Transparent pricing, annotation and collection.
Data Annotation is billed per accepted unit, QA and rework included. Data Collection is billed per accepted unit for repeatable capture, or per person-hour for open-ended work, whichever fits the task.
// how pricing works
Billed on evidence, not effort
Every batch is QA-sampled against your guideline. Units that pass are billed. Units that fail are corrected at our cost and re-reviewed before you see an invoice.
We price per unit of accepted work: per box, frame, minute or document that passes quality review. The rate you’re quoted covers the people: trained annotators, a dedicated project manager, QA sampling and rework.
Tooling is scoped per project. Where the work runs on your platform or on an open-source stack we set up, it costs you nothing extra. Where it genuinely needs a specific commercial platform, the licence is a named line in the quote, agreed before work starts and never discovered on an invoice.
Where a project lands inside a range depends on a small set of drivers: how many classes you need, how many attributes each object carries, how deep the QA sampling goes, and how dense the edge cases are. Each card below names the drivers that move its price, and a paid pilot batch pins your exact rate down before you commit to volume.
Hourly billing exists as a quiet second option for work where counting units doesn’t fit: guideline development, exploratory labelling, mixed tasks. Where it’s genuinely the cheaper structure for you, we’ll recommend it in the quote.
Specialised domain work such as medical, legal or highly technical content is estimated separately, based on the certifications and experience level the task needs.
// per-unit rate card
Cost breakdown by annotation type
Ranges per accepted unit, all-inclusive. Before you anchor on the bottom number, check the moves_price line: class count, attributes and QA depth are what separate a $0.03 box from a $0.07 one.
Cuboids and segmentation in point clouds: vehicles, pedestrians, infrastructure.
$0.10–$0.25 per object$10–$35 per scene moves_price: classes · points_per_scene · sensor_fusion
Transcription, sound-event tagging, speaker diarisation.
$1.50–$6.00 per minute$0.10–$0.50 per segment moves_price: class_count · language_tier · timestamp_precision
People, vehicles, products.
$0.03–$0.07 per box moves_price: classes · attributes_per_box · object_densityBuildings, crops, irregular shapes.
$0.07–$0.20 per polygon moves_price: vertex_precision · classes · edge_case_densityLane markings, river and path mapping, crack and defect tracking.
$0.03–$0.06 per line moves_price: curve_complexity · lane_rules · resolutionPose estimation, facial landmarks.
$0.02–$0.05 per keypoint moves_price: points_per_skeleton · occlusion_rulesScene understanding, pixel-level labelling.
$0.75–$3.00 per image moves_price: class_count · pixel_precision · scene_complexityObject tracking across continuous footage, interpolation-assisted.
$1.50–$6.00 per minute moves_price: objects_in_frame · classes · occlusion_handlingSampled frame sets and frame-accurate labelling, priced per labelled frame, not per minute.
from $0.02 per labelled frame moves_price: boxes_per_frame · attributes · qa_depthEntity tagging, sentiment and classification, priced per entity, sentence or document.
$0.02–$0.15 per entity$0.50–$3.00 per document moves_price: entity_types · language_tier · guideline_complexity
volume_pricing: 100k+ units → typically 10–15% off rate card · larger sustained programmes scoped individually
// worked examples
What projects actually cost
Three representative scopes, priced the way we’d quote them today. Your numbers will move with your guidelines and QA depth. That’s exactly what the pilot batch is for.
First-person footage, two hand boxes with five attributes each, frame-accurate.
Nursery recordings segmented and labelled for a sleep-monitoring model: cries, snoring and background events.
Short texts labelled for intent and tone against a 12-class guideline, double-reviewed.
// data collection pricing
Priced by the unit, or by the hour
Collection is billed one of two ways, decided when we scope the work. Repeatable capture, a photo, a sample, a completed form, a short clip, is billed per accepted unit: the same pay-for-what-passes-review principle as Data Annotation. Open-ended work that rides on a person’s time and judgement, interviews, moderated sessions, structured field research, is billed per person-hour.
// by the accepted unit
A traffic-sign photo from a phone, a soil sample from a field site, a completed household survey: each has its own per-unit rate, the same way every row on the annotation rate card has its own price. We set it once we’ve seen a sample of the unit and the volume: there isn’t one number that fits a phone photo and a lab-verified sample.
// by the person-hour
For work that can’t be counted in units, the rate is per person, per hour, and it tracks local income where the work happens, not a discount tier. A lower-cost economy can’t substitute for someone who has to be physically in San Francisco, or fluent in a dialect spoken in one province: if the work has to happen in a specific place, that’s a requirement, not a value trade-off. Choose where the work is based:
Web and public-record research, cataloguing, form-based sourcing, straightforward transcription.
