# Appen vs Toloka: How They Compare for AI Training Data (2026) Source: https://sellyouraidata.com/compare/appen-vs-toloka/ Updated: 2026-10-20 > Appen is an ASX-listed data company founded in Sydney in 1996 that reported US$230.8 million of revenue for 2025 and runs a catalog of licensable off-the-shelf datasets. Toloka is an Amsterdam-based AI data company founded in 2014, majority-owned economically by Nebius, that raised $72 million led by Bezos Expeditions in May 2025. Both rely on large contractor crowds, and neither is a confirmed buyer of existing business datasets. ## Key takeaways - Appen is public and its numbers are filed with the ASX: US$230.8 million revenue in 2025 and US$119.9 million in the first half of 2026, with China now the larger segment. - Toloka is private. Its only disclosed financing is a $72 million round in May 2025 led by Bezos Expeditions, and it publishes no revenue figures. - Appen lists a catalog of licensable off-the-shelf datasets, some marked as sourced from a third-party vendor, which makes it the more relevant of the two for anyone asking whether datasets are ever bought and resold. - Both have faced scrutiny over contractor pay and transparency: Appen over rater pay, rate cuts and reported military-related contracts, Toloka over reporting on Russian facial recognition clients and low task pay. Both disputed or answered some of this, and several claims are unadjudicated. - Neither company says in public materials that it buys proprietary business data from companies. ## At a glance | | Appen | Toloka | |---|---|---| | What they do | Training data for speech, text, image, video and agentic AI; annotation platform; licensable off-the-shelf datasets[^1][^2] | Custom datasets, expert evaluation, red teaming and benchmarks; self-serve data collection[^3][^4] | | Who they serve | Large language model builders (Appen says 80 percent of leading LLM builders), plus technology, automotive, financial services, retail, healthcare and government customers[^1][^5] | Frontier AI developers; Amazon, Anthropic, Microsoft, poolside, Recraft and Shopify were named in 2025[^6] | | How they source people | Crowd of 1M+ contractors (self-reported), via the CrowdGen platform[^1][^5][^7] | Data workers in 100+ countries and 40+ languages; Mindrift for domain experts[^3][^8] | | Pay model (as published) | Project-based; rates tied to local markets, "above minimum wage"[^9] | Per task on the crowd platform; Mindrift cites about $30 to $90 per hour[^8] | | Founded | 1996, Sydney[^1] | 2014; Amsterdam headquarters[^3] | | Listing and ownership | ASX: APX[^5] | Private; Nebius holds a majority economic stake without majority voting control[^6] | | Financials (cited) | 2025 revenue US$230.8M; H1 2026 revenue US$119.9M[^10][^5] | $72M round led by Bezos Expeditions (May 2025); no revenue disclosed[^11] | | Headquarters | Sydney and Kirkland, Washington[^1] | Amsterdam[^3] | Figures are as of October 2026. Appen amounts are in US dollars unless noted. > **For data sellers:** Of the two, Appen is the one that publicly lists licensable off-the-shelf datasets, and one filtered view of its dataset site shows the source as a "third-party vendor," so it sits closest to a data marketplace model.[^2] That does not mean it buys proprietary business data from companies, and no source we reviewed says so. Toloka's pages describe commissioning newly collected data and recruiting experts.[^3][^4] Both are mainly services vendors that recruit contractors. See [where to sell](/resources/where-to-sell/) and [the buyers directory](/buyers/). Whichever buyer you are weighing, do not let one buyer set your price: get competing offers. See [selling direct vs using a broker](/guides/selling-direct-vs-broker/), and if a buyer is already talking to you, [we review offers free](/offer-review/). ## Why this comparison matters Appen and Toloka are two of the longest-running names in AI data work. Both built large crowds of contributors who label, transcribe, rate and collect data for technology companies, and both have had to adapt as demand moved from simple labeling toward expert-written data, evaluations and agent training. They are structured very differently. Appen is a public company with half-year and annual reports filed with the ASX, so there is an unusually detailed paper trail of its revenue, losses and customer concentration. Toloka is a private company that has disclosed one funding round and little else. That asymmetry shapes this comparison: we can say a great deal about Appen's finances and relatively little about Toloka's. For a business owner thinking about the AI data market, both companies are useful as case studies in how this work is priced and sourced. They are less clearly useful as destinations for your own data, a point covered in the data-seller sections below. ## Appen ### History and founding Appen was founded in 1996 in Sydney, and its own account of its history starts with speech and language data for early natural language systems and moves through search relevance, image annotation, and in recent years data for large language models and agents.