7 points worth checking on virtual assistant agency startup

What to take away

Seven checks decide whether a virtual assistant agency startup survives its first year, and all seven can be run on paper before you take a client.

Tool prices quoted below are typical published list prices as of 2026. Wages and retainer rates move by market, so look them up in the sources named here rather than copying a number.

What you are actually selling

A retainer is a promise of availability. The client is not buying a finished object; they are buying the certainty that on Tuesday morning someone who already knows their inbox rules will be at it.

Your cost lands whether or not the client sends work, so idle reserved time is inventory spoiling on the shelf. The quiet client is the dangerous one: three weeks of nothing, then the whole month at once.

Write that sentence on the wall before you write a business plan. Most operating problems in this trade are versions of it.

The seven checks in brief

Check 1: which of the four delivery models you are actually staffing

Overflow capacity is not free. Upwork charges freelancers a flat 10% service fee, and Fiverr takes 20% of each order from the seller, so the marketplace rate is not the rate you pay.

Most agencies start in the first row and intend to reach the third. The move is not gradual. It happens on the day you first assign work to somebody the client has never spoken to.

Documentation and quality checks have to exist before that day, not be built afterward.

Choose the row you are actually staffing for. An agency that sells dedicated assignment and delivers from a pool is selling something it does not have, and clients notice within a quarter.

Check 2: size the bench from three of your own numbers

You do not need market data for this. You need three of your own numbers, and you can estimate all three on paper before you hire anyone.

Assistants needed is roughly the number of clients multiplied by R multiplied by U, divided by D. Run it twice: once with U at a low usage assumption and once with U near one. The second is what happens when three clients all get busy in the same week.

The gap between those two answers is your exposure. Cover it with a trained reserve, a stated turnaround window that lets work queue, or a contractual cap on how much of an unused month can be carried forward. Pick one deliberately. Agencies that pick none discover the answer during their first bad week.

D is the number new owners get wrong. Delivered hours are not scheduled hours. Every handover, every clarification, every rewritten instruction comes out of D, and in the first months of an assignment it comes out heavily.

Measure both numbers instead of estimating them forever. Toggl Track has a free tier and paid plans from about $9 per user per month. Harvest starts around $12 per user per month.

Checks 3 to 5: registrations, worker status and client data

Some of this takes an afternoon. Some of it changes what you are allowed to sell, so it belongs before the first invoice rather than after it.

The SBA business guide is a reasonable map of the general layers: planning, launch, management, growth. Use it to separate the parts you can settle yourself from the parts that need a professional who knows your state and your client mix.

Check 6: cost one delivered hour before you price it

Your price floor is not what competitors charge. It is what an hour of delivered capacity costs you, and you can derive it.

Take the fully loaded monthly cost of one assistant: pay, employer costs, tools issued, their share of supervision, and recruiting cost spread across expected tenure. Divide by D from the sizing step. That quotient is the cost of one delivered hour.

Everything above it pays for sales, admin, bad debt, unused reserved capacity and profit.

BLS occupational employment and wage statistics tables publish employment and wage estimates by occupation, industry, state and metropolitan area. Read them as a description of what the local labor market pays for adjacent office and administrative roles, not as a rate card. Your actual cost is what you must offer to hire and keep the people you need.

QuickBooks Online's entry plan, Simple Start, lists at about $30 a month and holds assistant cost against billed hours for the same period. That comparison is the number you reprice from.

Do this arithmetic before you look at any competitor's published price. Once you have seen a number, it anchors you, and an anchored owner quotes below cost with confidence.

Check 7: a staged opening that fails cheaply

Each stage tests one thing while the others are held still, so a failure stays small.

Stage one: one client, one service, delivered by you. You are testing whether the scope you wrote is the scope the client wanted. Log every task that arrives outside the written scope. That log is your real service definition.

Stage two: same service, three clients, still delivered by you. You are testing the arithmetic. Compare reserved hours to delivered hours. If U is far off your estimate, your pricing is wrong, not your effort.

Stage three: first assistant, one client transferred. You are testing documentation. Anything the assistant has to ask you is a gap in the written process. Write the answer down instead of answering it twice. Notion's free plan holds those pages, and Plus runs about $10 per user per month.

Stage four: second assistant, coverage overlap. You are testing whether the bench works when someone is away. Take a week off deliberately. What breaks in that week is your actual operating risk.

Calendly's free tier books one service line. Paid seats start around $10 a month when you need to route between people, and HubSpot's free CRM tier holds client records and scope notes until you outgrow it.

Do not run two stages at once. An agency that adds a client, a service line and a hire in the same month cannot tell which of the three caused the trouble that follows.

The monthly, quarterly and yearly reviews

Three habits keep the model honest once it is running.

Reconcile reserved against delivered hours every month, per client, and look at the outliers rather than the average. A client using far less than they reserve will eventually ask why they are paying. A client using far more is being subsidized by the quiet ones.

Review scope drift quarterly. Retainers do not usually fail on price. They fail because the work slowly became something else and nobody repriced it.

Re-read your own written scope once a year as though you were the client. If it does not say what happens when they send more work than they reserved, it will be settled in an argument instead.

Common questions

Do I need clients before I hire, or people before I sell?

Sell first, with a start date far enough out to recruit into. Hiring against hoped-for demand is how new agencies spend their working capital on idle time.

Should I specialize or stay general?

Specialize on the work, not on the client's industry. A narrow function is easier to document, easier to train, and easier to price. A narrow industry with a broad function gives you none of that.

Can I run this with contractors only?

That is a question about the actual working relationship, not about what you would prefer. Read the Labor Department page named above, then get advice on your specific arrangement before you build a bench on the assumption.

How long before it supports me?

Nobody can tell you that from a page, because it depends on your price, your delivered hours and how fast you fill the bench. What you can do is model it: your own arithmetic above, run monthly, will tell you sooner and more honestly than any published benchmark.