Selling customer support tickets and chats for AI training

Often yes: a company can license its customer support tickets and chats to AI buyers, after three checks. First, your customer contracts and the privacy notice in force when each ticket was written must allow it, because the FTC treats a broken privacy promise as a deceptive practice and buyers' counsel ask whether customer content is excluded or covered by the agreements. Second, the help desk the tickets live in decides how they may leave: Zendesk's exports are not enabled by default and not offered on Team plans, and its developer terms bar an app from using what it pulls for creating, training, or improving any general AI model, so the export has to be your own admin export, never a buyer's connector. Third, the names, emails, account numbers and agents in every thread must come out before any buyer reads one. What buyers pay for in a ticket is the resolution, the threading and the outcome: Prism asks for six months or more of multi-channel support logs with resolution tags, and Corpus Bridge counts outcomes attached and clean threading as adding value. Tickets usually sell as part of a company's records rather than alone: the median of the per-company sales on record is $28.3K per company, and one broker's illustration prices 96k Zendesk tickets at $27,500 inside a whole company's systems.

This guide is for a founder, COO or head of support whose help desk (Zendesk, Intercom, Freshdesk, Help Scout, Salesforce Service Cloud, HubSpot or ServiceNow) holds years of conversations with customers and how each one was resolved. It covers why AI buyers want those conversations, what they ask for in a ticket, how each help desk lets you export, the customer-data law and contracts that decide whether you may license them, de-identifying customers and agents, what tickets fetch as part of a company's records, preparing and delivering the export, the case of a closed company, and how a sale through AIData.Broker works. Every quote comes from the page it names, read on the day shown in its record.

This guide is research from public sources, not legal advice.

Contents

1. Why support tickets are valuable for AI agents
2. What buyers ask for in a ticket
3. Exporting tickets from each help desk
4. Whose words are in a ticket: customers, contracts and the law
5. De-identifying customers and agents
6. What tickets fetch as part of a company's records
7. Preparing and delivering a ticket export
8. A closed company's tickets
9. How a sale through AIData.Broker works
10. Questions people ask
11. What this means for you

1. Why support tickets are valuable for AI agents

Customer support is one of the first jobs AI companies are building agents to do, and one of the largest. The US Bureau of Labor Statistics counted 2,666,000 customer service representative jobs in 2025, and projects that number to fall over the next decade. An agent that answers customers has to learn what those representatives know: how a real customer describes a problem (badly, angrily, in fragments, with the wrong product name), which questions get to the cause, which answers resolve it, when to refund, when to escalate, and how to say no without losing the customer. That knowledge is written down in one place: the help desk. Every ticket is a worked example of a problem moving to a resolution, with the steps in between.

Public text on the web has very little of this. Companies publish their help-center articles, which say how things should work. They do not publish their tickets, which show how things actually went wrong and how a person fixed them. A help center tells a model that a password reset email arrives within minutes; years of tickets show every way that email fails to arrive, what the customer tried first, and what the agent did when the standard answer did not work. That gap is why buyers name tickets among the records they want. Turing, which buys companies' whole working records, says on its main page that communications, tickets, wikis and databases stand on their own and are often the most valuable part. Scale asks for records carried over with their complete history and metadata, and for conversations with threads, participants and reply structure preserved intact.

Agent builders also pay to simulate help desks, which shows how much they want the real thing. Prime Intellect, which funds training environments for agents, posted a bounty of $1,200 for an environment that simulates Intercom, so that an agent can practise working inside a support tool. A simulated help desk needs realistic tickets to fill it, and the most realistic tickets are the ones your customers wrote.

Four things in a help desk move its value most.

Resolutions. A ticket that ends with the customer's problem solved, and the agent's last reply saying how, is a complete example. A ticket closed for inactivity after one auto-reply is close to worthless. Buyers name this signal directly: Prism, Labelbox's data program, sets its requirement as support logs with resolution tags, and Corpus Bridge, a broker, describes the value as tickets marked resolved or not, with satisfaction scores and escalations.

Threading. A ticket is a conversation, and its value is in the order: the customer's first message, the agent's question, the customer's answer, the internal note to an engineer, the fix. Corpus Bridge counts clean threading, with each reply attached to the right question in the right order as adding value. An export that flattens a ticket into its first message and its status loses most of what a buyer pays for.

