Can a marketing agency sell its data to AI companies?

Usually not for client work, which client contracts, NDAs and ad-platform terms tie to the client. The agency’s own operating records (pitches, processes, templates, timesheets) are a separate question; as of October 2026, five providers name marketing agencies or their records.

What does a marketing agency keep that AI buyers ask for?

A record of how marketers turn a brief into a plan, creative and results, and how the client reacted at each step. Scale AI explains the demand this way: “AI companies want to improve their models for certain corporate personas such as marketers or financiers”. Mercor’s professional-services examples are “Client decks, SOWs, project files”, and FileYield’s marketing category lists “ad performance, SEO rankings, attribution, influencer metrics, and campaign datasets” (all as of October 2026).

Most of that is made for, or from, a client. Polyshares publishes export guides for Asana, Monday.com, ClickUp, Figma and HubSpot, and in an agency each of those tools holds both kinds of record.

What does one campaign leave behind?

Records on both sides of the client line. For example, a product launch run by a hypothetical 40-person agency:

Stage	Records	Usually lives in	What decides whether it could be licensed
Brief	The client’s brief, launch dates, sales figures	Email, shared drive	The client’s confidentiality terms
Strategy and media plan	Strategy deck, audience work, media plan and budget	Slides, spreadsheets	The ownership clause; built on the client’s information
Creative development	Concepts, rejected routes, working files, final assets	Figma, design files, shared drive	The ownership clause; client information inside the files
Review rounds	The client’s comments and sign-offs	Email, proofing tool, project board	The client’s confidentiality terms
Launch and optimization	Campaign settings, audiences, performance data	The ad platforms, reporting tools	Platform terms; the client’s customer lists
Reporting	Monthly reports, results deck	Slides, dashboards	Built from the client’s campaign data
Time and billing	Timesheets, change orders, invoices	Time tracker, project board, QuickBooks	The agency’s own records, describing client work
Retrospective	What worked, updated templates and checklists	Wiki, project board	The agency’s own, once client details are taken out

The last two rows are the agency’s. The first six are where the client’s contract, NDA and platform terms decide.

Which buyers say they want agency data?

As of October 2026, five providers name marketing agencies or their records on their own pages.

Provider	Type	What its pages say
Avelence	Introducer; buys no data	“Software businesses, agencies, managed-service providers, consultancies, and support-heavy companies are useful starting points”; “Agency and media” on its profile form
FileYield	Listing marketplace	A Marketing & Advertising category (source); sellers warrant they “own the data or have all necessary rights, licenses, and consents to sell it”
Polyshares	Licenses directly	A case study of a “Global marketing agency” paid $110,000
Scale AI	Licenses directly	“Marketing Agencies” among its industries
Troveo	Licensing agent	“Marketing & Advertising” among its snapshot industries, including a “Marketing Agency” with 100 to 119 employees, 13 years and 11 systems; “marketing/media/creative agency” in its assessment

Two others name agency records or the wider trade without naming agencies: Mercor (its professional-services records, above) and Telegraph Lab (affiliated), which lists “Media & creative” among its target industries.

Who owns the campaign work: the agency or the client?

The client agreement decides it, so read its ownership clause first. The Copyright Office’s Circular 30 (revised August 2024) says work an employee creates as part of regular duties is a work made for hire, with the employer as author, so the agency starts out owning what its employees make. Work commissioned from an outside firm becomes the client’s work made for hire only in nine listed categories, and only “if the parties expressly agree in a written instrument signed by them that the work shall be considered a work made for hire”. Otherwise ownership passes by assignment, which is valid only if “in writing and signed by the owner of the rights conveyed” (17 U.S.C. §204(a)).

Standard contract forms draw the line in detail. AIGA’s Standard Form of Agreement for Design Services treats final deliverables separately from preliminary works and “Working Files”, defined as “all underlying work product and digital files utilized by Designer to create the Preliminary Works and Final Works”. Even where the designer keeps portfolio rights, the form limits them “to the extent they do not contain Client’s Confidential Information”. For media buying, the ANA’s contract template, as updated in July 2018, includes language so that “advertisers have access to transaction data” even where a vendor or media owner claims rights in it.