[^1] It trades on the Australian Securities Exchange under the code APX.[^5] Its share price peaked at about $42 in 2021, when the company was valued at about $5 billion, per Small Caps, which does not state the currency.[^12] Ryan Kolln, previously chief operating officer, replaced CEO Armughan Ahmad, who left in February 2024.[^13] ### The Google contract and its aftermath In January 2024 Google ended its global inbound services contract with Appen, effective March. Google work had generated US$82.8 million of Appen's revenue in 2023 at a 26 percent gross margin, believed to be more than 30 percent of annual turnover, and the share price fell by more than 40 percent after the announcement.[^12] Appen responded with US$13.5 million of cost cuts and the closure of its Toronto and Bellevue offices.[^12] ### Recent financial performance Appen reports in US dollars. For the full year 2025, Appen's ASX announcement reported group revenue of US$230.8 million, up 4.5 percent on 2024 excluding the effect of the Google contract termination. Appen China contributed US$102.9 million (up 75 percent) and Appen Global US$127.9 million (down 21 percent). Gross margin was 40.3 percent, underlying EBITDA before foreign exchange was US$12.2 million, and cash was US$59.8 million at December 31, 2025.[^10] For the first half of 2026 (to June 30), Appen reported operating revenue of US$119.9 million, up 17.5 percent; including other income, revenue was US$120.8 million, up 17.0 percent. Appen China revenue rose 80.4 percent to US$76.2 million, and Appen Global fell 26.9 percent to US$43.7 million. The statutory net loss after tax was US$4.5 million, an improvement of 76.9 percent, and underlying EBITDA before foreign exchange was US$5.3 million versus a US$2.2 million loss a year earlier. No interim dividend was declared.[^14] Cash was US$44.7 million at June 30 with no debt, down US$15.1 million from December 31, and net cash used in operating activities was US$4.1 million.[^14] Appen reaffirmed full-year 2026 guidance of US$270 million to US$300 million of revenue and an underlying EBITDA margin of about 5 to 10 percent, and said Appen Global revenue is mostly project-based and skews to the second half of the year.[^5] It also identified about US$12 million of annualized cost reductions in Appen Global from using AI in operations, about 70 percent of which it expects to execute by the end of 2026.[^5] Revenue is concentrated: about 67.9 percent of group revenue in the half came from the top five customers, and Appen's announcement refers to Appen Global's "largest client" without naming it.[^14][^5] ### Products and business model Appen's current offering, as it describes it, includes frontier model alignment data (chain-of-thought reasoning, expert preference data, red teaming), agentic AI data such as agent trajectories and reinforcement learning environments, speech and multimodal data, physical AI data (LiDAR and robotics), coding datasets, and an annotation platform.[^1] It also claims SOC 2 and ISO 27001 certification and says 80 percent of leading LLM builders are customers, without naming any.[^1] Appen's 2026 results describe its two segments. Appen China includes Japan and Korea and, per Appen, has grown on generative AI projects including work for Chinese technology companies expanding internationally. Appen Global covers all operations outside the China group.[^14] Excluding Google, generative AI related revenue was 29.0 percent of total revenue in the half, up from 24.6 percent a year earlier.[^14] ### Off-the-shelf datasets Appen operates a separate dataset site advertising "over 300 audio, image, video and text datasets in over 80 languages" that are "licensable." In the filtered view we reviewed, 36 text and image datasets were listed, mostly academic and reference corpora, and the source filter showed only "Third Party Vendor."[^2] The page gives no license terms, and customers request quotes or contact the company.[^2] Appen's ASX announcements also cite "high-margin prebuilt datasets" as a contributor to Appen China's margin improvement in 2025 and the first half of 2026.[^10][^5] ### Contributors, pay and terms Appen's website says it has more than one million vetted contributors in more than 170 countries and 235 languages.[^1] Its August 2026 ASX release describes a crowd of more than one million contractors speaking over 500 languages in over 200 countries, and marks the language and country figures as self-reported.[^14] Appen's published position on pay is that contributor pay is "based on their local marketplace" and above minimum wage in every market; the 2022 post it appears in does not state contractor status or give figures.