Short answer. Four pricing models dominate annotation contracts, and each transfers a different risk. Per object is transparent when geometry counts are measurable and density varies. Per image or per video suits assets of stable complexity and penalises whoever guessed wrong about density. Per hour fits evolving guidelines and expert judgement but makes efficiency hard to compare. Project or subscription fits continuous pipelines at the cost of minimum commitments. Choose the model that mirrors the task's real cost driver.
Key takeaways
- The annotation pricing model decides who absorbs the variance when data turns out harder than the sample, and that variance is usually larger than the margin being negotiated.
- Per-object pricing is the most transparent model when the contract defines a billable object, covering occlusion, tracked objects and attributes, and the most disputed one when it does not.
- Per-image pricing prices the average and delivers the distribution; CVAT's published cost analysis assumes 23 objects per image across 100,000 images, or 2.3 million objects hidden inside one file count.
- Hourly pricing suits evolving guidelines and expert review, but is only auditable when expected units per hour on a reference task are agreed in advance.
- Mixed enterprise programmes work best under a master agreement with per-workstream pricing models, one acceptance definition and a change-control clause for re-pricing.
Why does the annotation pricing model matter more than the rate?
The pricing model decides who absorbs the cost variance when a dataset proves harder than the sample, and that variance is usually larger than the margin either side is arguing over.
A data annotation pricing model is the billing unit a contract uses to convert labelling work into an invoice, and it determines which party carries the risk of the data being denser, noisier or slower than expected.
A buyer who accepts per-image pricing on a dataset with wildly uneven object density has written the vendor an option. A vendor who accepts it has written the buyer one. Neither party usually notices until the second batch, so choose the model before negotiating the rate and before you compare data annotation vendor quotes in different units.
What are the four data annotation pricing models?
The four models are per object, per image or per video, per hour, and project or subscription. Each gives the buyer a different degree of control over what was billed.
| Pricing model | Best for | Main risk for buyer | Buyer control |
|---|---|---|---|
| Per object | Bounding boxes, polygons, keypoints, measurable entities | Dense assets become expensive fast | Very high when objects are easy to count |
| Per image / video | Stable complexity per file | You overpay on easy assets, or the vendor underprices dense ones and quality slips | High if asset complexity is consistent |
| Per hour | Complex, changing or expert tasks | Efficiency is hard to compare between vendors | Moderate; requires productivity metrics |
| Project / subscription | Continuous pipelines and reserved capacity | Minimum commitments, unused capacity | High if volume is predictable |
"Buyer control" is not about negotiating leverage; it is about whether you can verify what you were billed for. Objects can be counted; hours cannot, without productivity metrics agreed in advance. Rate ranges by modality are in the image, video and 3D/LiDAR annotation pricing guide.
When does per-object pricing work?
Per-object pricing works when a billable object has a definition that does not drift between the pilot and production.
Per-object pricing is a billing model in which the buyer pays a fixed rate for each annotated entity, such as a bounding box, polygon or keypoint, regardless of how many entities appear in a given image or frame. The definition of a billable object needs to answer three questions before the first invoice:
- Does a partially occluded object count? At what visible fraction?
- Does an object tracked across 200 video frames count as one object or 200?
- Do attributes count separately? A box with six attribute fields is not the same work as a bare box.
Answer these in the contract and per-object pricing is the cleanest model available. Leave them open and it becomes the most disputed one, because both parties have a defensible reading and the difference is the margin.
When does per-image or per-video pricing work?
Per-asset pricing works for homogeneous datasets where annotation time per file has low variance, such as product catalogue images, document scans or standardised inspection photos.
Per-asset pricing is a billing model in which the buyer pays a flat rate for each image, frame or video file, so the vendor absorbs the cost of dense files and the buyer overpays on sparse ones.
CVAT's published in-house cost analysis shows why a file count alone hides the workload: its worked example assumes an average of 23 objects per image across 100,000 images, producing roughly 2.3 million individual annotation objects. Two datasets with identical file counts can differ several-fold in cost.
Before accepting per-asset pricing, measure object density on a random sample of at least 200 assets and look at the spread, not the mean. If the 90th percentile is more than double the median, use per-object pricing instead. TELUS Digital's buyer guidance makes the same point: pricing varies widely by data type, and a credible estimate only follows a vendor review of the actual data.
