You have about $250 a month. Not $2,500. You asked where to find cheap virtual assistants and the first four replies were variations of “you get what you pay for,” which is not an answer, it is a shrug with a moral attached. Meanwhile you still have 500 supplier rows to retype and a day job.

Straight answer, then. Cheap help works, at rates that would make those replies uncomfortable, and when it fails it fails for a reason unrelated to the rate. The cheapest hourly rate and the lowest total cost are different numbers, and the gap between them is almost entirely your own checking time. A worker at $4 an hour whose output you must inspect in full is more expensive than one at $9 an hour whose output you can sample, because your hour is worth more than either of theirs.

Which means the discipline that matters at this budget is task selection, not negotiation. Cheap help succeeds on work where checking one unit is nearly free, and fails on everything else at any price.

Where the cheap end of the market actually is

Four channels, and their floors are made of different material.

Freelance marketplaces. Upwork publishes a median virtual assistant rate of $13 an hour and a typical range of $10 to $20. Read the footnote on their own page: those figures describe freelancers in North America with over 1,000 hours billed and a 90% job success score. That is the proven end, not the cheap end. Below it sits a long tail of newer profiles bidding lower to buy their first reviews, and no platform publishes rate data on that tail, so treat any figure quoted for it as anecdote. The fee mechanism is verifiable: Upwork’s pricing page lists a freelancer service fee of 0% to 15%, set per contract and fixed once it starts, so someone bidding $4 is bidding toward a net.

In-country job boards. On OnlineJobs.ph the floor is not an hourly rate at all, it is a subscription: a free tier with three job posts and 15 applications each, Pro at $69 a month or $299 a year, Premium at $99 a month or $349 a year. The pitch is “hire direct, no salary markups or ongoing fees,” and that is accurate. It is also the whole trade, because the markup you avoid is replaced by your own screening, contracting and payment admin. The worker’s rate is whatever you agree, so there is no published floor here, only a negotiation.

Micro-task platforms. Genuinely the cheapest labour available, priced in cents per unit, and the most restricted in what it can do. Amazon Mechanical Turk publishes a requester fee of 20% of the reward, an additional 20% on tasks with ten or more assignments, 5% for the Masters qualification, and per-assignment premiums from $0.05 to $1.00, so budget reward times 1.4 on anything with ten or more assignments. Suits tagging, classification and short extraction from public sources. Suits nothing that touches your systems.

Budget agencies. Their sites advertise “from $6 an hour.” That is a vendor’s advertised price, not wage data, and the margin means your worker receives materially less. No floor is publicly verifiable, but the mechanism is: a very low blended rate means a junior assigned to you, a high client-per-assistant ratio, or both.

The rework multiplier, with arithmetic

A real recurring job: 500 supplier rows a month, six fields each, into a fixed schema.

Worker A, $4 an hour. Ten hours, so $40 of labour. First-pass accuracy 92%, meaning 40 wrong rows, and you do not know which 40. Finding them means inspecting every row: at 15 seconds a row, 125 minutes, call it 2.1 hours. They redo the 40 rows for $3.20; you re-check those, 10 minutes more. Your time for the month is about 2.3 hours, and at $70 an hour that is $161. Total: roughly $204.

Worker B, $9 an hour. Slower, 11 hours, so $99 of labour. Accuracy 99%, five wrong rows. Because that accuracy has held for three months you check a random 50-row sample rather than the lot: 12 minutes, $15 of your time. Total: roughly $114.

The worker charging 2.25 times as much costs 44% less. Not because expensive workers are better people, but because one hour of your inspection time equals roughly 18 hours of a $4 worker’s labour, so the inspection term dominates the equation. A few dollars of rate difference is noise next to whether you must inspect all of the output or a sample of it.

Run it the other way, though, because this is what the scolding replies get wrong. Give Worker A 99% accuracy and the same sampling regime and A costs $55, beating B outright. Cheap does not lose. Cheap loses when it forces full inspection. And on some tasks full inspection is nearly free: if the check is “is the source URL column filled,” two seconds a unit, inspecting all 500 costs you 17 minutes and the cheapest competent worker wins.

Which is the thesis in one line. Cheap help works on tasks where you can afford to inspect 100% of the output.

