Appen review: the documented-dataset training data company
Answer in brief: Appen has supplied human data since 1996. Today it offers custom collection and annotation, frontier-model alignment data, agentic data, and an off-the-shelf catalog of 646 datasets across nine categories. Each catalog dataset comes with a spec sheet covering source, consent, annotation method, known limitations and license terms. Licenses are perpetual and non-exclusive with commercial training rights. This is independent research; we did not test Appen hands-on.
Top pick for: buyers who need licensed, documented datasets quickly, especially speech and multilingual. Visit Appen →
Where Appen sits in the directory
Appen spans three categories: human data, licensed datasets and collection. That makes it one of the most cross-category companies here. Appen is listed on the ASX (APX).
What it does for AI training data
- Off-the-shelf catalog: tasks and verifiers for RL, code repositories, academic text corpora, speech and audio, pronunciation dictionaries, enterprise company data, image and video, specialized and location data.
- Frontier alignment: chain-of-thought reasoning, expert RLHF, SFT and red teaming.
- Agentic AI: agent trajectories, RL environments and tool-use logs.
- Data collection: remote, supervised on-site and device-based collection, including wearables and AR/VR glasses.
Who it fits
Teams that want to license data and start training right away, compliance-heavy buyers who need provenance records, and programs that need many languages.
Why buyers pick Appen
- Provenance built in. Annotation quality is reported with Cohen's kappa or Krippendorff's alpha, along with adjudication protocols.
- Governance. SOC 2 Type II, ISO 27001, GDPR/CCPA handling and EU AI Act documentation.
- Reach. 500+ locales. Delivery formats include WAV, FLAC, JSON, JSONL and COCO. Samples are available under NDA.
- Long track record. One case study covers 10+ years of in-car speech collection in 20+ languages for an automotive OEM.
Honest drawbacks
- Catalog licenses are non-exclusive, so other buyers can train on the same data.
- Revenue is about $249M for the twelve months to June 2026 (compiled from filings), and FY2025 operating revenue was $230.8M. That is below the fastest-growing expert-data firms.
Pricing
Quoted per dataset or custom program.
Best alternatives to Appen
- TELUS Digital: high-volume multilingual data plus off-the-shelf datasets.
- Toloka: off-the-shelf datasets for agents and LLMs.
Shopping for ready-made datasets rather than a supplier? See bestaidatasets.com.
Appen FAQ
How many datasets are in Appen's catalog?
646 datasets across nine categories, according to Appen's catalog page.
Can I use Appen datasets commercially?
Yes. The standard license is perpetual and non-exclusive with commercial training rights.
Is Appen data exclusive?
Catalog data is non-exclusive. Custom programs are scoped separately.