Almost everything written about data pricing is either enterprise folklore (“value-based pricing!”) or useless (“it depends”). Here’s the practical version, with real numbers from a live marketplace.
Decide the unit of sale first — it matters more than the number
Pricing isn’t “what does my data cost,” it’s “what does one purchase contain?” Three workable units:
- Per call (inline API): one lookup, one row-set, one answer. The unit for data that’s consumed incrementally — enrichment, quotes, lookups.
- Per download (dataset file): the whole artifact, bought once per snapshot. The unit for training sets, backtests, bulk analysis.
- Free (with a purpose): discovery, trust-building, or the free half of a freemium ladder (free daily aggregate → paid hourly detail).
The mistake to avoid: accidentally selling everything for the price of one
call. If a single $0.01 request can return your entire dataset, that is
your price. Cap the per-call response (max_response_kb is a seller knob —
1 KB to 5 MB) so granularity is a decision, not an accident, and sell bulk as
an explicit download listing instead.
Know the fee math before picking numbers
On datapoint.market the protocol fee is added on top of your price — you
receive your full P; the buyer pays P + max(10% × P, $0.005). The $0.005
floor covers the real on-chain settlement cost of a paid call, which creates
one structural rule: the floor dominates below $0.05. At P = $0.01 the
buyer pays $0.015 (fee is 50% of your price); at P = $0.05 and above, the
fee is a flat 10%. Full worked table: rate card.
Practical corollaries: don’t price API calls at $0.001 to seem cheap — the buyer’s all-in cost is dominated by the floor anyway; and for datasets, price well above the floor (almost anything ≥ $1) so the fee reads as a clean 10%.
Real price points, not theory
From the live catalog and demand board today: inline API lookups cluster at $0.005–$0.035 per call; small datasets sell for cents to a few dollars; and on the requested-data board a buyer has posted a $25/download willingness-to-pay for an assembled EU electricity-prices parquet. That last number is the most instructive: the underlying data is public — the $25 is for assembly and maintenance. Curation is the product more often than the data is.
Agent buyers change the psychology
Two agent-specific rules that don’t exist in human pricing:
- Price under the reflex thresholds. Buying agents run budget policies — per-call caps and confirm-above thresholds. A call priced at $0.02 gets bought autonomously; one at $0.50 may require a human confirmation. For high-volume lookup data, being comfortably inside the autonomous band beats extracting the last cent per call.
- Evaluation is free, so quality is pricing power. Agents compare schemas, samples, join cards, and reliability stats before paying. A DS-ready, fusion-ready listing with a verified match-rate claim can charge multiples of an opaque one — and the opaque one doesn’t even get seen in join-handle search.
A starting recipe
- Anchor on buyer alternatives: what would assembling this themselves cost (hours × rate), and what do adjacent listings charge?
- Lookups: start $0.01–$0.05; watch conversion; move within the autonomous band.
- Datasets: start at “one hour of the buyer’s time” ($20–$100 for professional-grade), more if refresh cadence is the value.
- Version snapshots (new slug or re-publish) so recurring value is recurring revenue — buyers re-buy the new snapshot.
- Reprice freely:
update_endpoint(price_usd=…)is one call, and quotes are always pre-purchase, so nobody gets surprised.
Pricing data isn’t a one-time decision — it’s a knob you now hold on a live market with per-call telemetry. Start: sell a dataset · rate card · what buyers are asking for