Selling accounting and finance records for AI training: QuickBooks, Xero and NetSuite

Yes, a company can license its accounting and finance history to an AI buyer, and buyers ask for it by name: Nyne lists "QuickBooks, Xero, NetSuite and similar systems: invoices, estimates, AP and AR, and the transaction history behind them". Finance records teach agents real office workflows (raising an invoice, matching a payment, reconciling a bank feed, closing the month), and labs pay people to build exactly that work into practice tasks: Terac pays experienced QuickBooks users $110 an hour to write environments for accounting and sales workflows. But no records a company keeps need more removing before they leave: every invoice names a customer and its price, every bill names a supplier and its terms, loan files carry a lender's covenants, and payroll is personal data about staff. On exporting, the records we hold say this: Intuit's terms say "Your Content remains yours" but bar training or developing artificial intelligence or machine learning models or related technology using information derived from the Platform without authorization; Xero's developer terms say "API Data cannot be used to train, fine tune, adapt, or enhance any AI Models", so the books have to leave Xero through its own exports; and for NetSuite we hold only Oracle's commitment to self-service tools and processes to retrieve exportable data in a structured, machine-readable format. Buyers price finance records as part of the company's whole working record: Nyne says an established US business's accounting and customer records commonly fetch $100K or more, and the median of the per-company sales on record is $28.3K per company. The steps are: export the books yourself while the subscription runs, take out what you owe to customers, suppliers, lenders and staff, de-identify and aggregate the rest, and put a description of it before every buyer that fits at once.

This guide is for a founder, chief financial officer, officer or trustee of a company, operating or closed, whose books live in QuickBooks Online, Xero or Oracle NetSuite. It covers what a finance system holds and what buyers value in it, how each system lets the company export, the confidentiality the company owes to customers, suppliers and lenders, the personal data in payroll and in records about individuals, de-identification and aggregation, what finance records fetch as part of a company's records, preparing and delivering, the case of a closed company, and how a sale through AIData.Broker works. Every quote comes from the page or filing it names, read on the day shown in its record. Where buyers or vendors publish nothing on a point, this guide says so rather than filling the gap.

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

Contents

1. What a finance system holds, and what buyers value in it
2. What comes out of the books before anything is licensed
3. Exporting your books from QuickBooks Online
4. Exporting your books from Xero
5. Exporting your records from NetSuite
6. Confidentiality owed to customers, suppliers and lenders
7. Personal data: payroll, staff and individual customers
8. De-identifying and aggregating finance records
9. What finance records fetch as part of a company's records
10. Preparing and delivering finance records
11. A closed company's books
12. How a sale through AIData.Broker works
13. Questions people ask
14. What this means for you

1. What a finance system holds, and what buyers value in it

Most people think of a company's books as a set of reports: a profit and loss statement, a balance sheet, a cash flow statement, a tax return. Those are the summaries. Underneath them, an accounting system holds the work that produced them, entry by entry. In QuickBooks Online or Xero that means every invoice raised and every estimate sent, every bill entered from a supplier, every payment received and matched, every bank transaction imported and categorised, every journal entry posted at month end, and the chart of accounts that gives each entry its place. In NetSuite, which runs as a full enterprise system rather than a bookkeeping tool, the same ledger sits inside a wider record: sales orders, purchase orders, item receipts, inventory movements, revenue schedules, approvals, subsidiaries and the saved searches the finance team built to run the business.

That record is a precise history of how a company handles money, step by step, with every step dated and every amount balanced. A sale becomes an invoice, the invoice becomes a receivable, the receivable is paid, the payment is matched, the bank line is reconciled, and the month is closed. A purchase goes the other way: a purchase order, a bill, an approval, a payment run, a reconciliation. When something goes wrong (a short payment, a duplicate bill, a credit note, a write-off, an accrual reversed the next month), the books show what the team did about it. Every one of those steps has a right answer that the next entry confirms.

