Tweaky’s founders priced a website fix at $25 and kept half. Then they did the sums on what growth would take. Designers who code are running the same sums right now, with fewer people watching.
In the summer of 2012, Ned Dwyer put a price on a website fix: $25. Add a shopping-cart button. Swap the logo on a homepage. He and co-founder Pete Murray called each job a “tweak,” named the company after it, and kept about half of every sale. By January 2013, after working through the economics, the price was $39. At the 50% cut reported that month, the developer doing the work was left with $19.50. The company grew for 18 months, reached profitability, and was sold to GoDaddy in 2015. Even so, its founders have said they always knew the model would not scale.
So what broke? And what does it mean if you are a designer who can also write the CSS?
In this piece
- What a “quick fix” leaves in your pocket, with the arithmetic on screen
- Three real businesses built on small fixes, and what each changed to survive
- Where 2026 demand is growing, and the part of it that is a trap
- A checklist you can use within 24 hours
The sums behind the tweak
Dwyer described the pricing problem to SmartCompany, and the arithmetic is worth redoing, because one number in that interview does not hold up.
Derivation 1: what $5M meant at $25
- Inputs: $25 per tweak, the original price; a $5 million revenue goal (SmartCompany).
- Method: $5,000,000 ÷ 52 weeks = $96,154 a week. $96,154 ÷ $25 = 3,846 tweaks a week, about 770 every working day.
- Result: roughly 3,850 tweaks a week.
- Flag: the interview prints Dwyer as saying 40,000 a week. At $25, that is $52 million a year, ten times his own target. I used the arithmetic, not the quote. It may be a transcription slip or a different definition of revenue; the text does not say.
Derivation 2: what the developer kept
- Inputs: $39 per tweak; a platform share of 50%, which Dwyer called “variable” (TNW, January 2013).
- Method: $39 × 0.50 = $19.50 per tweak. At 15 minutes of work, that is $78 an hour. At 30 minutes, $39. At 60 minutes, $19.50.
- Result: the model paid a developer well only if tweaks were genuinely tiny.
Tweaky was also picky. At launch, 90 of the 1,200 to 1,500 developers who had applied had passed vetting: 6.0% to 7.5% (same TNW report). Half the price gone, a fixed price set by the platform, a tough screen to get in: the squeeze sat on the person doing the work.
A fixed price moves the risk of “quick” onto the person doing the fixing.
Same dispute, 2026, different tool
Todd Van Linda is a full-time illustrator in Florida, not a CSS person, but his story is the same argument. A client offered about $500 to repair 13 to 15 AI-generated illustrations for a children’s book, expecting each to take about 15 minutes. He said no. His rate is $65 an hour (Search Engine Journal, summarizing the Guardian).
Derivation 3: what the offer implied
- Inputs: about $500; 13 to 15 images; the client’s 15 minutes each; Van Linda’s $65 an hour.
- Method (client’s timing): 13 × 15 min = 3.25 hours, so $500 ÷ 3.25 = $154 an hour. 15 × 15 min = 3.75 hours, so $500 ÷ 3.75 = $133 an hour.
- Method (his floor): $500 ÷ $65 = 7.69 hours. 7.69 hours ÷ 15 images = 30.8 minutes each; ÷ 13 images = 35.5 minutes each.
- Result: on the client’s own timing, the offer paid $133 to $154 an hour, above his rate. He declined anyway, which suggests he expected each image to take longer than 31 to 36 minutes. That reading is mine; the report does not give his reasoning.
The fight was about minutes, not rate. A second freelancer in the same report, multimedia editor Nathan McConnell, spent two to three hours per card on a roughly 100-image AI-generated tarot deck.
A cleanup quote assumes the job is small. The file decides.
What the platforms keep
The cut is the visible part, so here it is, applied to the $39 price Tweaky settled on.
| Channel | What it takes | Keep on a $39 job | Confidence |
|---|---|---|---|
| Fiverr | 20% of seller earnings | $31.20 | High; consistent across Fiverr’s terms as captured by ConductAtlas and Upwork’s comparison page |
| Upwork | Variable 0% to 15% per contract since May 1, 2025; most freelancers report about 10% | $33.15 to $39.00; $35.10 at 10% | Medium; secondary summaries of Upwork’s docs (Jobbers) |
| Freelancer.com | Greater of 10% or $5 on standard jobs | $34.00 (the $5 fee is 12.8%) | Medium; quoted on a rival’s page, so check Freelancer.com’s own fee page |
| Codeable | 17.5% added on top of the expert’s estimate; recommended $80 to $120 an hour | n/a at $39; for one hour the client pays $94 to $141 | High; Codeable’s help center and pricing page |
Upwork also charges for proposals: $0.15 per Connect, typically 4 to 16 per proposal according to one fee guide, so $0.60 to $2.40. On a $39 job, that is 1.5% to 6.2%, paid whether you win or not.
