TL;DR. The two freelance marketplaces that have to show audited numbers are reporting the same two-year pattern from opposite ends of the market. Upwork’s active clients fell from 851,000 to 785,000 between 2023 and 2025, a drop of 7.8%, while gross services volume per active client rose from $4,867 to $5,129, up 5.4%. Fiverr’s annual active buyers fell from 4.027 million to 3.135 million, down 22.1%, while annual spend per buyer rose from $278 to $342, up 23.0%. Combined, the two listed client bases shrank by 19.6% in two years. Fiverr states plainly that buyers spending over $500 accounted for 66% of its marketplace revenue in 2025. The client count is the part that is disappearing; the depth per client is the part that is growing. A service business built on a queue of new projects is exposed to the shrinking half. A service business built on monthly contracts with a small number of clients is positioned on the growing half. The conversion is not a pricing trick. It is a specific sequence of four moves, made at a specific moment in the project.
The number nobody in the freelance advice market is looking at
Most writing about retainers argues from preference. Recurring revenue feels safer, so pursue it. That is true and useless, because it gives you no read on whether the market is moving toward or away from you.
There is a harder source, and it is the same one we used to price a completely different revenue line in creator sponsorship rates in 2026: audited filings rather than self-reported surveys. Upwork files a 10-K with the United States Securities and Exchange Commission. Fiverr files a 20-F. Both are audited. Both disclose, every year, how many clients transacted and how much each one spent. Those two series, read together, describe the actual shape of demand for independent services better than any survey of freelancers does, because nobody is being asked to self-report.
They have never been tabulated side by side. Here they are.
Verified data: client count against spend per client, two listed marketplaces, 2023-2025
| Metric | 2023 | 2024 | 2025 | Source |
|---|---|---|---|---|
| Upwork active clients (thousands) | 851 | 832 | 785 | Upwork 10-K, FY2025 |
| Upwork GSV per active client | $4,867 | $4,815 | $5,129 | Upwork 10-K, FY2025 |
| Upwork gross services volume (thousands) | $4,142,252 | $4,008,107 | $4,028,386 | Upwork 10-K, FY2025 |
| Upwork Marketplace revenue (thousands) | $586,099 | $662,108 | $682,883 | Upwork 10-K, FY2025 |
| Upwork Marketplace take rate | 15.4% | 18.0% | 18.7% | Upwork 10-K, FY2025 |
| Fiverr annual active buyers (thousands) | 4,027 | 3,630 | 3,135 | Fiverr 20-F, FY2025 |
| Fiverr annual spend per buyer | $278 | $302 | $342 | Fiverr 20-F, FY2025 |
| Fiverr marketplace GMV | not reported in this table | not reported in this table | $1,073.0m | Fiverr 20-F, FY2025 |
| Fiverr marketplace take rate | not reported in this table | 27.6% | 27.7% | Fiverr 20-F, FY2025 |
Upwork’s Marketplace take rate is Marketplace revenue divided by Marketplace gross services volume, which is a narrower denominator than total GSV; the two are not interchangeable. Fiverr’s 2023 GMV and take rate are not in the key-metrics table of the FY2025 filing and are left unstated here rather than reconstructed.
CEOtudent editorial framework: what the two filings imply together
| Derived measure | Upwork | Fiverr | What it says |
|---|---|---|---|
| Change in client count, 2023 to 2025 | -7.8% | -22.1% | The number of buyers is contracting at both ends of the market |
| Change in revenue per client, 2023 to 2025 | +5.4% | +23.0% | The buyers who remain are each worth more |
| Change in platform take rate | +330 bps (2023 to 2025) | +10 bps (2024 to 2025) | The intermediary’s cut is rising, not falling |
| Combined listed client base, 2023 to 2025 | 4.878m to 3.920m combined | -19.6% |
Derived by CEOtudent from the disclosed figures above. Percentage changes are computed from the reported values and rounded to one decimal place. The combined client base adds Upwork active clients to Fiverr annual active buyers; the two definitions differ slightly and a client active on both platforms is counted twice, so treat the combined line as a direction, not a headcount.
Two independently audited companies, different pricing models, different service tiers, different geographies of demand, and the same shape. Fewer clients. More per client. That convergence is what makes the reading trustworthy: it is not one company’s bad year.
