The short answer
Few expert data vendors publish hourly figures, and none publishes what a buyer pays. On 7 November 2026 the published references were Alignerr (Labelbox) at up to $150 an hour for experts, Prolific at an $8 minimum and $12 recommended for research participants, and Handshake AI's statement that pay varies by project.
Every buyer of expert data asks the same first question: what does an expert hour cost? The honest answer is that vendors almost never say. What a few of them do publish is what experts or participants are paid, which is a useful floor. This piece collects those figures as they stood on 7 November 2026, explains what they leave out, and shows how to use them.
Which vendors publish an hourly figure?
We read the pages of fifteen vendors that sell expert data. Three published anything about hourly pay.
| Vendor | Published figure | Who it applies to | Page |
|---|---|---|---|
| Alignerr (Labelbox) | Up to $150 an hour | Subject-matter experts in its network | alignerr.com/about |
| Prolific | $8 an hour minimum, $12 recommended | Research participants, paid by the researcher | Payment model help article |
| Handshake AI | Varies by project; shown before joining | Students and experts in its Fellowship | MOVE Program help article |
Prolific also charges researchers its own fee, separate from what participants earn; that article does not give the percentage. The other twelve vendors, including Surge AI, Scale AI, Mercor, micro1, Turing, Toloka and Snorkel AI, publish no hourly figure for experts or buyers on the pages we read. Their prices are quoted per project.
What do those figures leave out?
They are pay to the person, not price to the buyer. Between the two sit recruiting and testing, project management, quality review, tooling, the vendor’s margin and, for marketplaces, a platform fee. None of the vendors we checked publishes that markup.
They also mix very different people. Prolific’s figures are for general research participants: anyone who qualifies for a study, not a credentialed professional. Alignerr’s “up to $150” is a ceiling, and a ceiling says nothing about the average or about your field.
What about platform pricing?
Some vendors sell software and services together, and publish the software side. Labelbox meters usage in Labelbox Units (one unit for each image row in its Annotate product, twenty for each row in its Live Multi Modal/LLM product) but gives no dollar value for a unit. Kili Technology offers a free trial of $0 a month (two to five seats, a 100-asset limit), with its Grow and Enterprise plans priced by contract. Neither covers the experts themselves.
Are some projects priced per task instead?
Yes. Work built around automated verifiers, such as coding tasks and RL environments, is often priced per task. The best public figures come from Epoch AI, which interviewed people in the market in early 2026: the going rate for one task and its grader was “$200 to $2000 mostly”, five-figure tasks were unusual, and exclusivity raised the bill four- to fivefold. Epoch presents these as what interviewees said, not as contract data.
How do you turn published rates into a budget?
Start from time, not rates.
- Run a timed pilot to find real minutes per item for your task.
- Multiply by the number of items and divide by 60 for expert hours.
- Add review: spot checks add perhaps a quarter, double grading doubles it.
- Price the hours at an expert rate you choose within the published range.
- Treat the result as expert pay only; the vendor’s quote will be higher.
A worked case: 500 rubric-graded legal answers at 20 minutes each, with a second grader on half of them, comes to 250 hours. At $150 an hour that is $37,500 in expert pay before any vendor fee. Our project pricer does the same arithmetic with your numbers.
Does a higher rate buy better data?
Not by itself. A higher hourly figure usually buys a more senior or scarcer person, and that matters most for rubric design and for grading the hardest items. For high-volume ranking on clear criteria, the gain from paying more can be small. The cheapest way to find out is to split the pilot: give the same 50 items to two credential levels and compare acceptance rates against your rubric before committing the budget.
Why don’t more vendors publish rates?
We can only report what is published, not the reasons behind it. What we can say is that the published figures sit mostly on the worker side of these businesses, aimed at recruiting experts, and the buyer side is quoted project by project. For a buyer that means the comparison happens in scoping calls, which is where a clear brief earns its keep.
What should you ask for in a quote?
- The price per accepted item or per expert hour, and which one the vendor prefers.
- What the expert is paid, or at least the range, so you can see the markup.
- The review layer included in the price.
- Who pays for rework when items fail your rubric.
- What changes if you need exclusivity.
When you compare expert data companies on price, ask each one for the same unit, so that a per-item quote and a per-hour quote can be set side by side.
Published rates will change; the dated table above is the point of reference until our next check.
Sources for this piece
- Alignerr (Labelbox) about page (checked November 7, 2026)
- Prolific researcher help, payment model (dated 17 Jul 2026) (checked November 7, 2026)
- Handshake AI, Introduction to the MOVE Program (checked November 7, 2026)
- Labelbox billing documentation (checked November 7, 2026)
- Kili Technology pricing page (checked November 7, 2026)
- An FAQ on reinforcement learning environments (Epoch AI Gradient Updates, January 12, 2026)