Trained annotators, checked work
Annotators recruited and managed by Innovatiana, a calibration round before production, sampling reviews, inter-annotator agreement and gold-standard checks, with a dedicated manager.
Free tool · cost estimator
Annotation cost estimator
Estimate what a custom annotation request costs in credits and euros, how your plan changes it, and how it compares with labelling the same data with your own team. The inputs match the request form in Dataset Finder, so you can order the same thing in a few minutes.
- Same inputs as the order form
- Compared with in-house labelling
- No sign-up, runs in your browser
01 · What a request includes
Labelling, QA and delivery, handled
Fast first batch
A test batch usually within about 48 hours, so you can check the guidelines are understood before the full run. Even the largest requests are typically delivered within a month.
Your format, your tools
Delivery as COCO, YOLO, Pascal VOC, JSON, JSONL, CSV, XML or a custom schema. Annotators can work inside your Label Studio, CVAT, Labelbox, Encord or SuperAnnotate project.
Writing the guidelines is the step that decides quality. See how to write labelling guidelines, and how buying credits works.
02 · FAQ
Questions about cost and credits
What is a credit?
Credits are how custom data is paid for in Dataset Finder. A request uses credits according to its data type, use case, taxonomy and volume, and the request form shows the exact number before you submit. You buy credits in packages (100 credits = €75) or as a custom amount, and Pro includes 20 credits each month.
Why is it cheaper than labelling in-house?
Because the people doing the work are trained annotators supervised by a dedicated manager, not your engineers learning the task as they go. The comparison on this page assumes your own ML engineers would do the labelling at €60 an hour, with 30% more time for guidelines, review and corrections. It leaves out recruiting, training annotators and tooling, so the real in-house cost is usually higher still.
What makes a request cost more?
Richer data and richer labels. Video, multimodal and medical data take longer per item than text or images; use cases such as preference data, Gen-AI pairs and clinical annotation need more expertise; and more classes or more objects per item add effort. Asking the team to set up an annotation platform adds a one-off cost. Startup and Pro plans reduce the credits used.
Is the estimate the final price?
No. It uses the same inputs as the request form, but minutes per item vary within each range. The app calculates the exact number of credits for your request before you submit, and you are told if your balance is short.
How do plan discounts work?
Your plan changes how far credits go: Startup uses 10% fewer credits on custom orders and Pro 25% fewer, plus 20 included credits a month. Free orders at the standard rate. Every plan can buy credits.
How long does delivery take?
A test batch usually comes back within about 48 hours, and even the largest requests are typically delivered within a month. Very large or ambitious requests move to an Enterprise contract, so effort, cost and timeline can be planned together.
Do you work in my annotation tool?
Yes. Annotators are experienced with Label Studio, CVAT, V7, Labelbox, Encord, Roboflow, SuperAnnotate and Prodigy. If you have no platform, the team can set one up, which adds a one-off set-up cost.
Who does the labelling?
Trained annotators and domain specialists recruited and managed by Innovatiana, supervised by a dedicated Data Labeling Manager, with a calibration round, sampling reviews, inter-annotator agreement and gold-standard checks. Not an open crowd.
More free tools
Other tools for AI training data
How much data do I need?Pilot, workable and strong ranges for your fine-tuning task, with sources.EU AI Act dataset documentationA datasheet mapped to Article 10 and Annex IV, downloadable as Markdown or PDF.Annotation guidelines generatorLabelling guidelines with definitions, edge cases and format, exported to Word or PDF.Cohen’s kappa calculatorInter-annotator agreement and weighted kappa, with a confidence interval.