Fiverr’s own filing names the mechanism. It attributes part of the decline in annual active buyers to AI technologies reducing demand for what it calls “simple and low-skilled services” on the marketplace, and describes a widening gap between high- and low-skilled work. The bottom of the transactional market is being absorbed by tools. What survives is work that requires continuity, context and judgement, and that work does not arrive as a stream of discrete gigs.
There is a second number in the Fiverr filing that settles the argument. Buyers who spent over $500 accounted for 66% of marketplace revenue in 2025, up from 65% in 2024. Two thirds of the revenue on a platform famous for five-dollar gigs comes from the deeper end of its own client base.
What this means if you sell your own time
Read the pattern as a CEO reads a segment report, not as a freelancer reads a job board.
Your acquisition cost is rising and your retention value is rising at the same time. Fewer buyers in the pool means each new client is harder to win. More spend per buyer means each client you keep is worth more. Both movements push in the same direction: the return on winning a client has shifted from the first transaction to the twelfth.
The platform take rate is the tax on staying transactional. Our channel comparison in Upwork, Fiverr or direct clients in 2026 works through the per-engagement arithmetic; this is the two-year trend underneath it. Upwork’s Marketplace take rate went from 15.4% to 18.7% in two years. Upwork also disclosed that from May 2025 new contracts carry a variable talent service fee of 0% to 15% depending on project type, job availability and client demand, fixed for the life of that contract, while contracts formed before May 2025 keep the flat 10% rate. The specific percentage matters less than the structure: the fee attaches to the transaction. A client relationship that lives outside the transaction stream is not subject to it.
The work AI is eating is the work that was easiest to sell as a project. Discrete, specifiable, self-contained deliverables are exactly what a model does cheaply. Work that requires knowing what happened last quarter, who objects to what, and which constraint is real is not specifiable in a brief, which is why it does not get bought as a gig. It gets bought as an ongoing arrangement.
This is the CEO-and-student split in its plainest form. The student half keeps getting better at the craft. The CEO half notices that the unit of sale has changed, and changes what it sells before the market forces the issue.
Why most retainer pitches fail
Three failure modes, all of them predictable.
Selling access instead of outcomes. “Ten hours a month of my time” prices your availability, which is the one thing the client can obtain elsewhere. It also makes every quiet month look like waste to the person signing the invoice.
Pitching at the wrong moment. The end of a project is when the client’s attention moves on. The pitch has to land while the work is still producing visible results and before the relationship goes dormant.
Asking the client to imagine the value. A retainer proposal that describes what you would do is a request for faith. A retainer proposal that describes what already happened, and what will decay if it stops, is a request for continuity.
CEOtudent editorial framework: the four-step retainer conversion ladder
This is a CEOtudent framework, not a research finding. It is offered as a decision structure, and the arithmetic below is illustrative.
| Step | When | What you do | What the client hears | Failure signal |
|---|---|---|---|---|
| 1. Instrument the project | At kickoff, before any work | Agree on two or three measures the work should move, and record the starting values | “We will know whether this worked” | Client cannot name a measure. The engagement has no owner. |
| 2. Report against the instrument | Mid-project and at delivery | Show the measures moving, in the client’s own numbers | “This produced something” | Nothing moved, or nothing was measurable. Do not proceed to step 3. |
| 3. Name the decay | At delivery, not after | State specifically what stops working without maintenance, and on what timescale | “There is a cost to letting this sit” | You cannot name a decay mechanism. The work was genuinely one-off; take the referral instead. |
| 4. Propose the smallest credible contract | Within the delivery conversation | One monthly scope, one price, one review date, one exit clause | “This is easy to say yes to and easy to leave” | You proposed a large annual commitment. Expect a stall. |
The load-bearing step is step 3. A retainer is only honest when something actually degrades without ongoing attention: a positioning that drifts as competitors move, a pipeline that empties without maintenance, a system that accumulates errors, a body of content that ages. If you cannot name the decay, you do not have a retainer, you have a subscription the client will cancel in month four when they notice nothing is happening.
Step 4 matters for a reason the marketplace data supports. Fiverr’s revenue concentration in the over-$500 cohort did not come from buyers signing large commitments; it came from buyers repeating. The goal of the first retainer month is the second retainer month.
