Handshake AI data licensing: a seller's guide to its data partnerships program

What Handshake AI's data partnerships program buys, its published thresholds, price figures, calculator, timeline and process, what it leaves unpublished, and who it fits.

Travis Wheeler

By Travis Wheeler

Co-Founder, Commercial and Partnerships

Published

Three different things carry the Handshake AI name, and a search for it returns mostly the wrong one. Handshake is the student and early-career job network, founded in 2014 and now claiming 30 million knowledge workers and 1 million employers. Handshake AI is the same company's AI division, which supplies training data to AI labs and recruits paid fellows to produce it; its own page claims 188k+ fellows and $500M+ in payouts. Inside that division sits the data partnerships program, which buys operational data from businesses and licenses it to AI developers. This page is about the third thing. The fellowship listings are not for a company with records to license.

Everything below comes from Handshake's public pages and press coverage as read on November 12, 2026. Where Handshake has not published something, this page says so. Vaquero is a seller-side broker and is independent of Handshake.

What is the data partnerships program, and where does it sit inside Handshake?

Handshake's FAQ describes it in one sentence: "Handshake AI helps companies earn revenue by licensing their internal business data." The data is used to "train and evaluate AI models so they can better understand how real work gets done." The announcement blog post, dated October 20, 2026, says Handshake wants data on how businesses run, how decisions get made and how teams use tools. Asked which labs it works with, the FAQ answers "all of the leading frontier AI labs developing the next generation of models." Its enterprise page claims "8/8" of the largest frontier labs train with Handshake AI datasets. No lab is named on the program pages.

The program is new and the parent is not. Sacra dates Handshake AI's launch to January 2025. TechCrunch, reporting Handshake's acquisition of Cleanlab on January 28, 2026, said the human data business had launched about a year earlier and was forecast to end 2025 at $300 million in annualized revenue run rate. By April 2026, Dealroom, citing The Information, put AI training gross annualized revenue at nearly $1B, up from $550M in January 2026, with contractor costs at 60 to 70 percent of gross. The parent raised a $200 million Series F at a $3.5 billion valuation in January 2022, the last priced round we found. Its enterprise page claims 200+ in-house researchers and engineers.

The data partnerships team is a small, recent addition to that business. A Handshake job posting dated August 29, 2026 describes it as "a new, adjacent motion: going directly to companies to license their proprietary data," names "workflow data, code, support transcripts, and more" as the targets, and says a hire would be "one of the first 5-10 reps on the floor."

We found no coverage of the data partnerships program itself in TechCrunch, The Information, Fortune, Bloomberg or Axios as of November 12, 2026. Press on Handshake AI covers the fellowship business and its revenue. That dispute concerns individual fellows, not company data deals.

Who does it buy from, and what are the thresholds?

The license page: "We're looking for companies with 20+ full time employees & 3+ years in operation." The FAQ softens it: companies with 20 or more employees and three or more years of operating history are "typically more likely to be a fit," and eligibility "depends on the type, volume, and usability of the data." The calculator defaults to 20 employees and its years slider starts at 3.

Geography: the calculator's country options are USA, Canada, Europe and Other. No other restriction is stated. Language requirements are not published.

The program buys from operating companies. There is no published track for closed or winding-down companies. The one exception is by project, not company: a Gaming tab in the FAQ says Handshake is interested in "released titles, games in development, and canceled projects."

What records and verticals does it name?

Handshake AI is more specific than most buyers about record types by industry. The program page's industry tiles name:

Industry Records named
Finance and insurance Financial models, process documentation, underwriting workflows
Hardware and engineering Design documentation, specifications, test reports
Healthcare Operating procedures, training materials, administrative workflows
Legal Research workflows, process documentation, reusable templates
Logistics and supply chain Procurement workflows, shipping records, dispatch logs
Manufacturing Production records, quality reports, work instructions
Pharmaceuticals and life sciences Research documentation, lab protocols, development workflows
Professional services Project history, presentations, process documentation
Real estate Property documentation, market research, operating workflows
Retail and e-commerce Product catalogs, inventory workflows, merchandising plans
Software Technical documentation, issue histories, development workflows
Gaming (FAQ only) Game source code and development data, with supporting graphics, audio and design assets

Source for the eleven tiles: the program page. Source for gaming: its FAQ, which says "The focus is game source code and development data." An "Other" option asks you to describe your business and the records you hold. The FAQ's general description is "documents, communications, and records of business processes."

Systems list, and what AutoCAD at the top implies

The license page lists "Examples of sources we buy" in this order: AutoCAD, Dropbox, Figma, Gmail, Google Drive, Outlook, Monday.com, NetSuite, Notion, Paychex, Quickbooks, ServiceTitan, Slack, and "+1K more." The list is alphabetical, so AutoCAD leads partly by accident. What matters is that it is on the list at all. Together with the hardware and engineering tile, it makes Handshake AI one of two buyers we track that name CAD records on a public page; the other is AfterQuery. Our inference: CAD-heavy manufacturers and engineering firms have a stronger case here than at buyers whose pages only mention email and chat. Figma, ServiceTitan, NetSuite and Paychex on the same list mean design, field service, ERP and payroll exports count too.