[^9] ### Controversies and legal issues - **Google raters (2023).** NPR reported in 2023 that Appen-employed Google raters earned $14 to $14.50 an hour and were limited to 29 hours a week. Google said its wages and benefits standards apply only to people assigned at least 30 hours a week, so these raters were not covered. Google said its suppliers manage employment terms; Appen did not respond to requests for comment.[^15] The Alphabet Workers Union said in June 2023 that it filed unfair labor practice charges with the NLRB against Google and Appen as joint employers on behalf of six fired raters, alleging retaliation. The release is the union's account, and we did not find a ruling.[^16] - **Payment problems (2024).** Capital Brief reported in November 2024 that after Appen moved contractors to its CrowdGen employment system, some said they had not been paid for September. Appen has publicly said the timesheet and payslip issues were fixed, and the article does not quote a direct Appen statement.[^7] - **Rate cuts (2024).** Capital Brief reported in November 2024 that about a dozen contractors said rates had fallen by as much as 90 percent; one translator said hourly pay dropped from $24 to $3. The article did not include an Appen response.[^17] - **US military contracts (2026).** The Bureau of Investigative Journalism, republished by Rest of World on February 23, 2026, reported that military agencies awarded Appen about $17 million in defence contracts between 2005 and 2020, mostly for linguistic projects, and that Appen's gig workers, including Somali transcribers, were not told who the client was. One subcontract mentioned the Rivet Joint surveillance aircraft; the reporters found no evidence that the Somali contracts were related to it. The article says Appen did not respond to requests for comment.[^18] These are separate episodes, some a few years old, and several remain allegations. ## Toloka ### History and founding Toloka says it was founded in 2014 by Olga Megorskaya, who remains CEO, and began as a crowdsourcing and microtasking platform before turning to training data for large language models.[^3] It is headquartered in Amsterdam with offices in the US, Israel, Switzerland and Serbia.[^3] Toloka sits inside the group that was formerly Yandex N.V. On July 15, 2024, Yandex N.V. announced it had completed the divestment of its Russia-based businesses, in a deal valued at about $5.4 billion, and said the retained businesses would be developed under the name Nebius Group.[^19] Follow the Money reported in March 2024 that Toloka would stay with Yandex's international business.[^20] ### Funding and ownership On May 7, 2025, Nebius announced that Bezos Expeditions had led an investment in Toloka, with Shopify CTO Mikhail Parakhin also participating and becoming executive chairman. Nebius kept a significant majority economic stake but gave up majority voting control and said it would stop consolidating Toloka's results.[^6] Nebius's release did not state the amount; SiliconANGLE reported the round at $72 million.[^11] Nebius's own release describes Toloka as an AI data solutions business and says the transaction aims to give it greater independence and governance flexibility.[^6] We found no 2026 funding round, valuation or revenue figure for Toloka. ### Products and business model Toloka's site lists SFT demonstrations, preference collection for RLHF and DPO, auto-verifiable tasks for evaluation and reinforcement learning, custom human evaluation and red teaming, datasets (including egocentric video and Arena data), and Mindrift for sourcing domain experts.[^3] In a post dated April 2026 Toloka described Toloka Arena, a private benchmark platform where models are evaluated on hidden test cases. The first run scored 10 frontier models across seven enterprise domains using a pass^5 reliability metric, and Toloka released its evaluation harness, Toloka Forge, as open source.[^21] For physical AI, Toloka markets a self-serve platform for collecting egocentric video, claiming a first batch in about 3 hours, no minimum order, and over 200,000 contributors in more than 100 countries. Its sample open dataset of 63 egocentric videos cost $50 to collect.[^4] ### Customers Nebius named Amazon, Anthropic, Microsoft, poolside, Recraft and Shopify as Toloka customers in its May 2025 release, and SiliconANGLE named Anthropic and poolside.[^6][^11] Toloka's own site names Anthropic, Amazon and Microsoft.[^3] These are 2025-era statements, and we did not confirm them for 2026. ### Contributors, pay and terms Toloka says it works with data workers in more than 100 countries across more than 40 languages and 50 knowledge domains. It gives no total on its about page, and it does not state pay for its crowd workers there.[^3] SiliconANGLE reported a network of over 200,000 annotators and other professionals.