Take-home on a $39 fix (before tax and payout fees)
Direct client $39.00 ████████████████████
Upwork, 0% fee $39.00 ████████████████████
Upwork, 10% fee $35.10 ██████████████████
Freelancer.com $34.00 █████████████████
Upwork, 15% fee $33.15 █████████████████
Fiverr, 20% fee $31.20 ████████████████
Tweaky developer 2013 $19.50 ██████████
Source: my arithmetic from the fee schedules in the table above.The bigger gap is not between platforms. In 2021, WordPress consultant Chris Lema wrote that a $100 minimum to change a button label is out of proportion, and that this is exactly why people end up on Fiverr, where $5 or $10 tweaks are on offer. That is a 10x to 20x spread ($100 ÷ $10, $100 ÷ $5).
Platform fees are visible. Scope risk is the cut nobody itemizes.
Three ways out of the one-off trap
Subscription: WP Curve. It sold “unlimited small fixes” from $79 a month, and GoDaddy bought it in December 2016, per Post Status. That site’s writer saw it as closer to an acquihire or lifeboat than a victory lap, so read the exit with care. Plans also suit people who need fixes every month and strand people who need one, as Lema notes.
Packages: Elto, the renamed Tweaky. After the move to the US, it kept $39 tweaks but added packages such as $349 to grow an email list (SmartCompany). Its investor, Blackbird, credits it with 30,000 small-business customers (Blackbird), which is an investor’s own claim. Dwyer’s original logic was that many tweaks would add up to projects, and that is what would pay back the cost of finding customers.
Owning the inbound: Okupter. Justin, a solo Svelte developer, wrote on Indie Hackers in October 2023 that he had reached $6,000 a month for the second month running. All of it is self-reported. He copied a fixed-scope, two-week-sprint model; his blog (three to four posts a month, about 15,000 visitors a month) took over five months to produce a first paid client. He sells sprints, not CSS tweaks, so treat it as a path, not a price list.
All three changed the unit they sold: a plan, a package, a sprint. My reading of why: a buyer who needs one tweak never repays what it costs to find them.
Where 2026 demand is growing, and the catch
The Guardian, via Search Engine Journal, reports that Freelancer.com listings for correcting AI output rose 87% to 10,760 between August 2025 and June 2026. Upwork saw a 70% year-over-year rise in AI-remediation gigs, and Fiverr saw searches for “AI cleanup” grow more than twentyfold from 2023 to 2026. Those figures count listings, gigs, and searches, so do not add them or compare them. Freelancer.com’s biggest category was graphic design, not front-end code. A separate Fiverr analysis for KUOW found WordPress error-fix jobs up 712% from October 2024 to March 2025; that is platform-supplied data.
The headwind is documented too. A peer-reviewed study in Management Science found postings for writing and coding jobs most exposed to automation fell 21% against less exposed jobs after ChatGPT launched, and image-creation postings fell 17% after image generators arrived. Simple, well-specified fixes are what an assistant does best. What is left for people is the messy part: working out what someone else’s generated layout is actually doing. Freelancer.com’s CEO told the Guardian that cleanup can eat the time a draft saved.
Price the outcome, not the minutes.
Do this in the next 24 hours
- Write down five fixes you have really done, with the minutes each took. Not what you think they take.
- Set a floor. Price = target hourly rate ÷ (1 − platform fee). At $65 an hour, that is $81.25 on Fiverr (20%) or $72.22 on Upwork at 10%.
- Define one fix in writing: one component, one page, one breakpoint. List what is excluded, such as cross-browser bugs, plugin conflicts, and headless front ends.
- Add a look-first clause. A short paid triage, then a re-quote. Van Linda and McConnell both show why.
- Offer a second tier for repeat clients: a monthly block of fixes.
- Publish one before-and-after on your own site. Okupter’s slow blog is the only inbound channel in this piece that you can fully control.
Where this breaks down
- Most numbers for what small fixes pay are self-reported or supplied by platforms. I found no independent dataset on UI micro-jobs net of unpaid scoping.
- The Tweaky figures are from 2012 and 2013. Fee schedules and AI tools have both moved since.
- The 2026 cleanup statistics measure different things, and none states a price per job.
- Cleanup demand spans design, video, and writing. How much of it is front-end work is unknown.
- Acquired companies get written about. The ones that quietly shut down mostly do not.
The short version
That answers the question I opened with. Tweaky’s skill was never the problem; its unit was. A one-off $25 or $39 job could not carry what it cost to find the customer, and the person doing the work absorbed the rest. If you sell UI fixes and CSS tweaks, the same sums apply to you, with a 2026 twist: AI-damaged layouts are real work, and clients still quote them as if they take fifteen minutes.
What did your last “quick fix” actually take, and what did you charge? Put both numbers in the comments.