What does it publish about price?

Three figures appear on Handshake AI's pages. They describe different things and should not be read as one range.

Figure Where Wording
$100K to $4M+ Program page badge, license page, blog The badge reads "Earn $100K to $4M+." The blog says "Qualified partnerships typically range from $100K to $4M depending on the data."
Up to $100,000, rising to $400,000 to $800,000+ Calculator on the program page At the defaults of 20 employees, 5 tools and 3 years: "Your data could be worth up to $100,000." The page source sets the sliders at 0 to 150 employees, 0 to 16 tools and 3 to 10 years; the widest output they can produce is $400,000 to $800,000+.
$250M paid out to partners License and referral pages Shown beside "1M+ Handshake employers" and "1K+ Partner companies." Handshake does not say what "partners" counts.

The license page carries the disclaimer: "This is an indicative estimate only and is not an offer to purchase your data." Read the headline range the same way. For a 20-person company with three years of records, the calculator's own number is the one to plan around.

On the $250M figure, our inference: a program that described itself in August 2026 as a new motion with its first 5 to 10 reps, and announced itself publicly on October 20, 2026, is unlikely to have paid $250 million to companies for records. The fellowship page separately claims $500M+ in payouts to fellows. We do not treat $250M as a measure of company data deals.

How the calculator computes its output is not published. The program page also runs in more than one version: the copy we loaded in a browser on November 12 had the headline "Operational data is the future. Get paid for it." and no calculator or tiles.

How does intake work?

Handshake AI publishes a five-step process on the program page:

  1. Start a conversation. "Tell us about your business and the questions you're considering." The license form asks for name, email, phone, company name, website and country.
  2. Explore the opportunity together.
  3. Agree on scope and terms.
  4. Prepare the data.
  5. Complete delivery and payment.

The FAQ fills in the steps: review the intake form for fit, schedule a conversation, assess the data's value, agree scope and terms, coordinate delivery and preparation. "Your agreement outlines the approved uses, de-identification requirements, and payment milestones." The second intake page, for referrals, says "Our team directly reaches out within 1-2 days" and then "We scope their tools and agree to terms."

On timing, the FAQ is specific: "we can provide an initial review and pricing within 12–24 hours" once it has the details to assess the data, and "the full process typically takes 7–14 business days to finalize your transaction." No other buyer we track publishes a timeline that short. Hold Handshake to it in the agreement.

On scoping: "You determine exactly what is shared." Scoping means defining "which records are included and what sensitive information needs to be excluded" and reviewing "who may receive the data, how it may be used, and the safeguards required." The blog adds that everything is "scoped upfront, protected under NDA, and de-identified before use." The NDA appears only in the blog; the program and license pages do not use the word.

What are the license terms?

Published: the structure is a license. "Your company retains ownership of its data, and the scope, approved uses, protections, and compensation are agreed upon before any data is transferred." Payment follows "payment milestones" set in the agreement.

Not published: exclusivity, term length, what the payment milestones are and what triggers them, retention and deletion of raw data after use, who the downstream recipients are, and the agreement text. The license page says "Examples of sources we buy" while the FAQ says "licensing," so read the agreement for which one it is.

These are the terms that decide whether a deal is good. Several buyers in this market make exclusivity the default, and some ask for it in perpetuity. Ask for the draft agreement before scoping and read for exclusivity, any right of first refusal, the payment trigger, the deletion schedule, and what survives after delivery.

De-identification and who prepares the data

Three sources, read together. The license page: "All data is scoped, de-identified, and confidential." The program page tells the seller to "Understand who will access and prepare the data, including any required de-identification," which leaves open who does it. The FAQ says "our team handles the technical work to minimize the lift for your team."

Handshake's privacy policy, last updated July 3, 2026, defines a Data Partner as a business that shares information "to contribute to improve AI models" and says that before data is made available to AI labs "we apply technical and operational processes" "intended to identify and remove or obscure information that may directly identify individuals." That is the clearest statement that Handshake performs de-identification, and it sits in the privacy policy, not on the program pages.

One more data point. On the Latent Space podcast episode published November 9, 2026, a Handshake speaker described acquiring a bank's files and de-identifying everything as "really hard to do" and "capital intensive," and said labs cannot get scalable data "from, you know, buying company data and de-IDing it," so Handshake has experts generate it. That is the parent describing its main product. The data partnerships program is the smaller line.

Settle it in writing: who de-identifies, whether it happens before or after transfer, what the agreement's "de-identification requirements" are, and who carries liability if an identifier gets through.

