Introducers: Referral Programs of AI Data Buyers Compared

Introducers are advisers, investors, liquidators and other third parties who connect a company holding data with a program that licenses it for AI training, and many programs offer them a referral fee. As of 2026-11-09, 26 programs and intermediaries publish a referral offer for introducing a company, from Handshake AI's flat USD 10,000 per referred company to uncapped percentages such as Pangea.ai's 15 %, and 13 more publish none. Most fees are earned only after the company signs, delivers or is paid, and micro1 and Mode reserve the right to take a fee back.

Every amount below is the program's own claim, checked 2026-11-09, and rows run alphabetically, never by fee. This comparison is part of the independent reference on AI training data licensing: DataPacts has joined no referral program, carries no referral link and takes no cut of any sale.

Each figure carries its source and check date, and how we source, date and check every figure is set out in the methodology.

What is an introducer in AI data licensing?

An introducer is a person or firm that refers a company that holds data to a program or intermediary and is paid, if at all, only when a stated event happens. Introducers typically come from five groups:

• Accountants and fractional CFOs.
• Business brokers and M&A advisers.
• Investors winding down portfolio companies.
• Liquidators, trustees and administrators.
• IT providers and industry networks.

The introducer does not sell the data, sign the license or receive the company's payment. Regulr, a sell-side adviser, names business brokers, M&A advisors and fractional CFOs as its partners, on terms agreed in writing before introductions (checked 2026-11-09). People who refer contributors for paid AI training work take part in a different kind of referral, covered in the FAQ below.

Every referral offer on record, compared

Twenty-six programs and intermediaries publish a referral offer for introducing a company, and 13 more publish none. The table below compares each offer as stated on the program's own page, alphabetically, as of 2026-11-09. Three points help read it:

• "Not published" means the program says referrals exist but states no amount.
• Broker and Market rows are intermediaries: they place a holder's data with buyers and do not acquire it.
• Every cell is the program's own claim; no figure comes from third-party sites.

No referral offer is published, as checked on 2026-11-09, by Alpheva AI, Avelence (sourcing partners page without terms), DataSupply.ai (partner page unavailable), Endless, Getty Images, Labelbox Prism, Magma, Ooak Data, PIXTA AI (rates by separate application), Polyshares, SuperAnnotate, Sureform and Troveo.

How referral fees are calculated: flat, capped, tiered, greater-of and shared

Referral offers follow eight fee models, and the model matters more than the headline, because a percentage of a large deal can exceed any flat fee while an "up to" figure promises no particular amount. The table groups the 26 offers of the comparison above by model, as of 2026-11-09.

The same deal can produce different fees under these models. micro1 states, checked 2026-11-09, a flat USD 50,000 for enterprise deals up to USD 700,000 and "7.5% of the deal value for deals above $700,000", with a defunct-company schedule of USD 75,000 up to USD 1 million and 7.5 % above. Replay states the greater of USD 10,000 or 10 % of the company's upfront licensing fee, so its fee follows only the upfront part of a deal. Pewter states, as of 2026-11-09, 10 % of the licensing payment with a maximum of USD 100,000 per business. DataPacts computes no averages or sample fees from these schedules, because each applies to a different deal definition.

Referral fees are one column of a program's terms. The company-side figures, from what a program says it pays the data holder to its exclusivity and eligibility rules, sit in the table of AI data partnership payouts and program terms compared, each dated.

Conditions that decide whether an introducer is paid: trigger, attribution and clawback

Most programs pay an introducer only after the referred company signs, delivers or is paid, and several can take the fee back. The table compares the conditions of the 16 programs that publish them, alphabetically, as of 2026-11-09.

Four conditions decide whether an introduction ends in a fee:

• Trigger: the event that earns the fee, from a signature (Scale AI) and deal close (Handshake AI) to accepted data (micro1) or money received by the program first (Mode, Pewter).
• Attribution: a registration link, first referrer by date, or an introduction made before the company applies on its own, inside windows from 90 days (SourceX) to 12 months (AIData.Broker, HUMXN, Marin Data, Pewter).
• Clawback and holdback: micro1's terms allow a holdback and a clawback, and Mode can reclaim a fee within 12 months of payment, or without time limit under clauses (b) and (d).
• Written terms: Aba Compute, McCarthy's AI and Regulr agree terms in writing before the introduction; Specific Labs confirms them in writing before payout.

Introducer triggers run behind the company's own payment events, which programs set separately. The upfront, milestone, revenue-share and clawback terms that govern the company itself are set out in the guide to when and how AI data buyers pay.

Who pays the introducer, and whether the company receives less

Programs that say who bears the fee state it as paid on top of the company's payment or out of the intermediary's own fee, never as a deduction from the company; 22 of the 26 offers say nothing about it. The statements on record, checked 2026-11-09, fall into four groups:

• On top of the company's payment: Replay and Pangea.ai.
• Out of the intermediary's own fee: Pewter and AIData.Broker.
• Paid only after money has moved: among them Mode (after it receives the customer payment), Patronus AI (after the owner is paid), Pewter (after Pewter is paid) and SourceX (after it pays the company).
• Not stated: every other row of the comparison.

Some intermediaries are themselves paid referral fees by buyers, which is a conflict a company should know about. aitrainer.work states that "some buyers pay us a fee when someone we sent them completes a deal", and OpenTrain AI states that buyers that credit a referral may pay it a reward; neither names an amount. The Corner describes itself as an independent referral partner of micro1 and advertises "Refer companies. Earn $50K+ each", checked 2026-11-09, while the amount micro1 may pay The Corner is not stated. A company dealing with any intermediary can ask in writing which buyer pays it and how much. The directory of brokers and sell-side advisers lists their fees as published.