What cheap virtual assistants can and cannot be pointed at

Three properties make work cheap-compatible, and they compound rather than average. Objectively checkable output, so a second person holding only your spec reaches the same answer. Small uniform units, the same shape several hundred times. No business context required, so the answer never depends on knowing Henderson is on legacy pricing.

Passes: typing figures from supplier PDFs into a fixed sheet; deduping an export; loading posts as drafts with categories and alt text; tagging images against a closed list; renaming and filing a document folder; turning a defined list of companies into rows of name, exact title and source link.

Fails, predictably: anything where “done” requires your judgment. Replying to customers. Writing in your voice. Categorising bookkeeping entries, which is judgment wearing a spreadsheet costume. Deciding which leads matter. Anything needing your admin credentials, where the access decision is bigger than the task.

The failure mechanism is not ability. A cheap arrangement buys throughput across many buyers; it does not buy context accumulation, and context cannot be briefed cheaply. Point a $4-an-hour worker at “whatever comes up” and every task arrives contextless, so every task needs briefing, and that cost never amortises. You have bought a second job explaining your business in 20-minute instalments.

How to buy it: outcomes first, then one of three shapes

At the cheap end, hourly pricing is the worst available structure and the default everywhere. Switch to a fixed price per deliverable: $14 per 100 completed rows, a set price per transcribed audio hour, a set price per post saved as a formatted draft.

That puts the estimation risk where the information is. The worker knows their own speed and you do not. Billing hourly, you carry that risk blind and cannot tell slow-and-careful from slow-and-distracted, so you police time instead of quality. Per unit, they carry the speed risk and you carry the specification risk, which is the one you can reduce, by writing it down once. Price the unit against what it saves you: if 100 clean rows save you 90 minutes, $14 is a good buy whether that works out to $3 an hour or $11.

One caveat. Per-unit pricing pays for units that look complete, because completeness is the only property it measures. So attach the acceptance rule to the price in the same sentence: every field filled or marked unresolvable, source URL present, rework at no charge if a random sample fails. Say it before work starts, because a standard produced after delivery reads as a way to avoid paying.

Then pick one of three shapes.

• A defined recurring task list. The same six to ten tasks, same days, same acceptance rules, five hours a week in two blocks. The specification is written once and amortises across every week after, and the worker gets faster at a fixed per-unit price, quietly lowering your effective cost with no renegotiation.
• A fixed-scope project. One finite thing with an end: migrate the 400 old posts, clean the 6,000-row list, research 200 named companies. No retention problem, no cover problem, nothing to maintain. If your budget is $150 total rather than $150 a month, this is the only pattern worth considering.
• Part-time reserved hours. A fixed block, two hours every weekday morning, so you are a scheduled commitment rather than an interruption. Predictability is worth real money to a freelancer juggling five buyers, so reserved hours often price below ad-hoc ones, and availability at a known time is usually what you wanted anyway.

The pattern that fails is “10 hours a week of whatever comes up.” No specification to write, so every task is briefed from scratch and that cost recurs forever. Nothing to measure, so you cannot tell a good week from a bad one. And it demands the most context of any arrangement, which is exactly what the cheap end does not supply. “Whatever comes up” is a senior person’s job description, and buying it at a bargain rate is how people conclude that virtual assistants do not work.

The floor below which you are selecting for the wrong things

A quality-control argument, not a moral one. Rate determines which behaviours are affordable to the worker, and the ones you most need are unpaid: re-reading a batch before sending it, flagging an ambiguity instead of guessing plausibly, asking one clarifying question and losing 15 minutes to it. On a per-unit rate each of those costs the worker income. Price the work low enough and the only rational strategy left is speed plus plausible guessing, which you will read as carelessness when it was your pricing decision.

Two further selection effects sit at the bottom. You select for the newest profiles, people buying their first reviews, often fine at simple work and gone the moment they have a rating. And you select for undisclosed delegation: below the price of a careful human pass, work gets subcontracted or run through a tool unreviewed, and someone you never screened is holding your file.

There is no honest single global floor figure, because rates are set against local labour markets that differ by an order of magnitude and the published marketplace medians describe proven North American freelancers rather than the bottom. Set the floor relatively instead: pay at or above what a full-time local job in that person’s country pays per hour for the same skill, because that is the alternative they are comparing you against. Below it you are buying someone’s short-term cash problem, and their attention follows the better offer the day it arrives. The practical version: if your rate leaves the worker unable to spend a tenth of the job’s time checking their own output, you have bought unchecked work and will do the checking yourself. Put that tenth in the price. It is still cheap.

