How do AI companies price company data?

AI data companies price a company’s records as one package, not by the unit. Calculators start from headcount, years and country; the firm price adds what review finds about your systems, record types, connections and rarity, plus the license terms and the provider’s own demand. As of October 2026, none publishes a per-employee or per-gigabyte rate for business records.

What do AI companies look at when they price company data?

Mostly headcount, years of history, the systems and record types you hold, and how connected and rare the records are, though each provider names a different subset. As of October 2026:

Input	What providers say
Headcount	Replay: “Headcount is a proxy”. Troveo: value “scales with your systems, team size, and years of history” (Troveo).
Years retained	Avelence’s calculator asks for “retained years” (Avelence). Troveo: “Five years beats five months”.
Systems	Replay counts “which apps you run on and how much data sits in each” (Replay); Mercor’s pay depends on the tools connected (Mercor).
Record types	Handshake AI names “the underlying data sources” (Handshake AI). Sunset’s CEO told Forbes that healthcare and finance data command a premium (Forbes).
Volume	micro1 lists “dataset size and volume” (micro1). Polyshares disagrees: “The value is not in volume”.
Connectedness	Appen prices on “the number and connectedness of systems” (Appen). Sunset’s “data richness” is “internal traceability and cross-platform linkages” (Forbes).
Rarity	micro1: “the uniqueness of the operational knowledge”. License My Data: “how distinctive the operational knowledge is”. Telegraph Lab (affiliated) reviews “originality” (Telegraph Lab).
Quality and rights	Nyne: final value depends on “data quality, volume, rights, and the outcome of due diligence”.
Exclusivity	License My Data describes two ways to be paid for the same material: once, upfront, for an exclusive buyout, or on each placement of a non-exclusive license (License My Data).
Scope and cadence	Scale AI’s range scales “with cadence” (Scale AI); micro1’s $500K+ tier covers “ongoing participation across multiple teams” (micro1).
Demand	micro1 says it “currently prioritize[s] US companies, followed by other Western markets” (micro1).

Size and history are where every calculator starts, but providers say they do not settle the price. Troveo says “a smaller company with rare, specialized workflows can be worth more than a larger generic one”. Why connected, specialized records interest AI buyers is covered in why AI companies want our emails.

Which inputs does a calculator use, and which wait for review?

A calculator uses the inputs a stranger can answer in a minute; everything else waits until a provider sees your records. As of October 2026:

Calculator	Inputs it asks for
Avelence	Employees (20 to 10,000), retained years (1 to 50), headquarters region (source)
Handshake AI	Employees, years of operation, country: USA, Canada, Europe or Other (source)
micro1	Employees, years of data, number of business entities, share of data in English, headquarters, whether regulated data is present (source)
Telegraph Lab (affiliated)	Employees (20 to 200+), years in business (3 to 20+), location (source)
Troveo	Ten questions, including team size, years of history and the languages your data uses (source)

The year inputs ask different questions. Avelence asks how many years you have retained; Handshake AI asks how long you have operated, and Telegraph Lab how long you have been in business. For example, a company founded in 2004 that has kept email only since 2016 has 22 years in business but 10 retained years, and only the retained years describe what a buyer could license.

Avelence explains why its ranges are wide: each spans “the lowest lower estimate and highest upper estimate across three modelled licensing scenarios”. It treats location with the same caution. Its model gives UK and European companies a lower range than US ones, but Avelence says that does not establish a fixed regional discount in a real deal.

The rest of the inputs (record types, connections, rarity, rights, exclusivity) are judged in review. micro1’s estimator says the final offer depends on “data sources, data size, data quality, access terms, and due diligence”. How offers work covers the step from an estimate to a firm offer, and the guide the process around it.

Per person, per year, per gigabyte?

None of those, for business records. Troveo’s September 2026 pricing guide says “Buyers don’t price data by the gigabyte, and they rarely start from a rate card”, and as of October 2026 no provider in the Data Licensing Report directory publishes a unit rate for company email, chat, documents or software records.

• Per person. Headcount is an input, not a rate. Replay’s tiers group companies into bands of 20–50, 50–250 and 250+ employees, and how much is my company’s data worth? shows that published outputs do not rise in step with headcount.
• Per year. Years of history raise an estimate but are not priced one by one. Pay by period exists only where data keeps flowing: Scale AI says it designs partnerships so there is “opportunity for recurring revenue”, and Replay says new data “follows the same path, and continues to pay you” (Replay).
• Per gigabyte. No provider publishes one. Mercor and micro1 list volume as one factor among several, and Polyshares says value lies in “reasoning that was captured while the work happened” rather than volume.

Unit prices do exist for other kinds of data, which is where owners may have seen them:

Unit	Who publishes or reported it	What it applies to
Per page	$6 a page, up to $30K, Business Insider reported of a separate Handshake AI program (source)	Work documents an individual owns or may share, not an employer’s records
Per record	FileYield’s guide: $0.01 to $10 for raw data, up to $1,000+ for annotated data (source)	Generic datasets; FileYield gives no figure for company email or chat
Per codebase	HUD: a $5,000 baseline per codebase, rising with grading (source)	Source code
Per minute	$0.75 to $3 a minute for footage, the New York Post reported of Troveo’s video program in 2025 (source)	Video

If it’s worth this much to them, why would they tell me what it’s worth?

They don’t tell you what your records are worth. They publish what their program might pay a company with your headcount and history, before seeing anything, and they label it that way: Telegraph Lab’s calculator is “an illustration, so an actual offer may be lower, higher, or unavailable” (Telegraph Lab), and Handshake AI’s estimate “is not an offer to purchase your data” (Handshake AI).

Three things stay private, as of October 2026:

• What deals closed at. Handshake AI says everything in a deal is “scoped upfront, protected under NDA” (Handshake AI). Troveo’s guide says “most AI data deals are confidential, so the visible numbers are the edge of the market rather than its average” (Troveo).
• What the provider earns. Polyshares does not publish what it earns when it resupplies the data, Troveo does not publish its share of each sale, and Avelence does not publish what its data partners pay it (Polyshares, Troveo, Avelence).
• What its own buyers will pay. Most providers covered here license the records onward to AI labs they do not name, as their profiles note.

So the price for your records comes out when two or more providers price the same records on the same terms. One public case of several bidders pricing the same records, Spirit Airlines’ bankruptcy sale, is set out in how offers work. If you already hold a number, is this a good offer? covers how to judge it, and upfront payment or revenue share explains why the same records can be priced as one payment or as a share of later sales.