18 vendors, scored on what they publish
Best Data Annotation Companies
Find a labeling vendor you can check before the first sales call. We rank annotation companies on published pricing, modalities and security reports, then introduce you to the ones that fit your brief. We have not tested any of them, so pilot before you commit.
- $0.02+per bounding box, Label Your Data, managed labor
- $6an annotator hour, the usual hourly rate at Label Your Data
- $2.67an hour of platform use, by Segments.ai's own sum
- $99a month for 12 hosted users, HumanSignal Starter Cloud
Each figure is the vendor's own published price, never an offer. Only 3 of the 18 vendors print dollar prices; an annotator hour and a platform hour are different units. How each pricing model works.
The five you can check most before a call
Each vendor scores up to 16: published pricing (6), modalities covered (6), security reports listed (3), and selling both a platform and services (1). How we score.
- 11/16
Number 1: Kili Technology
Platform and servicesThe most checkable platform in the table: a free trial you can open today, and security reports named on the pricing page.
- Trial
- Free: $0, 2 to 5 seats, up to 100 assets
- Security
- SOC 2 Type II; ISO 27001
- 11/16
Number 2: Segments.ai
PlatformBuilt for robotics and autonomy teams labeling point clouds and fused sensor data, with a rate you can work out before a sales call.
- Price
- By its own sum, $2.67 an hour, on a 3,600-hour yearly minimum
- Security
- ISO 27001; TISAX
- 11/16
Number 3: HumanSignal (Label Studio)
PlatformThe open-source route: run Label Studio yourself for free, or pay a flat seat price for the hosted edition and bring your own annotators.
- Price
- $99/month hosted, 12 users in; +$49/month per added user
- Security
- SOC 2, on the Enterprise tier
- 11/16
Number 4: Label Your Data
Managed serviceThe only managed service in our table that publishes per-object and hourly rates, and it offers a pilot at no cost.
- Price
- $0.02+ per box; hourly work usually $6 an annotator hour
- Pilot
- Free pilot, no stated minimum
- 10/16
Number 5: Labelbox
Platform and servicesA broad platform covering five modalities, with labeling services on top. Usage is metered in its own units, so ask for the dollar rate.
- Free tier
- 500 LBUs free each month
- Metering
- 1 LBU per image row; dollar rate per LBU unpublished
A few of these five, and others below, may reward us for introductions that turn into signed work. Rank and verdicts come from the method alone.
Ranks 6 to 18
- SuperAnnotatePlatform and services · Plans published, no dollar rate · No security report found9
- DataloopPlatform · Quote on request · ISO 27001, ISO 27701, SOC 2 Type II8
- SamaManaged service · Quote on request · ISO 9001, ISO 27001, TISAX7
- Scale AIPlatform and services · Quote on request · No security report found7
- EncordPlatform · Quote on request · SOC 2 (type not stated)6
- AppenManaged service · Quote on request · No security report found5
- CentificPlatform and services · Quote on request · No security report found5
- TELUS Digital (AI Data Solutions)Managed service · Quote on request · No security report found5
- TolokaManaged service · Quote on request · No security report found5
- CloudFactoryPlatform and services · Quote on request · No security report found4
- InnodataManaged service · Quote on request · No security report found4
- Snorkel AIPlatform and services · Quote on request · No security report found3
- KeymakrManaged service · Quote on request · No security report found2
"No security report found" means none appeared on the pages we read, not that the vendor has none. Ask. Vendors may reward us for introductions; that plays no part in the order. We are independent of every vendor here: none owns, funds or endorses the site.
Pages each vendor's entry comes from
- Kili Technology: kili-technology.com
- Segments.ai: segments.ai
- HumanSignal (Label Studio): humansignal.com
- Label Your Data: labelyourdata.com
- Labelbox: docs.labelbox.com
- SuperAnnotate: superannotate.com
- Dataloop: dataloop.ai
- Sama: sama.com
- Scale AI: scale.com
- Encord: encord.com
- Appen: appen.com
- Centific: centific.com
- TELUS Digital (AI Data Solutions): telusdigital.com
- Toloka: toloka.ai
- CloudFactory: cloudfactory.com
- Innodata: innodata.com
- Snorkel AI: snorkel.ai
- Keymakr: keymakr.com
From a project brief to a shortlist
Send the outline of the work, never the data. We match it against what each vendor publishes and name the ones worth a call.
Project brief
Robotics perception, phase one
- Modality
- 3D point clouds with camera frames
- Volume
- About 250k frames over six months
- QA target
- Agreement with a gold set you label yourself
- Security
- A named SOC 2, ISO 27001 or TISAX report
- Who labels
- Either: your team or the vendor's
Invented example brief, not a real client.
Shortlist from our data
Segments.ai
PlatformLabeling software aimed at robotics and self-driving teams: camera images, LiDAR point clouds and fused sensor views.