[^11] Mindrift, operated by Toloka AI, is the freelance expert platform. Its own page describes "independent freelancers who set their own schedules," pay of roughly $30 an hour on general projects up to $90 an hour on specialist technical work, payment only for completed and approved submissions, and country restrictions. These are Mindrift's own claims.[^8] ### Controversies Follow the Money, with the Bureau of Investigative Journalism and Paper Trail Media, reported in March 2024 that Toloka supplied gig workers to Tevian and NTechLab, Russian facial recognition firms that supply Moscow's camera surveillance network, and that both firms were sanctioned by the EU in July 2023. Toloka said neither Yandex N.V. nor its Swiss, EU or US entities provided services to or received payments from the two firms, and that they had contracts with Yandex's Russian subsidiary.[^20] The Continent reported in August 2024, citing the Bureau, that five anonymous African freelancers said they did not know what their work was used for. The same article reported task pay of $0.10 to $8, with rejected tasks unpaid, and one worker earning about $7 over two months.[^22] Toloka disputed the findings, said it does not do business in Russia, said it has "never done business with sanctioned entities," and said its rates are competitive and that workers see fees before starting.[^22] We found no lawsuit against Toloka. ## Head to head ### Transparency and size Appen is the more transparent company because it is listed: filed numbers show where growth and weakness sit. Its Chinese operation now generates most of its revenue (US$76.2 million of US$119.9 million in H1 2026), while its Global business is shrinking and, per Appen, partly dependent on a largest client.[^5][^14] Toloka discloses almost nothing financial, so claims about its scale rest on its own statements and the 2025 round.[^6][^11] ### Ownership and independence Appen is independent and publicly held. Toloka is backed by a parent that has stepped back from voting control but remains the main economic owner, plus Bezos Expeditions.[^6] ### Product focus Appen is broader and more traditional: speech and language data across many locales, plus off-the-shelf datasets, with new lines in agentic and physical AI.[^1] Toloka leans toward new-style work: expert evaluation, benchmarks like Arena, self-serve collection and a separate expert platform.[^3][^21][^4] Neither is clearly better; they fit different buyers. ### People and pay Both manage very large contractor crowds, and both have been criticized in press reports on pay and transparency about clients.[^15][^17][^22] Appen publishes a local-market, above-minimum-wage policy; Toloka's Mindrift publishes hourly ranges for experts.[^9][^8] The figures describe different programs, so avoid comparing them directly. ### How each makes money Neither company publishes a clean price list. Appen's filings report revenue by segment and describe Appen Global's work as mostly project-based, which is why customer timing moves results so sharply.[^5][^14] Toloka's pages describe self-serve data collection without minimums, managed services for custom data and evaluation, and an expert marketplace through Mindrift, but no pricing beyond a single $50 example.[^4][^8] ### What to ask either vendor If you are evaluating either as a supplier, a short list of questions is more useful than a feature comparison. Ask who the contributors are and how they are screened, how quality is measured, what happens to your data after delivery, whether any of it is retained or reused, and whether work is subject to country restrictions on contributors. Public materials from both companies leave these points to the contract. For Toloka, ask which entities beyond Nebius hold economic interests, since the ownership percentages are not public.[^6] ## Strengths and weaknesses ### Appen - Strengths: filed half-year and annual reporting, a long track record, broad language coverage, a licensable dataset catalog, and strong growth in China.[^1][^5][^2] - Weaknesses: revenue concentration, a shrinking Global segment, negative operating cash flow in H1 2026, and recurring contractor-pay and transparency concerns.[^14][^17] ### Toloka - Strengths: well-known frontier AI customers named in 2025, fresh capital, an expert platform and a public benchmark product.[^6][^21][^8] - Weaknesses: no disclosed financials, past scrutiny over its Russian operations, and criticism of crowd-task pay.[^20][^22] ## Who each suits Appen suits buyers that want multilingual or locale-specific speech and text data at scale, a vendor with filed financials, or off-the-shelf datasets to license.[^1][^2] It may also suit annotation work in China, Japan and Korea, given that its Chinese business is now the larger segment.