app changes something.

Here is the honest part most automation pitches skip: the other 3 hours do not disappear. Someone still calls the customer who booked the wrong location, handles the refund, untangles the invoice dispute. With no VA, that someone is you. Automate-only does not eliminate the work. It promotes you into doing the exceptions.

Option C: both

The $3,000 build covers the 7 stable hours. A VA takes the 3 judgment hours: $105 a week, about $5,500 a year. Year one totals roughly $8,500 plus software. Year two onward, roughly $5,500 plus software. Against $18,000 a year for VA-only, and against an exceptions queue with your name on it for automate-only.

Cost line	VA only	Automate only	Both
Year one	~$18,000	~$3,000 + software	~$8,500 + software
Year two onward	~$18,000	Software + light maintenance	~$5,500 + software
If volume doubles	Cost doubles	Roughly flat	Only the small human part grows
Who handles exceptions	The VA, if trained	You, by default	The VA, with time to do it well

One more number worth knowing: the build versus the VA on the same 7 hours. The VA costs $245 a week for those hours. $3,000 divided by $245 is about 12 weeks. After three months the build has paid for itself. The honest caveat: that only holds if the process is stable. If the steps will change before the break-even date, the build is premature. Let the process settle first.

The usual right answer: automate the stable 70, staff the judgment 30

This is the split we end up at on most projects, and the strongest example from our own work is hiring itself. We built a recruitment funnel that moved more than 4,000 applicants through HubSpot, Zapier, and Eventbrite, and it eliminated 85 percent of the manual recruitment work. The 15 percent that stayed human was the part that should be human: interviews, borderline candidates, the careful no. The coordinator did not lose a job. They lost the spreadsheet shuffling and kept the decisions.

That is what the VA's role becomes in option C. They own the exceptions queue. They spot-check the automation's output, because they are the one who notices "the confirmation emails stopped going out on Tuesday." They handle the conversations no workflow should touch. They become the person who knows how the system works, which makes them more valuable, not less.

The conversation with your VA, if you already have one

If a VA is already doing this work, you owe them a direct conversation before any build starts. Say it plainly: the copy-paste is going away, the exceptions and the quality checks are yours, and that is the harder, better job. Then involve them in the build. Your VA knows edge cases nobody else does, like the client who always books by email instead of the form. That knowledge shortens the build and makes it correct. The worst version of this is the VA discovering the automation when their task list goes quiet.

What we would ask you next

On a call, after the short answer, these are the questions that follow:

1. Can you write the instruction sheet for this task today? If you cannot, you are not ready to hand it to a VA or to us. Documenting it is step one on every path.
2. Of the last 20 times the task was done, how many followed the exact same steps? Eighteen or more: automate that path. Under twelve: it is judgment work that looks like admin.
3. What happened the last time it was done wrong? Who noticed, and how long did that take?
4. Is the volume growing? Automation gets cheaper per run as volume grows. A VA gets more expensive.
5. How old is the process? If it changed last month and will change again next month, do not pay anyone to lock it in yet.

Your answers tell us whether this is a build, a hire, both, or neither.

The honest part: you might not need either

If the task takes under two hours a week, do neither. Batch it on Friday afternoons and spend the money on nothing. That is the cheapest correct answer, and we give it on calls more often than you would think.

If the work lives inside one tool you already pay for, you probably do not need a consultant. HubSpot can send the confirmation and the invoice reminder natively. An afternoon with the built-in workflow templates may close the gap.

If you already have a good VA, a strong middle path is to have them document the process, then automate the two or three most repetitive steps yourself on a starter plan. Our Zapier vs Make vs n8n comparison covers which platform fits which team.

Call someone like us when the process spans three or more tools, the error cost is real money (invoices, bookings, payroll), and the volume justifies a proper build. That is the core of our automation practice. We work at a flat $150 an hour CAD, scope quoted in writing before we start, hours never expire, no retainers. And if the right answer for you is a VA and a checklist, we will say so on the first call, because it sometimes is.

FAQ

Can AI handle the judgment 30 percent instead of a VA?

Some of it. AI steps are good at parsing messy input and drafting responses. We used one to turn unstructured shift requests into structured bookings, but we kept ownership locks and a human escalation path around it. Use AI to shrink the judgment pile, not to remove the person who owns the outcome. Anything involving money, legal exposure, or an upset customer keeps a human.

Is a $10 an hour offshore VA a better deal than automating?

A lower rate moves the break-even point, not the logic. At $10 an hour, our example build pays back in about ten months instead of twelve weeks. The costs that do not change with the rate are training time, turnover and retraining, and error rates on repetitive work, because tired humans make typos at any wage. If the process is stable and runs daily, automating the stable part still wins. If the process is young and changing, the cheap VA buys you time to let it settle, and that is a legitimate strategy.

How do we tell our VA we are automating part of their job?

Early and plainly. The honest frame: the repetitive steps are going away, the exceptions and quality checks become theirs, and that is a more valuable role. Then bring them into the build, because they know edge cases nobody else does and their notes make the build faster and more correct. The version to avoid is the VA finding out when their task list goes quiet.

What should we never automate?

Anything where the steps genuinely change every time. Anything less than about three months old, because you would be paying to lock in a process that is still moving. And any message where the recipient would feel cheated to learn no human wrote it: condolences, apologies for serious mistakes, negotiations. Also, never automate a process you cannot describe. Automating a mess gives you a faster mess.