Contradictions and third-party figures in referral offers

Several programs state their referral offer differently on two of their own pages, and DataPacts shows both versions rather than choosing one. The table sets the versions side by side, as of 2026-11-09.

The Mercor row matters most to an introducer: on eight listings the stated referral base is higher than the smallest deal the listing describes, as Mercor's own feed showed on 2026-11-09, so the base cannot be read as a share of the deal. Referral figures that circulate on third-party sites but do not appear on the programs' own pages are left out of every table on this page. An earlier micro1 forum figure from June 2026 was replaced by the current schedule and is not shown as current.

Disclosure, tax forms and estate rules for introducers, in brief

An introducer who recommends a program and is paid by it has a material connection that US endorsement rules require to be "disclosed clearly and conspicuously" (16 CFR 255.5(a), eCFR text of 2026-11-05), and micro1's Referral Program Terms require the same disclosure (s.5.4). Under the IRS instructions for Forms 1099-MISC and 1099-NEC, a US introducer paid USD 2,000 or more for a tax year beginning after 2025 receives a Form 1099-NEC, and the threshold may be adjusted for inflation from 2027. Pewter and HUMXN require a W-9 or equivalent before paying, and micro1 may require a W-9 or W-8.

In a US bankruptcy, estate professionals may not share their compensation (11 U.S.C. 504, with exceptions in 504(b) and (c)). Knowingly and fraudulently giving or receiving money for acting or forbearing to act in a case under title 11 is a crime punishable by a fine, up to 5 years in prison, or both (18 U.S.C. 152(6)), and no case applying it to data referral fees is on record. This is information, not legal advice, and the professional rules sit on the page on referral fees for advisers.

Benefits, risks, downsides and alternatives of introducing a company, including not referring

Introducing a company can earn a fee and help a client reach a buyer, but the fee depends on events the introducer does not control and creates a conflict that must be disclosed. The lists below rest on the programs' own statements, checked 2026-11-09.

Benefits

• A stated fee for an introduction, as each program claims it in the comparison above.
• Clients reach programs that publish their terms.
• Some fees come on top of the company's payment, as Replay and Pangea.ai state.

Risks

• No fee if attribution fails or the deal is never signed: micro1 attributes only through a unique link, and SourceX pays only after it pays the company.
• Clawback after payment under micro1's terms and Mode's agreement.
• Disclosure, professional-conduct and bankruptcy rules, including 11 U.S.C. 504 and 18 U.S.C. 152(6) around estate sales.

Downsides

• Long waits: Mode pays after its own customer pays, and Pewter owes no fee if Pewter itself is not paid within 36 months.
• A fee-paid recommendation can weaken a client's trust in the adviser.

Alternatives

• Introduce the company without taking a fee.
• Tell the client to apply directly to several programs.
• Use the DataPacts introducer form, which charges the company nothing and takes no cut.
• Advise the client not to sell or license at all.

Not referring is a complete answer for many clients, and the case for and against is weighed on the page on whether to sell or license data to AI companies at all.

How this comparison is built: alphabetical, dated, no referral links

Rows are ordered alphabetically and include every program whose referral terms DataPacts found, whether or not it would ever pay DataPacts. The sources are the programs' own referral pages, terms and agreements, checked 2026-11-09, as the methodology describes.

No row is ranked, badged or filtered by fee, and no amount is averaged, rounded or merged. Where a program states two versions, both stay visible.

DataPacts has joined no referral program, carries no referral code and takes no cut of any sale; if that changes, it is disclosed on the page explaining how DataPacts is funded and as a note here.

Where referral terms meet program terms, payment and the seller path

Referral terms are one column of a program; the pages below cover the rest of it.

• How company data partnership programs take in, review and pay for company records.
• Who buys AI training data, every buyer compared.
• How administrators and liquidators sell data assets in insolvency.
• Selling data from a shut-down or defunct company.
• Checking a message: is that data licensing email legitimate.

Frequently asked questions

Which AI data programs pay referral fees?

As of 2026-11-09, 26 programs and intermediaries publish a referral offer for introducing a company, and 13 publish none. The offers range across eight fee models, from flat sums (Handshake AI, McCarthy's AI) to greater-of formulas (Mode, Replay, SourceX); none is ranked here.

Does the company receive less when an introducer is paid?

Replay and Pangea.ai state the fee is paid on top of the company's payment, and Pewter and AIData.Broker pay it from their own fee. The other programs do not say, so a company can ask in writing before it signs.

Can an accountant accept a referral fee?

It depends on the accountant's professional rules and on disclosure. US endorsement rules require a paid connection to be disclosed clearly and conspicuously (16 CFR 255.5(a)), and the professional and tax rules are covered on the page on referral fees for advisers.

Do AI training job referral bonuses work the same way?

No: contributor referral programs, such as the Handshake AI Fellowship and Mercor expert referrals, pay on a referred worker's hours, not on a company's license. Those programs are compared on the page on AI training jobs and platforms, for contributors.

Does DataPacts earn referral fees from these programs?

No: DataPacts has joined no referral program, carries no referral links and takes no cut of any sale. Inclusion and order on this page never depend on fees, and joining any program in future would be disclosed on the disclosure page and on this page.