[^5][^14] Toloka suits AI teams that want to commission new data, evaluations or benchmarks, including physical AI data, through a self-serve or managed workflow.[^3][^4] Its Mindrift platform suits experts looking for freelance work, subject to its eligibility rules.[^8] ## What this means for data sellers The key distinction is between commissioning new data and licensing existing data. Both companies mostly do the first. Toloka's public pages describe collecting new data on demand.[^4] Appen does both: it collects custom data and also lists a catalog of licensable datasets, some from third parties.[^2] That makes Appen the only one of the two with a visible resale channel, but there are limits. The dataset site we reviewed showed a single source category, "Third Party Vendor," and gave no license terms or information on how vendors join.[^2] We found no program in either company's materials for businesses to submit proprietary data, and Appen's results announcements speak of project-based work and prebuilt datasets rather than acquired archives.[^5] Practical points if you hold data: 1. **Check terms first.** A catalog of licensable data tells you resale exists, not what vendors are paid or how rights are handled. 2. **Expect scrutiny of provenance.** Both companies have faced press questions about who their work serves and how contributors are treated.[^18][^20] Buyers care about consent and rights, so see our [trust page](/trust/). 3. **Start with your own inventory.** Our [assessment](/assessment/) and [data types](/data/) pages explain what kinds of business data are typically of interest. SellYourAIData is not affiliated with Appen or Toloka, and nothing here says either buys data from you. **Do not let one buyer set your price.** Any single buyer's published range, program page or first offer is that buyer's number, not the market's. The sensible move is to get competing offers on the same data, compare scope, exclusivity, indemnity and payment timing side by side, and have someone independent look at the terms. Our guide to [selling direct vs using a broker](/guides/selling-direct-vs-broker/) explains how, and if a buyer is already talking to you, [we review offers free](/offer-review/). ## What we could not verify - Appen's largest customer is not named in the sources we found. - Appen's contributor, language and country counts differ between its website and its ASX release, and are self-reported.[^1][^14] - Toloka's revenue, valuation and current ownership percentage are not disclosed; the round amount comes from press coverage rather than Nebius's release.[^6][^11] - Where the Bureau of Investigative Journalism's original reporting was not accessible to us, we relied on Follow the Money, The Continent and Rest of World. - Worker accounts of pay and awareness of clients are anecdotal, and allegations in union, investigative and press reports have not been adjudicated. ## Related reading For broader background, see our [guide to selling data to AI companies](/guides/selling-data-to-ai-companies/) and the [Scale AI vs Surge AI](/compare/scale-ai-vs-surge-ai/) comparison. [^1]: [About Appen](https://www.appen.com/about) (Appen, accessed October 2026) [^2]: [Off-the-shelf (OTS) Datasets](https://datasets.appen.com/data-types-image-and-text) (Appen, accessed October 2026) [^3]: [About Toloka](https://toloka.ai/about) (Toloka, accessed October 2026) [^4]: [Egocentric video data collection](https://toloka.ai/egocentric-video) (Toloka, accessed October 2026) [^5]: [H1 FY26 results](https://announcements.asx.com.au/asxpdf/20260827/pdf/0739d4q7lyh3w2.pdf) (Appen, ASX announcement, August 2026) [^6]: [Nebius welcomes Bezos Expeditions as lead investor in AI data business Toloka](https://nebius.com/newsroom/nebius-welcomes-bezos-expeditions-as-lead-investor-in-ai-data-business-toloka) (Nebius, May 2025) [^7]: [Appen says it fixed its payment fiasco but some contractors insist they're still waiting](https://www.capitalbrief.com/article/appen-says-it-fixed-its-payment-fiasco-but-some-contractors-insist-theyre-still-waiting-5248200e-dfc8-4ba2-8571-995ad6f5f53e/) (Capital Brief, November 2024) [^8]: [Is Mindrift legit? What you need to know before joining](https://mindrift.ai/blog/legit-mindrift) (Mindrift, accessed October 2026) [^9]: [AI ethics: how fair pay benefits our crowd contributors](https://appen.com/blog/ai-ethics-how-fair-pay-benefits-our-crowd-contributors/) (Appen, June 2022) [^10]: [FY25 results and FY26 outlook](https://announcements.asx.com.au/asxpdf/20260225/pdf/06wq7ntr51bhgl.pdf) (Appen, ASX announcement, February 2026) [^11]: [AI data provider Toloka raises $72M in funding](https://siliconangle.com/2025/05/07/ai-data-provider-toloka-raises-72m-funding/) (SiliconANGLE, May 2025) [^12]: [Appen to slash costs, close North American offices in wake of Google deal termination](https://smallcaps.com.au/appen-slash-costs-close-north-american-offices-google-deal-termination) (Small