You have a shortlist. Two agencies somebody vouched for, a freelancer profile with 400 hours logged, and a Google Doc called “VA tasks” with four bullets on it: email, calendar, research, misc. Finding the person was supposed to be the hard part. It turns out to be the document.

That doc is why most first VA hires die around week six. The owner hands over a pile of tasks, the assistant does them at 70% of the owner’s standard, the owner starts checking everything, checking takes longer than doing, and three months later the verdict is “VAs don’t work for my business.” The assistant was usually fine. The brief was not.

What a virtual assistant should actually do is not decided by a job description you copy from a blog. It is decided by two weeks of logging your own time, sorting that log by value and by drain, and handing over whole processes rather than loose tasks. A task is “book the flight.” A process is “travel, from the moment a trip is confirmed to the moment the receipts are in the accounting folder.” Delegating tasks makes you a dispatcher. Delegating processes gives you the hours back.

Two weeks of logging beats any job description template

Start here, because your memory of your own week is not reliable evidence. Research in the Bureau of Labor Statistics Monthly Labor Review in 2011 compared what people said they worked against their time diaries: estimates averaged 39.5 hours against 36.3 diary hours, and the bigger the estimate, the bigger the overestimate. People reporting very long weeks are furthest off. You are probably one of them, and the tasks you believe are eating your week may not be the ones doing it.

So log it. Fifteen-minute blocks, phone timer or spreadsheet, every working hour including the ten minutes lost reformatting a proposal. Two weeks rather than one, because week one is always atypical and you know it.

Then sort. The framing that circulates in business books puts work in four buckets by what an hour of it is worth: $10 work (data entry, chasing invoices, reformatting, scheduling), $100 work (writing a proposal, running a hiring call, fixing a process), $1,000 work (selling, pricing, key relationships, hiring decisions), and $10,000 work (the two or three calls a year that change the business). It is not precise and does not need to be. Its job is to make you notice that Tuesday morning went on $10 work while the $1,000 work slid to Friday.

Add a second column the buckets miss: energy. Mark each block as drains me, neutral, or feeds me. A 40-minute weekly reconciliation you dread costs you more than 40 minutes, because it poisons the three hours after it.

Then the uncomfortable pass. Mark the $10 and $100 rows you keep for reasons that are not business reasons. Everyone has them: the inbox you let nobody touch because answering fast is part of how you see yourself, the invoicing you do personally because it is the moment you feel like a real company. Write them down. Those are the ones you take back in week three and then blame the assistant for, and naming them now is the only defence I know of.

What you end up with is concrete: how many hours a week are recoverable, which processes they cluster into, and which two or three are high-volume and low-judgment. That is your first hire’s job. Not a wish list. A shortlist out of your own data.

What a general virtual assistant does well, and what goes wrong

Generalists are good at one shape of work: repeatable, rule-based, verifiable, low-consequence when it goes wrong once. What fits:

• Inbox triage and calendar. Sorting, labelling, drafting replies for approval, protecting blocks, ending scheduling ping-pong, briefing you before each external meeting.
• Research and list building. Eighty companies matching a profile, contacts pulled, a competitor’s pricing page checked monthly.
• Data entry and CRM hygiene. Deduping, chasing missing fields, logging activity, keeping the pipeline honest.
• Travel, end to end. Options, booking, calendar entries, confirmations, itinerary, receipts filed.
• Invoice chasing. Reminder on day 31, second on day 45, escalation to you on day 60 with the history attached.
• Social scheduling and transcription. Loading approved content, comment triage with an escalation rule, meeting notes turned into actions, SOPs for everything they learn.
• First-pass customer replies. The 60% of enquiries with a known answer, from approved templates, anything unusual flagged.

Now the list that gets people hurt. These look delegable, and are not, at least not to a generalist:

• Bookkeeping. Categorising transactions in Xero is not data entry. A generalist miscoding a year of expenses creates a mess your accountant bills you to unpick.
• Ads management. Someone following a checklist can also spend the budget on a broad-match keyword with no negatives attached. Money that moves needs a specialist.
• Design and copywriting in your voice. A VA with Canva access produces something adequate that slowly makes your brand look adequate. Drafting from your bullets is fine. Publishing in your name unreviewed is how a VA post embarrasses you.
• Anything requiring judgment about a specific client. Discounts, deadline renegotiations, complaints. This becomes delegable later, to a person you have come to trust, never to the role.