Channels and languages. Customers write by email, chat, social messages and phone. Prism's requirement is for multi-channel support logs, and Corpus Bridge counts languages other than English as adding value because they are underrepresented in public training text. A company that supports customers in Spanish, German or Japanese holds something rarer than one that supports them only in English.

Links to other systems. A ticket that links to a bug report in Jira, a pull request that fixed it, a refund in the billing system or a CRM record ties the conversation to what the company did about it. Sunset, a wind-down firm that buys closing companies' records, told Forbes that it prices on how traceable records are and how well they connect, and that a ticket tied to a specific code commit is worth more than a standalone document. The link only carries value if the other end comes too.

What lowers value is the opposite: tickets with no resolution, conversations handled entirely by a chatbot, a help desk used only for a year, or threads cut into pieces when the company changed tools. A plain count before you talk to anyone, split by channel, by year and by status, tells you most of what a buyer will ask first.

2. What buyers ask for in a ticket

The buyers that publish what they want from support data agree on more than they differ. Here is what their own pages say.

Read together, those rows describe one shape. A buyer wants each ticket whole: every public reply and internal note, in order, with who wrote it (as a role or a stand-in, not a name), when, through which channel, and the ticket's fields. It wants the outcome on the ticket: solved or not, the satisfaction rating if the customer gave one, whether it was escalated, and the tags or categories your team used. It wants enough of them, over enough time: Prism's floor is six months, and more years show more products, more failures and more ways of fixing them. And it wants the context around them: the macros (canned answers) your agents used, the help-center articles they linked, and where it exists, the link from a ticket to the bug or the refund it led to.

Three things in that list are worth preparing before any buyer asks.

Your status and tag scheme. Most help desks let teams invent tags freely, and after a few years a help desk has tags nobody uses, three spellings of the same category and statuses whose meaning changed when a manager changed. A short note that says what each status and the main tags meant, and when the scheme changed, turns a messy field into a labelled one. That is what "resolution tags" means in practice: a buyer can tell a ticket that was solved from one that was closed.

Satisfaction ratings. A rating attached to a ticket is the closest thing a buyer gets to a judgement of whether the agent did well. It is also a measure of a named employee's work, which is why section 5 treats agents as people to protect, not just customers.

Internal notes. Notes between agents, and from agents to engineers, often hold the actual diagnosis. They are also where staff write things they would never say to a customer. Decide whether they are in scope, and if they are, read them in the same review as the replies.

What buyers do not want is just as clear from their pages. None asks for a CSV of ticket subjects and statuses alone, and a report that loses the replies is not a ticket export. Buyers also want to know where a conversation came from: a thread handled by your own staff is different material from one handled by a chatbot, and Zendesk's own help says its export cannot include AI agent tickets, which is one sign of how vendors separate the two.

3. Exporting tickets from each help desk

The help desk controls the gate here, and each vendor's gate is different. Two questions decide every case: can your own administrator export the whole thread through the vendor's built-in tool, and what do the vendor's developer terms say about an app that pulls tickets through the API? The second question matters because buyers increasingly offer to connect to your systems for you, and several vendors now forbid exactly that for AI training. Buyers' counsel treat data pulled through a connector from a system like Zendesk as a red flag, and the safe answer is the one buyers' counsel describe as good: each system exported by its own admin with the vendor's native export tool, with the method recorded. Of the 18 business systems whose terms we have read, 13 let the customer export its own data without a limit that matters for a sale, while 6 bar using data pulled through their API to train AI.

Sources: each vendor's terms and help pages as held in its record on our rights pages, for example Zendesk and Salesforce.

Zendesk

Zendesk is the help desk with the clearest rules, and they are strict. Its help page on exporting ticket data says that "data exports are not enabled by default in Zendesk accounts": the person who holds the account's top role has to ask Zendesk Customer Support to switch them on. Once they are on, an administrator can export tickets, users and organizations as JSON, CSV or XML. JSON is the format to ask for: Polyshares' system guide for Zendesk names a full JSON export from Admin Center, and it is the format that keeps each ticket's comments together. The same record notes that the native export is available on Suite Growth, Professional, Enterprise and Enterprise Plus, and not on Team plans, whose only route out is the REST API.

That is where Zendesk's developer terms bite. Clause 3.6 of the Zendesk Developer Terms, effective 12 August 2025, says a developer must not use API data for anything other than its application's function, " including creating, training, or improving any general AI model, such as, but not limited to, a large language model", must not build applications that allow bulk export of API Data, and must not create persistent copies, archives, indexes, or long-term data stores from it. Our reading, set out in the record, is that a connector or script that pulls tickets through the API to build a training set, or to clean them for that purpose, appears to fall within that clause, and that this would include a Team-plan customer whose only route out is the API. The native export, switched on through Support and run by your own administrator, is the route least exposed to those terms. If you are on a Team plan and want to license your tickets, the cleaner path is to move to a plan that offers the native export before you export, and to record that you did.

Count before you export. Polyshares' guide points to a ticket count from Explore or the API; Explore, Zendesk's reporting tool, also gives counts by channel, status, year and satisfaction rating, which are the first figures a buyer asks for.

Intercom

Intercom's record is thinner, and it says so. Its customer terms were not reviewed, and the plan gating of its conversation exports was not verified. What the record does hold is Intercom's developer terms, which say an app "will not scrape, or otherwise create copies of any data obtained using the Intercom Service". There is no explicit AI clause, but a buyer's connector that copies your conversations to build a training set is copying data obtained using the Intercom Service. The same rule as Zendesk follows: export it yourself, through Intercom's own export tools, run by your own administrator.

One Intercom term is often misread. Intercom's additional product terms say it contractually restricts Third Party AI Providers from using Customer Data for training. That is a promise about Intercom's own AI suppliers (the models behind its AI features), not a limit on what you may do with your own conversations. It does tell you something useful, though: if your team used Intercom's AI features, some conversations were partly answered by a model, and a buyer will want to know which.

Salesforce Service Cloud

Service Cloud keeps tickets as cases inside the same Salesforce org as the rest of your customer records, so Salesforce's general terms and export tools apply. The Salesforce Main Services Agreement has no clause barring a customer from using its own Customer Data to train AI; it does require use in line with Salesforce's acceptable use policies and bars using the services to build a competitive product or service. The Data Export Service in Setup produces full exports weekly on Enterprise, Performance and Unlimited editions, and every 29 days on Professional and Developer, and the record notes that third-party guides report the download links expire after 48 hours. Case comments and email messages attached to cases are separate objects from the case itself, so make sure the export includes them, or the thread is lost.

The date that matters most is the end of the contract. After termination, Salesforce will make your data available for export only if you ask within 30 days after the effective date of termination; after that it has no obligation to keep it. A company that cancels Salesforce before exporting has a month.

HubSpot Service Hub

HubSpot's customer terms say "You own and retain all rights to the Customer Materials and Customer Data". Its developer terms, updated in May 2026, restrict using data accessed through the APIs to train, fine-tune, or improve AI or machine learning models, with an exception for single-customer use, which a multi-customer training set is not. So the same rule applies: export from HubSpot itself. The record notes that exports are made by users with export permission, and that Super Admins see an audit log of exports. HubSpot also says it may use Customer Data to train its own AI models unless you opt out in your settings; buyers sometimes ask whether a system's vendor has already trained on the data, so note whether you opted out and when.

ServiceNow, and internal tickets

ServiceNow mostly holds a different kind of ticket: IT requests and HR cases from your own staff, not conversations with customers. They can be valuable for agents that work inside companies, but they carry different people. The record advises excluding HR Service Delivery and security incident tables by default, because HR cases hold health, grievances and investigations. ServiceNow's agreement lets you retrieve your data, including a standard database export on written request within 45 days after termination, and the record found no clause restricting your own use of it. Note whether your instance shared data with ServiceNow's own AI features, which the record says it uses to improve its Now LLMs unless a customer opts out.

Freshdesk, Help Scout and the rest

We have not yet read Freshdesk's or Help Scout's terms into our records, so this guide states nothing about their plans or their developer terms. The same two questions apply to them and to any other help desk: does your administrator's own export carry every reply and note of each conversation, and do the vendor's developer terms say anything about an app that pulls conversations for AI training? Read the vendor's export help page and its developer or API terms, save a dated copy of each, and put both in your rights file.

4. Whose words are in a ticket: customers, contracts and the law

Ownership is rarely what stops a ticket sale. The conversations were collected by your company, in its own help desk, and the vendors say so too: HubSpot's terms, for one, say the customer owns and retains all rights to the Customer Data. What stops a sale is the people in the tickets, and the promises your company made to them and about them. A buyer's lawyers ask 27 rights questions, and 16 of them stop a sale until answered; for tickets, five matter most.

Your privacy notice, and the FTC

The first check is your own words. Section 5 of the FTC Act makes unfair or deceptive acts or practices in or affecting commerce unlawful, and the FTC has long treated a privacy policy as a binding promise. In February 2024 its staff applied that to AI training directly, warning that a company that reneges on its user privacy commitments risks running afoul of the law, including by changing its terms quietly and after the fact. The rule that follows is simple and tedious: each ticket is governed by the privacy notice in force when it was written. Buyers' counsel ask what the privacy policy in force at collection said about sharing or selling data, and the red flag they name is " 'We never sell your data' in an old privacy policy".

So collect every version of your privacy notice and terms of service since your first ticket, note the dates each was in force, and read each for three kinds of sentence: a statement that you do not sell or share personal information, a purpose limit ("we use your information to provide support"), and a clause allowing you to share de-identified or aggregated information. The last is the one you hope to find, because a de-identified licence then fits the promise. The FTC's remedies are why buyers care: in the Everalbum case the order required deletion of the models and algorithms it developed by using the photos and videos uploaded by its users. A buyer whose model trained on records collected under a forbidding promise has its model at risk, and it prices that risk into every ticket you offer.

State privacy law

California treats any transfer of personal information to a third party for monetary or other valuable consideration as a sale, with notice and opt-out duties attached, and most state privacy laws follow a similar pattern. Tickets are personal information almost line by line: names, email addresses, phone numbers, order numbers, addresses, and whatever customers volunteer about themselves. The way through is de-identification. Under California's law, personal information does not include consumer information that is deidentified, and the definition has three parts: reasonable technical measures, a public commitment not to re-identify, and a contract that binds every recipient ( 1798.140(m)). Buyers' counsel ask whether the de-identification meets the CCPA definition, including a public commitment and contract terms binding the buyer. The guide to selling customer data has a table of every state law in force.

Your customer contracts: whose tickets are they?

This is the check that most often decides a ticket sale, and it turns on what your company sells. If you sell to consumers, the tickets are your own relationship records with your customers: yours to license once the people are protected. If you sell to businesses, look more closely. The customer data guide sorts a company's records into three buckets, and tickets can fall into two of them.

Your own support tickets with your customers are your relationship records. A ticket in which a customer's administrator reports a bug in your product is yours, though it is full of personal information about your customers' staff.

Tickets your product holds for your customers are their content. If your company sells a help desk, a chat tool or an outsourced support service, the tickets in your systems are your customers' conversations with their own customers, held under their contracts. Under California's law a service provider's contract must prohibit it from selling or sharing the personal information, and the regulations allow a service provider to use it only for the listed purposes. In Europe, a processor that decides its own purposes becomes a controller in breach. Buyers' counsel list exactly this as a red flag: support tickets, customer success call recordings or customer-uploaded files in the dataset; contracts that restrict use to providing the service. De-identification does not cure a contract that limits use to providing the service.

Even your own support tickets can carry your customers' confidential information: a screenshot of their dashboard, their configuration, a pasted log with their customer names in it. A good rights file treats attachments and anything copied out of a customer's own systems as customer content by default, and the usual answer is to leave attachments out of the first licence altogether. Read your enterprise customers' contracts and data processing agreements as well: many large customers negotiate a clause on how their support interactions may be used, and a customer that said "only to provide support" is a customer whose tickets stay out.

Tickets can also hold other companies' material: a vendor's escalation, a reseller's account, a partner's integration details. Buyers' counsel ask whether that material is excluded or reviewed, and the largest company-records sale on record drew objections from five third parties over their own data inside it.

Sector law: money, health and recorded calls

Three kinds of support data carry their own law.

Financial services. A bank, lender, insurer or other financial institution may not hand customers' nonpublic personal information to an outside company unless it has given its privacy notice and each customer has had the chance to opt out first. Support tickets at a fintech are full of exactly that information, so for a financial institution, de-identification to a strict standard is the only realistic route.

Health. Patient support chats are named as a red flag in the health rights question: clinical notes or patient support chats with free text; data held as a business associate. Protected health information can be sold only after de-identification by HIPAA's safe harbor or expert determination, and a business associate agreement may forbid it entirely. A health company's tickets need an expert determination, not just a scrub.

Recorded calls. If your support data includes call recordings or their transcripts, the recording law comes with them. In California every party to a confidential call must consent to its recording ( Penal Code 632), and for any call with a cell or cordless phone on one end California requires the consent of all parties with no confidentiality condition. Buyers' counsel name as a red flag sales or support call libraries recorded under 'for quality and training purposes' notices only: a notice that a call may be recorded for training your own staff does not say a recording may be licensed to train someone else's model. Many holders leave call recordings out and license the text tickets alone.

Customers outside the United States

Customers in Canada, the EU and the UK bring stricter rules. Canada's federal privacy law has no legitimate interest ground, so a new use needs the person's consent unless the data is anonymised to the point of no serious possibility of re-identification; the red flag is names, phone numbers and emails stripped while the free text still points to people, which is what a lightly scrubbed ticket looks like. In the EU, data is anonymous only if no one can be singled out, linked or inferred, and the red flag buyers name is names swapped for tokens with no written test. If a buyer is a foreign company, or may pass the data to one, US rules on bulk sensitive personal data also apply, and buyers must be screened whatever the scrub.

The rights hub holds every one of these questions with its sources, and the employee messages guide covers the staff side in depth.

5. De-identifying customers and agents

Every buyer of company records expects personal data to be removed before or as part of delivery, and support tickets are among the most personal records a company holds. Customers write their names, addresses, order numbers, card fragments, health complaints and family circumstances into tickets because they want help, not because they expect to be read by anyone else. The question is who removes it, to what standard, and when.

Buyers differ. Scale says it anonymises everything inside your own systems before anything is extracted. Turing removes personal data and secrets itself, in isolated processing, and you approve the scope first. Prism says it removes all personal data as data is exported, before transfer into its infrastructure, and that it uses read-only connectors to your source systems to do its own ingestion and scrubbing. Polyshares says identifying fields, customer names, emails, and account numbers are stripped by a specialist PII partner before it or any lab sees the record. Corpus Bridge says the standard is negotiated with the buyer, and you approve it before anything is delivered. Each of those models means someone other than you reads your raw tickets first, and a connector model runs straight into the developer terms of section 3. Decide which you accept before you choose a buyer.

Tickets have particular shapes that a de-identifier must handle.

• Requesters and organizations. A help desk export keeps a users file and an organizations file mapping IDs to names, emails, phone numbers and companies. Replace them with stable stand-ins so each conversation still shows who replied to whom, which is what buyers value. Our method links every mention of a person to one stand-in, because a person counts as protected only when every mention of them is replaced, and draws each stand-in from a keyed hash under a key made on your computer, with a new key for each delivery so two buyers' copies cannot be joined.
• Identifiers in the text. Customers type order numbers, tracking numbers, account IDs, addresses, phone numbers and card fragments into the conversation itself. These need text-level detection checked against each identifier's own pattern, not just a swap of the users file.
• Signatures and quoted email. Email tickets carry signature blocks, disclaimers and the quoted history of earlier messages, each full of names and titles.
• Agents. Your support staff are people too, and tickets name them in every reply, in signatures, in internal notes and in satisfaction ratings that measure their work. California staff have had full privacy rights over their work data since 2023, and buyers' counsel ask how California employees' rights were handled; staff in the EU need a lawful basis for the reuse. Replace agents with role stand-ins (agent, senior agent, engineer), so the pattern of escalation survives while the person does not.
• Sensitive topics. Health, finances, children and family matters appear in support conversations more than anywhere else in a company. Our content screen flags records about health, children, beliefs, sex life, union membership, ethnicity, crime, biometrics or a person's finances so that you read them and leave them out, rather than masking them line by line.
• Attachments. Screenshots, invoices, photos of damaged products and log files cannot be scrubbed like text. Leave them out unless a buyer's use needs them and each one can be reviewed.

Removing names is not enough. A study in Nature Communications estimated that 99.98% of Americans would be correctly re-identified using fifteen demographic attributes, and a ticket's free text holds more than fifteen: the city, the product configuration, the unusual complaint, the date of the outage. That is why our method keeps a person's stand-in within each file by default, so the clues about one customer cannot be gathered across the whole help desk, and links them across files only when a buyer's stated use needs it.

Our de-identification standard sets out the method we advise: removal before anything leaves you, by you or an independent provider you choose, with a group-size floor of 11 for any delivery and 300 random records read by a person before it goes. Our free de-identification tool runs in your own browser, so your original records stay on your computer, and on company Slack, email and documents it measured an F1 of 86.8%, finding 95.8% of people's identifiers. No tool is perfect, which is why the standard pairs it with a person's review and a certificate from a provider you choose before each transfer. The model comes from the largest company-records sale on record, where payment was tied to delivery of records certified as de-identified to California's standard by an independent agent.

The licence completes it. California's standard needs a contract that binds the buyer, and our position is that every licence binds the buyer not to attempt re-identification, not to resell the raw data, and to bind any permitted recipient to the same terms, which is also what our standard asks of every buyer.

Exclusion and de-identification work together. Exclusion removes whole groups that cannot be licensed: tickets from customers whose contracts forbid it, a health product line, call recordings, attachments, tickets about legal disputes. De-identification makes what remains safe to deliver. Do exclusion first: it is cheaper, it is auditable from a list of filters, and it shrinks what the de-identifier has to read.

6. What tickets fetch as part of a company's records

No buyer we have read publishes a price for a help desk on its own. They price a company's whole working record, in which the help desk is one system, and they price it higher when more systems connect. Here is what the public record shows for records that include tickets.

Across every public offer for one company's records, the middle half sits at $174K to $693K per company. The middle of the buyers' offers is $316K per company, and the median of the sales that actually happened is $28.3K per company. Published tiers sit above what closes, partly because sales are disclosed more often for small closed companies, and the prices guide explains the gap.

Four patterns in those rows matter for a holder of tickets.

Per ticket, the numbers are small; per company, they are not. Corpus Bridge's illustration works out to about 29 cents a ticket, against about 2 cents an email in the same illustration. Tickets are priced above email because each one is a complete, labelled example. But a company does not sell its tickets one at a time: Turing's per-artifact rate applies only when bundled with a codebase, and the whole-record prices above are what a company actually receives.

Tickets raise the price of the whole package. Buyers treat a company's records as more valuable when its systems connect. Ooak's estimator values a 51 to 100 person operating company with four or more years at $85k to $121k with six tools, against $16k to $22k with only three, and Handshake says the highest payouts include data from 8 or more named applications. A help desk whose tickets link to the bug tracker, the code and the CRM is the kind of connection that moves a record from the lower row to the upper one.

Closed companies get less, and less every month. The market pays for a closed or closing company's records about 0.65 times what it pays for the same company operating. Turing's intake says records lose value fast once operations cease, naming 60 to 90 days for a CRM; a help desk ages like a CRM, because the products and policies it describes stop changing. Section 8 covers what to do about it.

The first offer is rarely the best. Business Insider reported in October 2026 that small businesses receive six-figure offers that expire in 48 hours for their internal records. Buyers quote very differently because they buy on different theses, which is why refusing a short deadline and collecting several offers is the single most useful thing a holder can do. Our valuation puts a range on your own company from these records before you talk to any buyer, with no upload.

7. Preparing and delivering a ticket export

The preparation guide sets out the full order for any company's records. For a help desk, it comes down to these steps, most of which need no buyer at all.

1. Confirm your plan, your roles and your dates. Note the help desk, the plan, when you started using it, any earlier tool you migrated from, and who holds the account's top role. On Zendesk, check that your plan offers the native export ( not Team). If a contract is ending, note the date: Salesforce gives 30 days after termination to ask for your data, ServiceNow 45 days.
2. Count before you export. Tickets by year, channel, status, language and satisfaction rating, and the share with an internal note or a link to another system. Polyshares and Turing both start from how much data sits in each system, for how long and with how many seats, and estimates are fine at that stage. Polyshares' intake asks for record counts, not files or exports.
3. Gather your promises. Every version of the privacy notice and terms of service since the first ticket, with dates, and the contracts of any customer that negotiated how its support data may be used. This is the rights file's first page.
4. Set scope with filters, not by hand. Write down which tickets are in and which are out, as rules: by date range, product line, brand, customer, channel and form. Exclude customers whose contracts forbid it, regulated product lines, call recordings and attachments for a first licence. A rule you can rerun is auditable; a hand selection is not.
5. Run the export yourself. Use the help desk's own admin export, run by your own administrator, on the full date range. Ask for JSON where it is offered, because it keeps comments with their ticket. Never install a buyer's app to pull the history: that is the route the export-route question treats as a red flag.
6. Keep each ticket whole. Do not edit the export by hand. Keep the tickets, their comments and notes in order, the users and organizations files, the fields and tags, and your macros and help-center articles if they are in scope, so threads and outcomes stay intact.
7. Write the scheme down. A short note on what each status, priority and main tag meant, when the scheme changed, which conversations were answered by a bot, and which tool the tickets came from before a migration. That note is what turns tags into the resolution tags buyers ask for.
8. Exclude, then de-identify. Remove out-of-scope tickets from the copy, then de-identify the rest to the standard you chose (section 5). Draw any sample for a buyer from the de-identified copy, never from the raw export: buyers' counsel ask whether any sample has already been shown to a buyer before the rights review, because a sample is itself a disclosure.
9. Write down how you did it. Who exported what, when, from which plan, through which tool, with which filters, and who de-identified. This becomes your rights file and your proof of title: buyers have been offered stolen copies of companies' data, and one buyer described a group that did not have the rights to sell this data.
10. Agree delivery and acceptance in the licence. The format, who de-identifies and to which standard, objective acceptance criteria and a review window.

Keep apart what you describe to buyers and what you deliver. Buyers price from the description first: counts, years, channels, languages, the share of tickets with a resolution and a rating, what links to what, and what stays out. Polyshares says it reads an anonymised sample before it prices. The full export moves only after a licence is signed, and only the scope the licence names.

8. A closed company's tickets

A company that is closing often holds a valuable help desk and has little time to use it. Three questions decide whether its tickets can still be licensed: does the history still exist, who can sign, and who can run the export.

Does the history still exist? A help desk keeps its tickets while the subscription is paid. After that, each vendor's own window applies: Salesforce makes data available only if you ask within 30 days after the effective date of termination and then has no obligation to keep it, and ServiceNow provides a database export on written request within 45 days after termination. For Zendesk, Intercom, Freshdesk and Help Scout we hold no record of what happens after cancellation, so treat cancellation as the end. The first act of any wind-down that may license its records is to keep the help desk paid until the export is done, and on Zendesk, to ask Support to switch exports on before anything else, because that request takes time.

Who can sign? The company licenses, not its founders personally, and the right person signs for it. Buyers' counsel ask for the legal entity that collected the data to be the licensor, with the board's approval, and for a wind-down, the board's resolution and any investor consents. In Delaware a dissolved corporation continues for three years to wind up its affairs, including disposing of its property, and must adopt a plan of distribution that provides for its claims. In a chapter 7 bankruptcy the trustee signs, because the trustee's duty is to collect and reduce to money the property of the estate; a sale of personal information then runs through section 363(b)(1), which holds the sale to the debtor's privacy policy unless an ombudsman is appointed and the court approves. Buyers ask that the right officeholder signs. Our estates page sets out the path for trustees, assignees and dissolved companies.

Who can run the export? The administrator who held the help desk's top role may have left. Transfer the role to an officer or the trustee before you need it, because the vendor's support team will deal only with someone who holds it, and Turing accepts an intake only from someone with admin authority.

A closed company's tickets carry the same people questions as an operating company's, with one difference: the privacy promises are now frozen. Whatever the privacy notice said on the last day is what governs, and in bankruptcy it governs the court's decision too. That pushes most closed companies toward thorough de-identification of their own support tickets, and away from anything their customers' contracts reserved.

9. How a sale through AIData.Broker works

AIData.Broker represents companies selling their data to AI buyers. We act only for the holder, as its agent in the sale, for holders based in the United States, the United Kingdom and Canada. For a holder of support tickets, a sale through us runs like this.

The valuation comes first, with no upload. The valuation asks about your systems (help desk, years, tickets, channels, languages, what else the company ran) and returns a range from the public records in section 6. If you add an export, it is measured on your own computer. The valuation binds no one.

You keep the data and deliver it yourself. Under the mandate, delivery under a licence is from you, or from a delivery or de-identification provider you choose, to the buyer directly. We do not send or show a buyer any part of your tickets, and the buyer pays you directly. That keeps your export on the route the help desks' terms allow: your own admin export, not a connector.

You decide who removes personal data. The mandate says we are not the de-identifier of record, including where you use our de-identification tool on your own computer, and that you decide who removes personal data and to which standard. We advise that it is removed before anything leaves you, by you or an independent provider you choose, as our de-identification standard describes.

Every buyer that fits sees it at once. Every buyer whose published requirements fit your records receives a description that does not identify you; each that gives its confidentiality undertaking receives the same datasheet, with one deadline for offers. Offers come back on the same terms, so you can compare price, de-identification model, payment timing and exclusivity side by side. Our free offer check reads any offer you already have against the published terms of the buyers we track.

The fee is due only out of money you receive. It is a success fee on what you are actually paid, band by band: 15% of the first $250,000, 10% from $250,000 to $1 million, and 6% above $1 million. On a $300,000 sale the fee is $42,500. Mandates are exclusive for 2 months, with a tail of 12 months for buyers we introduced, and we do not take mandates where the expected sale is below $26,000. Several buyers publish an introduction fee; in a sale we run we decline it and ask the buyer to add it to your price instead, so we earn the same whichever buyer you choose. How it works sets out every step and term.

Questions people ask

Can a company sell its customer support tickets and chats to train AI? Often yes, after three checks: that customer contracts and the privacy notice allow it, that the help desk the tickets came from does not forbid exporting them for AI, and that customers' and staff's personal data is removed before any buyer sees it. Tickets with their resolutions are among the records buyers ask for ( Prism, Corpus Bridge).

Does Zendesk allow exporting tickets for AI training? Zendesk's native export is off by default until Support switches it on, and not offered on Team plans. Its developer terms bar an app from using API data for creating, training, or improving any general AI model, so export through your own administrator, not a buyer's connector.

Can a buyer connect an app to our help desk to collect the tickets? Some buyers offer it ( Prism uses read-only connectors), but Zendesk's and HubSpot's developer terms restrict using API data for AI training, and Intercom's bar an app from creating copies of any data obtained using the Intercom Service. Buyers' counsel treat connector-pulled data as a red flag. Export it yourself.

What makes a support ticket valuable to an AI buyer? The resolution, the threading and the outcome: buyers ask for resolution tags, satisfaction scores, escalations and clean threading, several channels, languages other than English, and links to the bug or refund the ticket led to.

How much are support tickets worth? Buyers price a company's whole record rather than tickets alone. One broker's illustration prices 96k tickets at $27,500; public offers for one company's records have a middle half of $174K to $693K per company, and the median of disclosed sales is $28.3K per company. The valuation gives a range for yours.

We sell a support product to other businesses. Can we license our customers' tickets? Generally no, unless each customer agrees in writing. Those tickets are your customers' content, held under contracts that usually restrict use to providing the service, and California requires a service provider's contract to forbid selling or sharing the personal information. Your own support tickets with your own customers are different.

Do we need our customers' consent? In the US, usually not for de-identified tickets, if your privacy notice at the time did not promise otherwise; records that meet California's definition are not personal information. Canada, the EU, financial institutions, health data and recorded calls set a higher bar (section 4).

Do our support agents have to agree? Agents are employees whose names and performance are in every ticket. Replace them with role stand-ins, and handle California staff's rights and EU staff's lawful basis; the employee messages guide has the full answer.

Our company has closed. Can we still license its tickets? Often yes, if the help desk still holds them and the right person signs: in Delaware a dissolved corporation continues for three years to wind up, and in bankruptcy the trustee acts. Keep the subscription paid until the export is done; see the estates page.

What this means for you

Your help desk may be the most useful record your company holds for an AI buyer, because it is the only place where thousands of real problems and their real fixes were written down as they happened. Buyers name tickets, ask for their resolutions and pay more for records that include them. It is also one of the most constrained records you hold: the help desk decides how the tickets leave, your privacy notice and customer contracts decide whether they may be licensed at all, and every ticket is full of customers and staff who never expected to be read by anyone else.

So do four things now, before any buyer conversation. Check your help desk plan and its export route, and on Zendesk ask Support to switch exports on. Count your tickets by year, channel, language and outcome. Collect every version of your privacy notice and the contracts of customers who negotiated how their support data may be used, and write the filters that keep their tickets out. And get a range for the whole record from the valuation, so that the first offer you see is measured against the market rather than against nothing. Then export it yourself, de-identify it to a standard you chose, and put it before every buyer that fits at once. The rest of the What sells section shows the other records buyers pay for beside your tickets.