- Labels 3D and sensor data
- ISO 27001, TISAX
- Prices published
Dataloop
PlatformData platform for unstructured files, used to run multimodal pipelines from raw data through to labels.
- Labels 3D and sensor data
- ISO 27001, ISO 27701, SOC 2 Type II
- Quote on request
Sama
Managed serviceAnnotation and validation by Sama's own workforce on its own platform; vision and generative AI work above all.
- Labels 3D and sensor data
- ISO 9001, ISO 27001, TISAX
- Quote on request
Vendors with 3D coverage and a named security report, in rank order. Any of them may reward us for an introduction that becomes signed work; that has no bearing on who appears.
Which vendors cover your modality?
Coverage is what each company says it labels. Depth varies: a vendor that lists audio may do far less of it than one built around speech.
Vendors per modality, of 18
Managed service, platform, or both?
The choice usually follows one question: who will do the labeling? 7 of the 18 vendors sell both, and many teams use each at a different stage.
Managed service
Trained annotators, project managers and quality review supplied for you, charged per item or per hour.
Fits: no spare people, or one large dataset to build.
Labeling platform
Your own people get the tool, usually billed by seat or by usage, and the data can stay closer to home.
Fits: ongoing work, sensitive data, in-house experts.
What should you check before you sign?
The risks in annotation contracts are rarely about price. They sit in who sees your data, who owns the output, and what happens when quality slips.
Ask for the report, not the badge
A SOC 2 Type II report covers controls over a period; ISO 27001 certifies a management system for a defined scope. 6 vendors here list one or more. Check the dates and which entity is covered.
You own the labels
The raw data, the labels and the guidelines stay yours, and the vendor may not train its own models on your data without written consent.
Who does the work, and where
Employees or contractors, which countries, and whether they can download files. If data cannot leave the US, put it in the RFP.
A remedy for poor quality
Write the metric, the target and the fix (free rework, then a credit) into the order. Our post on measuring annotation quality covers gold sets and agreement scores.
How an annotation project usually runs
To start, write the brief: the RFP guide lists what vendors need to price it.
Scope and guidelines
You define classes, shapes and edge cases; the vendor asks questions and may help write the guide.
Paid pilot
A few thousand of your items, scored against a gold set you labeled yourself.
Calibration
The guide is tightened where annotators disagreed, and targets are set from pilot results.
Production
Regular deliveries, each audited on a sample, with rework for misses.
Close-out
Final delivery, written confirmation your data copies are deleted, and a review before renewal.
Questions about choosing an annotation vendor
Which is the best data annotation company?
On published information, Kili Technology ranks first: its pricing page offers a free trial and cites both SOC 2 Type II and ISO 27001. The best company for your project depends on modality, volume and whether you want a managed service or a platform, so treat the ranking as a shortlist, not a verdict. We have not run projects with any of these vendors; a paid pilot on your own data is the real test.
How much does data annotation cost?
Few vendors publish prices. Of the 18 here, three show dollar figures: Label Your Data (two cents a box and up, or about $6 an annotator hour), Segments.ai ($2.67 an hour of platform use by its own conversion) and HumanSignal ($99 a month for a hosted team of 12). The other fifteen quote on request. See the pricing guide.
Should we outsource data annotation or do it in house?
Outsource to a managed service when you have no spare people or one large dataset to build. Keep it in house on a platform when the work is ongoing, the data is sensitive or your own experts must settle edge cases. Many teams do both: start managed, then move routine work in-house later. Of the 18 vendors, 7 sell mainly a managed service, 4 mainly a platform and 7 sell both.
Is a managed annotation service or a labeling platform better?
A managed service supplies the annotators and the project management; a platform gives your own team the tool. Teams without spare people usually start managed. Teams with in-house labelers or strict data controls often prefer a platform. Compare the two.
Which annotation companies hold SOC 2 or ISO 27001?
On the pages we read, 6 of the 18 vendors list a SOC 2 report, ISO 27001 or both: Kili Technology, Segments.ai, HumanSignal (Label Studio), Dataloop, Sama, Encord. Ask any vendor for the current report and its scope before you share data.
Do vendors pay to be listed or ranked?
No vendor can pay for a place. Some reward us after an introduction turns into a signed project, and none of that money touches inclusion, rank or wording.
Get a quote
A ZIP code, a reply address, then a few taps describing the work. We match the outline against what each vendor publishes and write back with names; no vendor hears about you until you pick one.
Keep data, samples, label guides and logins out of this form. The outline on its own is enough for us to work from.
How this site earns: as an introduction service, it may be paid by the annotation company you are introduced to.
Received. Next steps from our side:
- Your outline is set against each vendor's published modalities, minimums, security reports and pricing.
- We write to you naming the vendors we think fit, and why.
- Nothing goes to any vendor until you reply to say which introductions you want.