Read each master services agreement and statement of work for:

• Ownership. Whether deliverables, working files and rejected concepts pass to the client, and when.
• Data. Whether campaign, audience and performance data is defined as the client’s.
• Confidentiality. What counts as confidential, and how long the duty lasts after the account ends.
• Use limits. Whether client materials may be used only to perform the services.

Have your own lawyer read those clauses against any license before client material goes in.

Do client NDAs reach our Slack and email?

Often, because that is where client information travels. Frankfurt Kurnit’s October 1, 2026 commentary notes that “Email and chat contain other companies’ confidential information, often covered by NDAs and commercial agreements”, and that “De-identification also addresses whether a record can be linked to an individual, not whether its content is confidential”. An unreleased product, a client’s budget or its sales figures stay confidential after every name is replaced.

Prospects count too. A pitch the agency wrote is its own work, but the RFP and briefing a prospect shared may have come under an NDA signed before the pitch. Channels and folders named for clients can be left out as a block; leaving out records explains how.

What about ad-account data and clients’ customer lists?

Both come with terms the agency did not write.

• Data pulled through Meta’s APIs. If the agency’s reporting tool or connector reads data through Meta’s platform, Meta’s Platform Terms (updated February 3, 2026) list “Selling, licensing, or purchasing Platform Data” among prohibited practices. Platform Data includes “data anonymized, aggregated, or derived from such data”, and a tech provider may process it only “on behalf of and at the direction of your Client”. Other ad platforms have their own terms; read them for the tools you use.
• Customer lists for email and custom audiences. California’s privacy law defines a “service provider” as one working under a written contract that prohibits it from “Selling or sharing the personal information” (Civil Code §1798.140(ag)). If a client is covered by that law and treats the agency as its service provider, their contract has to say so.

What counts as the agency’s own records?

The records of running the agency rather than serving a client: new-business pitches, the creative and strategy process written down, templates and checklists, resourcing and timesheets, finance, hiring and training, and internal chat about the agency itself. Of everything an agency holds, this is closest to Scale AI’s stated interest in how marketers work.

Polyshares’ case study gives one example: a global marketing agency paid $110,000 for a “Perpetual exclusive license to the operational record. Anonymized before delivery.” The case does not say what the record contained, and Polyshares’ intake says “Data you hold for your own clients is out of scope”. A perpetual exclusive license also stops the agency licensing the same records elsewhere; selling versus licensing covers those terms.

What could an agency’s records be worth?

One published figure is agency-specific: Polyshares’ $110,000 case above, undated and self-reported. The other figures are providers’ general statements as of October 2026, not offers or averages: Polyshares “$100K to $2M, and above for large records”; Scale AI “$10K–$1M+ illustrative value per data partnership, scaling with cadence”; Troveo “AI labs and startups are actively paying six figures for company data exports”. Ask which records any figure is based on, since Polyshares, for one, puts data held for clients out of scope; how much company data is worth explains why ranges differ.

How does an agency owner start?

By sorting client folders, channels and boards from the agency’s own, and counting only the agency’s.

1. List the systems. Email, Slack or Teams, the shared drive, project boards, design tools, time tracking, CRM and accounting. Note the oldest year in each.
2. Mark client spaces. Client folders, client channels and client boards go on an exclusion list until the contracts are read.
3. Pull the contracts. MSAs, SOWs and NDAs for current and past clients, plus any data processing terms, go to a lawyer.
4. Count the agency’s side. The free sample tool reads a Slack export, mailboxes (.mbox or .eml) and CSV exports from project and CRM tools in your browser, and counts a documents folder by type and date without opening a file. Nothing is uploaded unless you choose to. Record the totals in an inventory.
5. Apply. The application needs no files; state that client work is out unless clients have agreed.