Per-Datapoint Pricing: Examples & Companies

What is it

Per-datapoint pricing is a billing unit where each individual data measurement or signal ingested is metered — common in cloud cost intelligence and ML evaluation platforms.

The unit sits across two pricing cultures that draw the line differently. In cloud FinOps platforms, a “datapoint” is a cost signal — a metered reading pulled from an AWS, GCP, or Azure billing API, or from an AI vendor like OpenAI or Anthropic. Finout ingests these signals into its MegaBill unified cost layer; Vantage turns them into cost reports, budgets, and anomaly alerts across 20+ providers. The value scales with how many signals get tracked, so “monitored cloud spend” — a buyer-legible proxy for datapoint volume — gates their pricing tiers.

In ML evaluation and training-data platforms, a datapoint is a stored, structured example. For Humanloop, it was one model completion enriched with inputs, human feedback, and evaluator judgments — the primary 2023 billing unit before it switched to the simpler “log” meter in 2024. Scale AI meters labeled records and annotation tasks — the same per-datapoint logic under a different name, applied to training data rather than observability.

The datapoint’s appeal is granularity: cost tracks the volume of data processed, not seat count or elapsed time. Its challenge is definitional — what counts as one datapoint is not self-evident, which makes the boundary a value-metric decision.

How it works

Per-datapoint billing typically operates in one of three modes: a quota-bundled subscription where each tier includes a fixed pool of datapoints per month, a committed contract that agrees an annual volume of data units, or a spend-gated proxy where the plan tier is set by the volume of cost signals tracked (expressed as monitored cloud spend).

Humanloop’s 2023 pricing shows the quota model most clearly. The Team plan bundled 100,000 datapoints/month at $1,000 (~$0.01 per datapoint at the included rate), overage at $4 per 1,000 — a 60% volume discount over the Starter overage of $10 per 1,000. Annual billing cut the platform fee 30%.

Unit math (Humanloop 2023): Monthly bill = plan fee + max(0, datapoints used − quota) × overage rate per 1,000 ÷ 1,000. A Team-tier customer processing 250,000 datapoints in a month would pay $1,000 + (150,000 × $4 / 1,000) = $1,600.

Scale AI applies the same per-unit logic at far larger volumes with no public rate card. Third parties cite indicative self-serve figures near 2 cents per image and 6 cents per annotation on the Data Engine, with the average enterprise contract reported near $93,000/year — implying large committed volumes negotiated per deal. Its free trial (first 1,000 labeling units, first 10,000 images) is the only publicly defined threshold. The introduction to usage-based pricing covers how committed-volume metering compares with pay-as-you-go.

Finout and Vantage abstract the datapoint meter behind spend tiers. Vantage’s free Starter covers up to $2,500 of monitored spend (3 users, 6-month retention); a team monitoring ~$6,000/month needs Pro (up to $7,500). The real variable is the count of cost signals ingested from AWS, Snowflake, OpenAI and others — spend is its proxy. Finout does the same: third parties report its Business tier near ~$1,000/month covering up to ~$500,000 of annual cloud spend, a scale-of-estate proxy for the datapoint volume its MegaBill layer processes.

Companies using this

Four corpus companies meter on datapoints across two categories: cloud cost intelligence (Finout, Vantage) and ML evaluation or training data (Humanloop, Scale AI). Each one’s value scales with the volume of structured data signals it processes, which makes the datapoint a natural meter even when the pricing page names it differently.

Explore this theme in the knowledge graph →