When is hourly pricing the right model?
Hourly pricing is the honest choice when guidelines are still evolving, when expert judgement is required, or when the task cannot be standardised into countable units.
Hourly pricing is a billing model in which the buyer pays for annotator or reviewer time rather than for output units, which suits adjudication, taxonomy design, complex 3D scenes and exploratory labelling. Its weakness is comparability: two vendors quoting the same hourly rate can differ by a factor of two in output.
Mitigate it by agreeing productivity metrics up front: expected units per hour on a defined reference task, reported weekly, with a review trigger if actual output diverges materially. That makes an hourly contract auditable without turning it into a unit contract.
What does project or subscription pricing actually buy?
Project and subscription pricing buys reserved capacity rather than labels, and the discount it offers is payment for the forecasting risk the buyer takes on.
Subscription pricing is a billing model in which the buyer prepays for a committed volume or capacity over a fixed term in exchange for a lower unit rate. CVAT's published 2025 example illustrates the mechanism: 100,000 objects at $0.10 each is $10,000, while a prepaid subscription with a volume estimate for the same expected work is illustrated at $0.05 to $0.075 per object, roughly $5,000 to $7,500. This is not an industry price; it demonstrates that prepayment and commitment shift forecasting risk to the buyer and get paid for it. All figures quoted here are published third-party examples for illustrative scenarios, not industry averages and not Lifewood prices.
Accept a minimum commitment only after modelling expected utilisation honestly, including months when your ML team is retraining rather than labelling. Unused reserved capacity is the most common way a "cheaper" contract becomes the expensive one, a pattern that also runs through the in-house vs outsourced annotation cost comparison.
Which pricing model matches which annotation task?
Countable, stable tasks suit per-object pricing; uniform assets suit per-image pricing; temporal, 3D and expert work suit hourly or project pricing; and continuous pipelines suit subscription with volume tiers.
| Task | Recommended model | Why |
|---|---|---|
| 2D bounding boxes, stable ontology | Per object | Countable, verifiable, density-fair |
| Product catalogue classification | Per image | Uniform complexity, low variance |
| Video tracking with occlusion | Per hour or per project | Temporal work resists unit definition |
| 3D LiDAR cuboids | Per hour, per frame or project | Object density and point quality vary heavily |
| Medical or expert review | Per hour | Judgement time is the cost, not the count |
| RLHF preference ranking | Per task or per hour | Comparison quality depends on reviewer time |
| Continuous production pipeline | Subscription with volume tiers | Reserved capacity is the actual deliverable |
| Guideline design and calibration | Fixed fee | It is a project, not a production run |
The LiDAR row is where a flat per-object rate fails most expensively, which is why managed autonomous driving annotation is usually scoped per sequence or per frame.
How should mixed annotation workloads be priced?
Mixed programmes should run under a master agreement with a separate pricing model per workstream, a single acceptance definition, and a change-control clause for re-pricing.
Large enterprise programmes rarely contain one standardised task forever: a programme may start with image bounding boxes, add video QA, expand into 3D point clouds, and later require multilingual text or LLM evaluation. Forcing all of that into a single billing unit produces one of two outcomes: the buyer overpays on the parts that do not fit, or the vendor loses money on them and quality follows.
The change-control process should trigger on a new modality, a taxonomy revision, or a shift from pilot to production; a contract with no such clause forces both parties to pretend the work has not changed. Per-workstream pricing is more work to negotiate than a single rate card, and the only structure that survives two years. The top 10 large-scale AI data annotation and labelling companies are, for the most part, the vendors already used to operating this way.
How does Lifewood price annotation work?
Lifewood scopes pricing per project rather than publishing a universal rate card, because the four models apply differently to different workstreams inside the same programme.
What stays constant across workstreams is the acceptance standard: a 95%+ accuracy SLA with dual-layer human review, and below-threshold batches reworked at Lifewood's cost. That is what makes a mixed-model contract workable: the billing unit can vary by workstream; the definition of an accepted unit does not. The reasoning behind that threshold is set out in what accuracy standard to require from an annotation vendor.
Coverage across text, image, audio, video and 3D point-cloud work through 40+ delivery centres across 30+ countries in 100+ languages means a workstream can change its pricing model without changing supplier. Every AI data service Lifewood runs uses the same acceptance framework.