Illustrative arithmetic: why the conversion changes the shape of a year
| Scenario | Clients per year | Average value per client | Annual revenue | New clients you must win |
|---|---|---|---|---|
| Project-only | 12 | $4,000 | $48,000 | 12 |
| Half converted | 6 projects + 3 retainers at $1,300/month | $4,000 / $15,600 | $70,800 | 6 |
| Retainer-led | 2 projects + 5 retainers at $1,300/month | $4,000 / $15,600 | $86,000 | 2 |
Illustrative CEOtudent example, not survey data. Figures are chosen to show the structure and are not claims about typical earnings. The arithmetic: 12 x $4,000 = $48,000; (6 x $4,000) + (3 x $1,300 x 12) = $24,000 + $46,800 = $70,800; (2 x $4,000) + (5 x $1,300 x 12) = $8,000 + $78,000 = $86,000.
The revenue difference is the visible part, and it is the one our analysis of how long it takes to replace a salary with independent income treats as the timeline variable. The part that matters more here is the last column. In the retainer-led row you have to win two clients in a year instead of twelve, in a market where the audited filings say the client pool shrank by roughly a fifth in two years. You are competing for far less of a contracting resource.
What the data does not say
Three honest limits.
The Upwork and Fiverr filings describe platform-mediated demand. They do not measure direct client relationships, which is precisely the channel this piece recommends, and no audited public dataset does. The inference that the same depth-over-breadth pattern holds off-platform is reasoning from the mechanism, not from a measurement.
Both companies note limits on their own metrics. Upwork’s 10-K states that it tracks active clients and GSV per active client with internal tools that are not independently verified by any third party, and that the metrics are affected by fraud, spam and fake accounts. Take the trend, not the third decimal place.
Neither filing tells you that retainers pay better than projects for any individual. They tell you that the number of buyers is falling and the spend per buyer is rising. What you do with that is a business decision, and it depends on whether your work has a genuine decay mechanism.
FAQ
Is a retainer just a discount for buying in bulk?
It should not be. Priced as bulk hours, a retainer is worse than project work: you accept lower rates in exchange for predictability, and the client accounts for it as a cost line. Priced as maintained outcomes, the comparison is not to your hourly rate but to what the client loses when the work stops.
How long should the first contract run?
Short enough that saying yes is not a decision the client has to escalate, with a named review date. The evidence that recurring revenue concentrates in repeat buyers rather than large commitments, visible in Fiverr’s over-$500 cohort accounting for 66% of marketplace revenue, points to earning renewal rather than negotiating length.
What if the client says the project is finished?
Then it may be. Step 3 of the ladder is a test, not a technique. If nothing degrades without you, the honest move is to close well and ask for an introduction. Manufacturing a decay mechanism that does not exist is how service businesses lose reputations.
Does the marketplace decline mean freelancing is shrinking?
The filings do not support that reading. They show platform-mediated buyer counts falling while spend per remaining buyer rises, and Fiverr attributes part of the decline to AI absorbing simple, low-skilled services specifically. That is a change in what gets bought through a marketplace, which is not the same as a change in total demand for independent work.
Should I leave the platforms entirely?
The take-rate data argues for reducing dependence, not for a clean break. Upwork’s Marketplace take rate rose 330 basis points in two years and its post-May-2025 talent service fee is variable by project type rather than flat. A platform is a reasonable acquisition channel and an expensive place to keep a relationship that has already proven itself.
Sources and further reading
- Upwork Inc., Annual Report on Form 10-K for the fiscal year ended 31 December 2025, filed with the United States Securities and Exchange Commission, 13 February 2026. Key financial and operational metrics; Marketplace revenue and take rate definitions; talent service fee structure.
- Fiverr International Ltd., Annual Report on Form 20-F for the fiscal year ended 31 December 2025, filed with the United States Securities and Exchange Commission, 12 March 2026. Key performance indicators; marketplace GMV, take rate and buyer cohort disclosures.
- United States Securities and Exchange Commission, EDGAR full-text filing archive, for the primary documents cited above.
Both filings are primary, audited disclosures. Every figure in the verified-data table above is quoted from them directly; every figure in the derived tables is computed from those quoted values and labelled as such.
This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.
This post is also available in:
