What you are buying when you pay a freelancer by their passport

Short version: we are in Bengaluru and we quote first-world prices to first-world clients. We have also been on the other side of that invoice. So this is the argument we make when a prospect asks why we do not cost a fifth of what a studio in Austin costs, written out with the receipts, because the answer turns out to be good for the client as well as for us.

The discount, measured

Start with the size of it. In 2021 the International Labour Organization surveyed about 12,000 platform workers for its report on digital labour platforms. Workers on freelance platforms averaged US$7.60 an hour. Across all online web-based platforms the average was US$3.40, and half of the workers earned under US$2.10 an hour. Workers in developing countries earned 60 percent less than workers in developed countries on freelance platforms, and that gap held after the ILO controlled for basic characteristics and the types of tasks performed. That is the same tasks at 40 cents on the dollar.

A team at the Oxford Internet Institute (Vili Lehdonvirta, Isis Hjorth, Helena Barnard and Mark Graham) went one level deeper, with the full transaction records of one of the largest platforms: 362,220 projects over six months, narrowed to 34,352 writing jobs bought by clients in the United States and Canada, because writing is commodified enough that the same job is the same job wherever it is done. The mean worker in Kenya earned $4.26 an hour, a third of the mean worker in the United States. About 35 percent of all the variation in pay rates was explained by the worker's country alone.

Then they checked whether the cheaper workers were worse. The platform records skill tests passed, projects completed and client feedback. On that index Bangladesh ranked slightly above the United States, and most of the lower-income countries ranked above the high-income ones. Holding competence equal changed the country gaps by almost nothing. What did close the gap was two things that have nothing to do with the work: the wage level in the worker's home country, and how well known that country is in the client's country, which the authors measured through search volume. Add those two and the unexplained country effect falls from 35 percent to 5.

The mechanism is not mysterious. There are more people bidding than jobs to bid on, so the price settles near the lowest rate at which each bidder can still justify the work, and that floor is set by the next-best job at home. A worker in Manila told the researchers she priced a whole day online at $30 because that matched her call centre wage. A worker in Nairobi put the other half of it in one line: "Third-world people are only offered low-skilled jobs."

The view from the cheap end

Falcon has done first-world work for a first-world client from a desk in Bengaluru at under two dollars an hour. It is a normal arrangement here rather than a hard-luck story, and that is the point. Nobody involved thought the work was worth two dollars. That was simply the rate at which the next person in line would have said yes.

The accounts from workers match the data. In a Hacker News thread in 2013 an engineer in India wrote that $10 an hour would have been dream pay, that his first programming job paid 14,000 rupees a month, and that being paid less meant people had no opportunities to begin with and took whatever was offered. In a 2020 thread a commenter laid out $4,000 a year for engineers in smaller Indian cities against $180,000 for a new graduate at Google, and noted that the same engineer's pay multiplies the moment they land in California with the same skills. A reply from Bengaluru corrected the extreme: engineers at Google India make about $50,000 against $200,000 in Mountain View, a four-times gap, and the $4,000 figure is the floor at the big IT consultancies. Both are true. The spread inside the country is as wide as the spread across the border.

An American who worked alongside one of those consultancy teams described people who quit for a raise of under a hundred dollars, because they had been hired for their price and for nothing else, and knew it. That is what a reservation wage looks like from the inside: no loyalty in either direction, because none was paid for.

The platforms do not soften this. The Fairwork Cloudwork Ratings 2025, run by researchers at Oxford and WZB Berlin, rated sixteen of the largest online work platforms. Four could show that workers consistently earn at least the local minimum wage after costs. Amazon Mechanical Turk, Fiverr, Freelancer and Upwork were not among the four. Of the workers surveyed, 31 percent had gone unpaid for completed work at some point and 38 percent had been paid late.

One more thing about the cheap end that clients rarely price in. The worker's rent is in rupees. Their tools are not. Our own infrastructure bill for this site is about $73 a month, the same $73 that Vercel and Supabase would charge in Ohio. A laptop, a phone, a Claude subscription, a domain: all dollar-priced. Wherever a living-cost discount is justified, it applies to a shrinking share of what a knowledge worker actually spends.

How much of it is cost of living

The arithmetic, since "cost of living" is the phrase every discount arrives wrapped in, goes like this.

Numbeo, with data updated this month, puts the cost of living in Bengaluru 73.2 percent below Austin, Texas, rent included: you would need about $1,820 here to live the way $6,800 lives there. The World Bank's purchasing-power conversion factor for India in 2024 was 20.45 rupees per international dollar, against a market exchange rate of 83.67 rupees to the US dollar, which says a rupee buys about four times what its exchange value suggests. Both routes land in the same place. Living costs can explain a gap of roughly three and a half to four times, for a person who spends every rupee at local prices.

So take a $50 hour, the bottom of the published range for freelance Shopify work in the United States. Living costs, generously, would justify paying about $12 to $13.40 for the same hour from Bengaluru. The offers that actually circulate sit well under that line.

Anything below the living-cost line is not rent. It is bargaining power, or the assumption the Nairobi worker described, and it helps to know which one you are buying.

The companies that pay by location and say so plainly agree with this reading. GitLab's handbook multiplies a San Francisco benchmark by a location factor, and it says the factor follows the cost of hiring in a market rather than the cost of living there, which is another way of saying what people in that place will accept. At the other pole, Basecamp has paid everyone at the 90th percentile of San Francisco rates regardless of where they live since 2017, Reddit dropped its US geographic pay zones in 2020, and Buffer removed its two lowest location bands in 2022, raising 55 of its 85 people by an average of $10,265. None of these companies is a charity. They decided that the pool of people who apply when pay does not depend on the passport was worth the money.

What the discount buys

The buyer-side accounts in the same threads are consistent, and they are not flattering to the discount.

A consultant recommended five skilled engineers and a year for a project; the CEO hired eighty offshore developers instead, on the advice of a CTO who turned out to be a fraud, and the first week's commits were copied from a public repository. Another engineer has spent much of a career on "reshoring" projects, cleaning up codebases bungled by budget teams, and in the same comment says the best developers he has worked with were Indian and paid the same as he was. A third made the mechanism explicit: outsourced projects go to the lowest bidder while hiring is held to a high bar, so the difference in outcomes is self-selection rather than geography. Pay top dollar and offshore works fine.

The subtler failure is attention. A person billing $6 an hour cannot live on one client, so they take four. The honeymoon lasts until the other three start, and then sprints stretch, the same bugs come back, and switching costs have quietly become hostage costs. Nobody cheated. The price simply bought a quarter of a person.

There is a 2026 version of this with a new name. BetterUp Labs and Stanford's Social Media Lab surveyed 1,150 US employees last September and found that 40 percent had received "workslop" in the previous month, AI-generated work that looks finished and is not, at almost two hours of cleanup per instance, about $186 a month per employee. Upwork's own hiring report for the same month shows demand for quality assurance up 9 percent and project management up 17 percent, and reads the two together as businesses paying to check work they thought they had bought. A cheap hour is now often a fast prompt and a copy button. Whoever is downstream pays for the checking.

One honest caveat, because the evidence deserves it. Paying a given person more does not make that person's work better. Winter Mason and Duncan Watts tested this on Mechanical Turk in 2009 and found that higher pay bought more output but no higher quality; workers anchored on the new rate and felt no more obliged. What price changes is who shows up and who stays. Companies that pay the same everywhere report application pools full of exceptional people from poorer countries, precisely because the door is open. The retention half is measured too. After Sam's Club raised wages, productivity rose 16 percent, turnover fell 25 percent and sales rose 25 percent over two years, in Zeynep Ton's account for Harvard Business School, and Costco turns over about 8 percent of its staff a year against about 60 percent at other retailers. Gallup puts the cost of replacing one employee at half to twice their annual salary. Replacing a freelancer who knows your store is not cheaper.

What paying properly buys

The other side of the ledger, from the same threads. A founder whose devops engineer is in India says he was the best applicant, is paid accordingly, and is not cheap. A European manager had terrible results outsourcing whole projects to India and then hired individual Indian engineers at European salaries; they are now among the top performers on his team, and his advice is to skip the peanuts. An engineer summed up the market: there are plenty of excellent developers in India, there are just none working for $10,000 a year, at least not for long.

That last clause is the whole business case. Skill at a low price is a temporary condition. The person either gets paid what the work is worth, by you or by the next client, or they stop doing the work. Paying properly is how you keep the person you liked. Generosity has little to do with it.

For our part, we price the outcome. The Foothold audit on our front page is $1,500 to $2,500 for a fixed scope, whether it is bought from Bengaluru or from Boston, and our monthly retainers are quoted the same way. A client is not buying our hours or our rent. They are buying a store that leaks less money next month, and that is worth the same amount in every time zone.

If you hire across a border

Six things we would want a client to do, having sat on both sides.

Price the outcome, not the postcode. Ask what the work is worth to your business, then negotiate from there. A fixed scope with a fixed price removes the hourly-rate theatre entirely.

If you take a living-cost discount, know its ceiling. For India it is about four times, and only on the share of someone's life that is spent in rupees. A rate that implies a ten- or twenty-times gap is not a living-cost adjustment.

Ask who does the work and what they get. If an agency bills you $60 an hour and the person at the keyboard sees $6, you have bought the layer rather than the work. The threads above are full of what the layer delivers.

Pay on time. Over a third of platform workers have been paid late and nearly a third have gone unpaid. Being the client who pays on the day is a competitive advantage that costs nothing.

Budget for the checking. If the rate is low enough that the work must be reviewed line by line, add the review to the price before you compare it with the quote from the person who does not need reviewing.

Plan to keep them. The person who knows your catalog, your theme and your ad account is worth more each month. Treat the rate as the cost of that memory persisting.

Receipts

• ILO, World Employment and Social Outlook 2021, executive summary: the hourly earnings, the median, and the 60 percent gap after controls.
• Lehdonvirta, Hjorth, Barnard and Graham, "Global earnings disparities in remote platform work: liabilities of origin?" (2021, open access): the 362,220 projects, the 35 percent, the competence index, the Manila and Nairobi interviews.
• Fairwork Cloudwork Ratings 2025: the sixteen platforms, the four, the non-payment and late-payment figures.
• Numbeo, Austin against Bangalore, read 21 September 2026.
• World Bank: PPP conversion factor and official exchange rate for India, 2024.
• GitLab's compensation handbook, Basecamp on how it pays (DHH, 2017), CNBC on Reddit's 2020 change, Buffer on location-independent salaries (2022).
• The Hacker News threads, with the accounts paraphrased above: 2013, on hiring in India for $10 an hour; 2020, should a remote salary follow location; 2022, on outsourcing; 2024, on location-based pay.
• Harvard Business Review, "AI-Generated 'Workslop' Is Destroying Productivity" (September 2025) and Upwork's hiring report for September 2025.
• Mason and Watts, "Financial incentives and the 'performance of crowds'" (2009).
• Harvard Business School on Zeynep Ton's good jobs research and Gallup on the cost of turnover.
• Our own prices are on the front page and on /services.