Caps, 2024) [^13]: [Appen CEO Armughan Ahmad replaced by COO Ryan Kolln](https://www.capitalbrief.com/briefing/appen-ceo-armughan-ahmad-replaced-by-coo-ryan-kolln-68fa32d1-cd2b-4426-9db6-5ae8d2141e17/) (Capital Brief, February 2024) [^14]: [Appendix 4D and Interim Financial Report, half-year ended 30 June 2026](https://announcements.asx.com.au/asxpdf/20260827/pdf/0739ct3d5q4g32.pdf) (Appen, ASX announcement, August 2026) [^15]: [Google's 'Ghost Workers' Demand to Be Seen, Seeking Better Wages and Benefits From Tech Giant](https://www.kqed.org/news/11945357/googles-ghost-workers-demand-to-be-seen-seeking-better-wages-and-benefits-from-tech-giant) (KQED/NPR, April 2023, updated February 2025) [^16]: [Press release: raters illegal firing ULP](https://www.alphabetworkersunion.org/press/raters-illegal-firing-ulp) (Alphabet Workers Union-CWA, June 2023; union statement) [^17]: ["Slap in the face": Appen accused of slashing pay rates for contractors](https://www.capitalbrief.com/article/slap-in-the-face-appen-accused-of-slashing-pay-rates-for-contractors-c7f385c1-b839-4804-aa2e-b842d7698778/) (Capital Brief, November 2024) [^18]: [Gig workers and the US military: Appen](https://restofworld.org/2026/gig-workers-us-military-appen/) (Rest of World, republished from the Bureau of Investigative Journalism, February 2026) [^19]: [YNV announces successful completion of the divestment of its Russia-based businesses](https://nebius.com/newsroom/ynv-announces-successful-completion-of-the-divestment-of-its-russia-based-businesses) (Nebius, July 2024) [^20]: [Russia relies on online gig workers to fuel surveillance mechanism, and a European company is helping it](https://www.ftm.eu/articles/dutch-yandex-daughter-supplies-russia-with-surveillance-software) (Follow the Money, March 2024) [^21]: [Toloka Arena: independent evaluation of agentic intelligence](https://toloka.ai/blog/toloka-arena-independent-evaluation-of-agentic-intelligence/) (Toloka, April 2026) [^22]: [Serving Putin, without even knowing it](https://continent.substack.com/p/serving-putin-without-even-knowing) (The Continent, August 2024) ## FAQ **What is the difference between Appen and Toloka?** Appen is an Australian-listed company founded in 1996 with a long history in speech, search and text data, now split into Appen China and Appen Global. Toloka is an Amsterdam-based company founded in 2014 as a crowdsourcing platform and now focused on data for frontier AI developers, with Nebius holding a majority economic stake but not majority voting control. **How big is each company?** Appen reported US$230.8 million of revenue for 2025 and US$119.9 million for the first half of 2026. Toloka does not publish revenue. Toloka says it works with data workers in more than 100 countries, and its egocentric video page cites over 200,000 contributors. Appen says it has more than one million contributors. **Do Appen or Toloka buy datasets from companies?** Neither says so in the public materials we reviewed. Appen runs a catalog of licensable off-the-shelf datasets, and its dataset site lists the source of the datasets in one filtered view as a third-party vendor. That shows Appen resells data from some outside parties, but it does not show that it buys proprietary business data. Toloka's pages describe commissioning newly collected data. **How much do contributors earn?** It depends on the project. Appen says it pays above local minimum wage, with pay tied to local markets; Toloka's expert platform Mindrift cites roughly $30 an hour for general projects and up to $90 an hour for specialist work, while reporting on Toloka's crowd tasks described pay from $0.10 to $8 per task. These are not directly comparable programs. **What controversies surround each company?** Appen: Google rater pay and a 2023 labor charge filed by a union against Google and Appen, contractors' reports of payment problems and rate cuts in 2024, and February 2026 reporting on US military contracts. Toloka: investigative reporting that it supplied gig workers to Russian facial recognition firms through its Russian branch, which Toloka disputes, and worker complaints about low task pay. See the sections below for sources and responses. **Is Appen's stock a good indicator of the AI data market?** It is a rare public data point, but it is a narrow one. Appen's revenue is concentrated: about 68 percent of its first-half 2026 revenue came from its top five customers, per its interim report, and Appen Global's work is mostly project-based. This page does not provide investment advice. **Should a company with proprietary data approach Appen or Toloka?** Based on public materials, neither is an obvious buyer. A company holding a proprietary dataset would need to ask each directly whether it licenses or resells third-party data, and compare the options on the SellYourAIData buyers page.