If the budget is not yet enough for ongoing help

Two one-off purchases beat a thin ongoing arrangement, and both fit inside $250.

Buy documentation as a deliverable. Not onboarding, and not an SOP you keep promising yourself. Hire someone competent at writing, six to ten hours at a fixed price, to sit on a screen-share while you do your five recurring tasks and then produce the procedures plus the acceptance rule for each. That document is the asset that makes every later cheap hire work, and unlike a person it never churns. Buy it rather than write it, because writing it down is the step that has already failed to happen for as long as this has annoyed you.

Or buy one automation. Zapier’s published free plan covers 100 tasks a month, with Professional starting at $19.99 a month. If the job is “when a form arrives, add a row to the sheet and send the templated acknowledgement,” the free tier may cover it outright, and the paid one costs less than four hours of the cheapest assistant alive. It runs at 3am, has a zero error rate on the path you tested, and never needs re-briefing. Two limits: it handles the deterministic middle and dumps every exception back on you, and automating an unstable process locks in the wrong version, so automate only what has not changed in three months.

Sorting rule for the two. Fixed shape and low volume, automate it. Fixed shape and high volume, buy it per unit from a person. No fixed shape at all and you do not have a cheap option, you have a hiring decision you are trying to solve with a discount.

A three-question test before you post the job

Answer these about your task. More predictive than rate shopping.

• Can you write the acceptance rule in one sentence, right now, before any work starts? Something like: “each row has all five fields filled or marked unresolvable, with a source URL.” If you cannot, you have no basis on which to accept or reject, so you will redo the work and call the worker the problem.
• How long does checking one unit take compared with doing one unit? Under about a fifth and cheap help is strongly positive, with full inspection affordable. Around half and the arithmetic above collapses. Equal, and it has negative value at any rate, because you are doing the work twice.
• What does one undetected error cost, and who finds it? If the honest answer is “a customer,” “the tax authority,” or “I find out in March,” you are not buying cheap help. You are buying a liability with a small invoice attached.

Three yeses and buy it today, per unit. Any no and you have two better options than negotiating harder: split the task so the judgment stays with you and the mechanical remainder passes, or accept that this work needs a proper hire and stop pricing something that was never going to work cheap.

If the work passes the test but has outgrown one freelancer, that is a volume problem rather than a budget one, and it is what AB7 Solutions runs as a service: back-office and BPO teams for data entry, list building, transcription and CMS production, data annotation and labelling with human-in-the-loop review, and small automation builds where a workflow beats a person. Send the task and the acceptance rule and we will say which of those it needs, including when the answer is “automate this and keep your money.” Call +1 321 341 7733, email ab@ab7solutions.com or director@ab7solutions.com, or see what the teams cover at www.ab7solutions.com.

Questions that come up next

I have $60 in total, not per month. What do I buy? One fixed-scope batch of the most mechanical thing on your list, priced per unit, from one person, with the acceptance rule in the brief. Do not start an hourly arrangement with $60. You will spend it on briefing and finish nothing.

Is a monthly job-board subscription worth it for one hire? Only if you compress the hiring into that month. Post once, screen one round, hire for a defined project, cancel. At $69 for a Pro month on OnlineJobs.ph the fee pays for itself if you hire inside 30 days and is dead money if the post sits there. Paying $299 for a year before you have hired anyone once is optimism, not a saving.

Can I start cheap and trade up with the same person? The best outcome available at this budget, and it has a clear trigger. When a worker’s sample checks pass three consecutive batches, raise the rate before they ask and cut your inspection further. What you are paying more for is not their hours. It is permission to stop looking.

Sources: Upwork, virtual assistant hourly rates (stated on the page as North American freelancers with 1,000+ hours and 90% job success); Upwork published pricing, freelancer service fee 0-15%; OnlineJobs.ph employer pricing; Amazon Mechanical Turk requester pricing; Zapier published pricing. Marketplace and vendor figures are platform-published marketing numbers rather than wage surveys, and are described as such in the text.