The dividing line is not difficulty. It is whether a wrong answer is recoverable within a day, which is a better test than any skills list.

A VA, an executive assistant and a specialist are three different hires

People use these three words interchangeably and then get surprised by the price. A general virtual assistant executes defined processes. You supply the process, they supply the hours and the reliability. Offshore direct hires sit at the bottom of the market; managed services publish numbers you can check. Time etc lists $39 an hour at 10 hours a month, falling to $36 at 60. Prialto publishes $1,600 a month for a 55-hour unit plus a $250 setup fee, near $29 an hour, with an engagement manager and backup cover included.

An executive assistant is a different job. They hold context, anticipate, and make small decisions for you without asking. Not a more experienced VA, a different skill, and the market prices it that way: US Bureau of Labor Statistics figures for May 2025 put the median wage for secretaries and administrative assistants at $48,310 a year, and for executive secretaries at $76,590. Roughly a 59% premium for the judgment layer.

A specialist offshore role is where most of the value hides once your audit is done. A dedicated bookkeeper, a CRM operations person, a medical biller. Narrower, deeper, and usually cheaper per unit of output than a generalist attempting the same work, because they are not learning on your time. As a domestic reference point, BLS puts bookkeeping and accounting clerks at a median of $50,670 in May 2025.

My opinion, held firmly: if the audit shows one dominant process, hire the specialist for it rather than a generalist who will eventually attempt it anyway. If it shows twelve small ones, hire the generalist.

From doing it yourself to not checking it at all

Every process you hand over moves through five stages, in order. Skipping one is what people mean when they say delegation didn’t work.

Do it. You still do the task, but now you narrate it. Record your screen, noting every point where you chose between two options and why.

Document it. Turn the recording into an SOP. Write it yourself for the first two or three. After that the assistant writes it from your recording and you review once.

Delegate it. They run it while you watch the output, not the method. Resist correcting anything cosmetic for a fortnight. If the invoice went out right, leave the subject line alone.

Verify it. Sample rather than review everything. Three of ten items, then two of twenty, then spot checks. Write the rate down or it silently reverts to all of them.

Stop verifying. The stage nobody reaches. Set the condition in advance: after four consecutive weeks with no material error, the process is theirs and you see only the monthly numbers. Without a written stopping rule the checking never ends, the assistant never owns anything, and you have paid for help while keeping the mental load.

Gallup’s entrepreneur research is worth a mention, with the caveat that it is correlational rather than experimental. Across 143 CEOs from the 2014 Inc. 500 list, those scoring high on Delegator talent had generated 33% greater revenue in 2013 than those scoring low.

An SOP a stranger could follow

Most SOPs fail because they document the happy path only. The assistant hits the first unexpected case and either freezes or improvises, and improvisation is what you were trying to avoid. A usable SOP has seven sections and fits on a page.

• Trigger. “Every weekday at 9am ET,” or “an email arrives at billing@ with an attachment.”
• Inputs and access. Systems, logins by vault entry name rather than the credential itself, files, templates, who to ask.
• Steps. Numbered, one action each, imperative, with screenshots for anything visual. If a step hides a choice, that choice belongs in decision rules.
• Decision rules. The if-then table. “Under $500 and under 30 days late, send reminder A. Over $2,000 or over 60 days, stop and flag to me.” This section is the actual value of the document.
• When something is unexpected. A written default. Mine: do not guess, do not skip silently, park the item in the Blocked column with one line on what is unclear, carry on with the rest. That one instruction prevents most of the damage a new assistant can do.
• Done means. A finish condition someone else could check. “All 12 invoices logged, status column filled, nothing left in the Pending view.”
• Escalate when. Named situations, plus how fast and by which channel. Client anger, anything legal, money above a threshold, anything a customer asks twice.

Access, credentials and where the work lives

Three decisions on day one, all of them annoying to fix later.

Credentials go in a shared vault, never in a message. 1Password, Bitwarden or similar, on a business plan, with a vault you own and can revoke in one click. Passwords pasted into chat live forever in a history you do not control. Where the vault supports it, share access without exposing the password at all.

Delegated mailbox access, not your login. Google Workspace supports mail delegation, so your assistant reads and drafts from their own account with their own audit trail, and Microsoft 365 has the equivalent. Handing over your password means no real two-factor authentication, no record of who did what, and an offboarding that becomes thirty password resets. Better still, a mailbox they own outright: enquiries@, or billing@.

Work lives in a task system, not WhatsApp. Asana, ClickUp, Trello, Notion, any of them. What matters is that every task has an owner, a due date and a status, and that “what happened with the Henderson quote” is answerable without scrolling a chat. Run delegation through WhatsApp and you stay the only index of what exists, which is the exact load you were trying to shed.

The first month, in order of blast radius

Sequence the handover by what a mistake costs, not by what annoys you most. The instinct is to offload the worst job on day one, and that job is usually the one where errors reach customers.

• Week 1: internal and reversible. One documented process. Calendar, file organisation, CRM cleanup, a research list. Nothing a customer sees, and they write their own SOP for it.
• Week 2: internal, higher volume. A second process of similar risk. Inbox triage in read-and-sort mode, with drafts they do not send. Reading those drafts tells you whether their judgment is moving.
• Week 3: customer-facing, with a net. Their drafts go out, template-answerable ones only, and you see everything for a week. Add a process with a deadline attached, like the invoice chase, so they learn that dates are real here.
• Week 4: one thing they own. Hand one process over completely, decision rules included, and look at output only. Say out loud that they own it, because assistants often wait for permission nobody thought to give.

How to tell in 60 days whether this is working

Four measurements. The first is the one people skip.

Hours reclaimed, and what replaced them. Redo a one-week log at day 60. If eight hours came back and filled with different $10 work, the hire is technically succeeding and strategically pointless. The return sits entirely in what fills the gap.

Error rate, on a defined process. Material errors per hundred items: anything needing rework, an apology, or money. Expect weeks one to three to look bad. You want a downward line, not a low starting point.

Escalation frequency and shape. Escalations should fall over six weeks, then flatten, never hit zero. Zero is not mastery, it means someone is guessing. If the same question returns weekly, that is an SOP defect, not a person defect.

Whether you have stopped checking. If at day 60 you still review every item on the week-one process, the problem is upstream of the assistant.

Three failure modes, stated plainly. Hiring before anything is documented: you train by interruption, which costs more hours than the work did, and the hire fails on your inputs. Part-time money, full-time expectations: 20 paid hours, same-day responses expected across twelve, from someone with three other clients whose expectations are clearer than yours. The owner who cannot stop: still doing the task, quietly, on Sunday. If that is you, pick one process and delete your own access to it. Usually nothing bad happens, which is the point.

The audit often reveals one more thing: some of what you listed for a human should not go to a human at all. Recurring data movement between systems, a report rebuilt from the same three sources every Monday, routine CRM updates. If your log is full of that, the answer may be an automation plus a smaller VA role. AB7 Solutions works both sides of that split, placing remote professionals and back-office teams and building CRM and workflow automation for the parts that should not need a person. For a second opinion on your time audit before you hire against it, call +1 321 341 7733, email ab@ab7solutions.com or director@ab7solutions.com, or see the service list at www.ab7solutions.com.

Questions that come up next

How many hours should I start with? Halve the recoverable hours from your audit for month one, because documenting and reviewing eats the difference. Say up front that it grows once processes are running.

What if I cannot describe my own process? Then you do not have a process, you have a habit. Record yourself doing it three times. The variations between those recordings are the decision rules you never wrote down, and they are the reason nobody else can do it.

How long before it actually saves me time? Month one costs you time. Month two is roughly break-even. The gain lands in month three, and only for processes that reached the stop-verifying rung. Anything still checked item by item at day 90 was either documented badly or should never have been delegated.

Sources: Robinson et al., “The overestimated workweek revisited,” Monthly Labor Review, June 2011 (ATUS 2003-07); Gallup, “Delegating: A Huge Management Challenge for Entrepreneurs” (143 Inc. 500 CEOs, 2014); BLS Occupational Outlook Handbook, secretaries and administrative assistants (May 2025); BLS Occupational Outlook Handbook, bookkeeping, accounting and auditing clerks (May 2025); Time etc published pricing; Prialto published pricing.