$6–$9 per hour moves_rate: volume · criteria_specificity · locationNative-language interviews, moderated recordings, structured field surveys.
$10–$16 per hour moves_rate: language/skill · session length · locationClinical, legal, technical or otherwise regulated data capture; rare-language or credentialed work.
$16–$25 per hour moves_rate: certification · compliance · locationWhat determines the rate?
Bands are set by income, not a label we chose: lower-cost economy is our own certified workforce, the right fit whenever the work has no location or language requirement, the same as on the annotation rate card · mid-cost economy runs roughly 2× the tier, higher-cost economy (US, Western Europe) roughly 3.5×, tracking average local pay, not quality or effort. We’re not paying less because it’s cheap for us to: the rate reflects the cost of living where the work happens, and we’re glad to pay more where the engagement and the client’s budget call for it. See the World Bank’s income classification of economies for the country-by-country data, or tell us what the task requires and we’ll quote the right band exactly.
deelabgigs
Prefer a matched team to a quote?
DeeLab Gigs runs the same annotation and collection work through a different front door: you post the need, we match the people, your estimate comes back fast.
Describe the work on deelabgigs.com with as much detail as you have: task, volume, skills, region, timeline.
We match against our vetted, Academy-trained talent by skill, experience, availability, willingness and region.
The shortlist goes to your assigned Business Manager, who turns it into a cost and timeline estimate, fast.
// start small
Start with a paid pilot
A flat USD 250 pilot batch: your data, your guidelines, full QA reporting. The fee is credited in full against your first production invoice. Judge us on output, not promises.
/pricing/faq
Pricing questions, answered
q_01What counts as an “accepted” annotation?
Every batch goes through sampled QA review against your guideline. Units that pass are accepted and billed. Units that fail the sample are corrected at our cost and re-reviewed before they’re billed. You never pay for work that didn’t meet the bar.
q_02What happens when errors are found after delivery?
Flag them. Sampled QA failures are ours to fix, free, and that applies after delivery too. Report an issue inside the review window agreed in your statement of work and we correct it at no cost.
q_03Who sees my data?
Every annotator is an individually NDA-contracted member of our own workforce, trained and certified through DeeLab Academy. We don’t subcontract to third-party firms or crowd platforms. Access is limited to the assigned project team, and we can work inside your environment if your security policy requires it.
q_04What currency do you invoice in?
Quotes are in USD. Invoices are issued in USD or SGD (we’re a Singapore company), and SGD conversion uses the Xe mid-market rate on the invoice date, with no exchange-rate padding.
q_05Is there a minimum project size?
The paid pilot is the practical minimum: a flat USD 250, credited against your first production invoice. There’s no minimum volume commitment after that: scale up or stop on the evidence.
q_06When do volume discounts apply?
From roughly 100,000 units, volume pricing typically takes 10–15% off the rate card. Larger, sustained programmes are scoped individually.
q_07Can I still be billed hourly?
Yes. Some work, such as guideline development, exploratory labelling or mixed tasks, is priced better hourly than per unit. Where that’s genuinely cheaper for you, we recommend it in the quote.
q_08Which annotation platform do you work on?
Whichever fits the project: your platform, our own stack, or an open-source tool we set up and run. Where the work needs a specific commercial platform, its licence cost appears as a named line in the quote and is agreed before work starts.
q_09Can annotation happen in a specific country or region?
Yes, on request. Annotation is remote, digital work, so by default we staff it from our own certified global team: there’s no location the work itself requires, so this is the most cost-efficient fit. Where compliance or your project specifically requires annotators based in a higher-cost region (the US, Western Europe), the hourly rate and delivery cost scale up to reflect that region’s local labour cost, typically 2–3.5× the standard rate card. Often the cheaper answer is data residency instead: your data stays in your environment and our standard team works inside it.
q_10How is Data Collection priced?
Whichever basis fits the task. Repeatable capture, a photo, a sample, a completed form, is billed per accepted unit. Open-ended work that rides on a person’s time, interviews, moderated sessions, research, is billed per person-hour, and there the rate follows local income where the work happens: a lower-cost economy for work with no location or language requirement, or a mid- or higher-cost economy at roughly 2–3.5× the tier when the task requires it, the same logic as Data Annotation, priced by what the work needs and what it costs to live where it happens, not by which team is cheapest. Describe what you need collected and we’ll tell you which basis fits and quote it within 48h.
q_11What’s the difference between a quote here and a DeeLab Gigs estimate?
Not cost: it’s the same company, the same Talent Pool, and the same underlying pricing either way, and we reply within 48h no matter which door you use. The difference is fit. DeeLab Gigs is a fast, self-serve form: fill it in and we can get things rolling quickly once we see who’s available and who fits. For more custom needs, contacting us directly is often the better route, since your account team can properly scope it. Use whichever fits how you work.
// tell us about your data