How Handshake AI differs from other buyers

  1. It names engineering and CAD records. AutoCAD is on the license page's systems list, and the hardware and engineering tile asks for design documentation, specifications and test reports. Most buyers' pages list email, chat and documents only.
  2. Its calculator has a "number of tools" input, with a slider from 0 to 16 and a default of 5, alongside employees and years. It is the only buyer calculator we have seen with that input, and it matches what buyers say privately: more systems with history means more value.
  3. It publishes a floor and a timeline: 20 or more full-time employees and three or more years, pricing within 12 to 24 hours and 7 to 14 business days to close. Polyshares and AfterQuery publish no floor, micro1 publishes 30, and no buyer we track publishes a shorter timeline.
  4. Its intake offers three regions plus Other: USA, Canada, Europe. Several buyers state US-only or say nothing.
  5. It sits inside a company with a $3.5 billion valuation at its last priced round and an AI data business reported at nearly $1B gross annualized revenue. Payment risk is lower than at a venture-stage buyer. Attention risk is higher: the program is a small new team inside a large one.
  6. It has a gaming track that buys game source code, including canceled projects. No other buyer we track asks for that.

What does Handshake AI not say?

  • Exclusivity and term
  • The payment milestones and what triggers each
  • Retention and deletion of raw data after use
  • Who the downstream recipients are, by name
  • Language requirements
  • Whether it buys from closed companies, outside the gaming FAQ
  • How the calculator computes its output
  • What the $250M "paid out to partners" figure counts

None of these is unusual for this market. Most buyers publish a range and a process and leave the contract terms to the agreement. The published pages are not enough to decide on. The agreement is.

Who is Handshake AI a fit for?

A good fit, on the published material: US, Canadian or European operating companies with 20 or more full-time employees, three or more years of records, and operations or engineering records in the systems it lists. Manufacturers, engineering firms, software companies, logistics companies and game studios match its tiles and FAQ directly.

A weaker fit: companies under 20 staff, closed companies, and companies whose records are mostly email and chat without process documentation. Those sellers have other buyers.

Alternatives to Handshake AI

  • Polyshares. No published headcount floor, prices on the length and texture of the record, exclusive by default per its data-handling page.
  • micro1. Publishes a 30-plus full-time employee floor for operating companies and a separate program for closed companies.
  • Telegraph Lab. Publishes a $100K to $4M range, a 20-employee and three-year floor, per-system guides and an estimator; terms left to agreement.
  • AfterQuery. Program announced October 23, 2026; accepts closed companies; names CAD records; no price anchors published.
  • Mercor. Ingests through connectors and, per its CEO's public post, targets firms with 100 or more employees.

Find out where your records land

Enter your office headcount and years of records on the Vaquero estimator for a range built from published deals, not one buyer's headline. A person replies within one business day with a view on which buyers are paying for records like yours. No files, no names, nothing to sign, and nothing to pay at any step. The buyer pays Vaquero's fee at closing.

Sources

All read November 12, 2026.

Questions and answers

Is this the same Handshake as the student job site?

Same company, different program. Handshake runs the college recruiting network. Handshake AI is its division that supplies expert-generated training data to AI labs and recruits paid fellows to produce it. The data partnerships program sits inside that division and licenses company records. The fellowship job listings are for individuals, not for companies with records to license.

Is the Handshake AI data partnerships program legitimate?

It is a published program from a company founded in 2014 and last valued at $3.5 billion in January 2022, with a program page, a license page, a calculator, a five-step process and an FAQ. The program itself is new: a job posting from August 2026 called it a new, adjacent motion with the first 5 to 10 reps, and the announcement blog post is dated October 20, 2026. Exclusivity, term and retention are not published. Get them in writing.

My company has fewer than 20 staff. Will Handshake AI buy?

Its license page says it is looking for companies with 20 or more full-time employees and three or more years in operation. The FAQ softens that to companies that are typically more likely to be a fit and says eligibility depends on the type, volume and usability of the data. Under 20, expect a harder conversation here and look at buyers with no published floor.

How long does the Handshake AI process take?

Its FAQ says that once it has the details needed to assess your data it can provide an initial review and pricing within 12 to 24 hours, and that the full process typically takes 7 to 14 business days to finalize the transaction. That is the shortest published timeline of any buyer we track. Treat it as a target, not a commitment, until it appears in your agreement.

Does Handshake AI pay $100K to $4M?

That is its headline range. Its blog says qualified partnerships typically range from $100K to $4M depending on the data. Its own calculator reads up to $100,000 at the default inputs of 20 employees, five tools and three years, and tops out at $400,000 to $800,000+ with every slider at maximum. The license page calls the calculator an indicative estimate and not an offer.

Who de-identifies the records?

The license page says all data is scoped, de-identified and confidential. The FAQ says Handshake's team handles the technical work and that the agreement sets de-identification requirements. The privacy policy says Handshake applies processes intended to remove or obscure information that may directly identify individuals before data is made available. Confirm in the agreement who does it, when, and who carries liability if an identifier gets through.

Can I get quotes from Handshake AI and other buyers at the same time?

Yes, until you sign. Handshake AI publishes no exclusivity term, so read the draft agreement for one before signing anything that restricts you. A seller-side broker runs quotes from several buyers in parallel and negotiates with one.

Does using a broker change the price Handshake AI pays?

The license fee is the amount paid to you. A broker's value is the second and third quote, a counsel read of your client contracts before scope is priced, and negotiated terms. The buyer pays Vaquero's fee at closing. You owe nothing.

What could your records be worth?

See my range