That is why buyers ask for finance records. Nyne, which buys companies' historical business records, names "invoices, estimates, AP and AR, and the transaction history behind them" from QuickBooks, Xero and NetSuite. Handshake's data partnerships page lists among the sources it buys NetSuite, Paychex and QuickBooks, beside Gmail, Google Drive and Slack. Ooak Data includes financial records among the company records it takes, beside documents, project management data, CRM and code. Replay prices a company's operating data (code, docs, CRM, finance, support history), Red Deer Data bids on a company's business software records (communications, orders, finance, SOPs), and Grepped's range covers company records that include finance records. Labelbox's Prism names SAP among the systems whose operational data it licenses, beside Salesforce, Jira and ServiceNow.

Why buyers want it is plain from what AI labs are building. An analysis of the reinforcement learning environment market by Epoch AI quotes a builder saying that after maths and code, "we're now seeing significant growth in enterprise workflows: tasks like navigating Salesforce, filing reports, or manipulating spreadsheets". A model meant to work as a bookkeeper's or a controller's assistant has to learn what that work looks like in real books, with real mistakes and real corrections, and public text holds almost none of it, because companies do not publish their ledgers. The labs' own spending shows the demand: Terac pays experienced QuickBooks users $110 an hour, for at least two weeks of work, to build RL environments and write their tasks for accounting and sales workflows, and pays corporate finance professionals $1,000 per task to build environments from financial models and narrated walkthroughs. A company's real books are the material those environments imitate.

Four things in a finance system move its value most.

Workflows with their outcome. An invoice followed by its payment, its matching and its reconciliation is a complete example of accounts receivable. A bill followed by its approval, its payment run and the bank line that cleared it is a complete example of accounts payable. A month that closes with its accruals, its adjustments and its reconciliations is a complete example of a close. The value lies in the sequence and in the check at the end: the books either balance or they do not, which gives a buyer a way to score what a model does.

Exceptions and corrections. Clean, repetitive entries teach little. The entries that matter are the ones where a person had to think: a payment that matched two invoices, a supplier credit applied against the wrong bill and then moved, a bank feed line that needed a rule, a journal that reversed an error, a write-off with a memo explaining why. Memo fields and the notes attached to transactions are short, dense and written by people who do the work, which is the kind of material buyers pay more for: Mode.Inc's chief executive told reporters that value rises with the number of white-collar employees.

Years and volume. Finance records often reach back to the company's first sale, and estimators price years directly. Nyne's estimator for an operating company's accounting, customer and scheduling records returns a higher range for the same headcount with eight to fifteen years than for three to seven, and its ranges rise again with size, up to 250 or more employees with fifteen years or more.

Links to the rest of the company. A ledger rarely stands alone. Invoices tie to the quotes and contracts in a document store, to the opportunities in a CRM, to the emails in which a customer disputed a charge, to the tickets in which a billing error was fixed. A wind-down firm told Forbes that a record tied to the work it produced is worth more than a standalone document, and Ooak's estimator shows the effect in numbers: six tools including Gmail and Drive at $85k to $121k, against $16k to $22k for the same company with only Jira, GitHub and HubSpot. Books delivered with the systems around them carry more of the company's story than books alone.

What lowers value is the opposite: a ledger kept only for the year-end accountant, entries posted in one batch a quarter, a few months of history, or books migrated from one system to another with the old transaction detail dropped and only opening balances carried over. A plain count before you talk to anyone (transactions per year, invoices and bills per year, how far back the detail reaches, how many entries carry a memo, which other systems link to the ledger) tells you most of what a buyer will ask first.

2. What comes out of the books before anything is licensed

A finance export holds everything at once: the work, and a great deal of material the company may not pass on. Buyers and their counsel expect the second kind to be found and excluded before anything leaves, and one law firm advising holders on these sales tells them to exclude high-risk data entirely rather than relying on de-identification. Our register's own notes on each system say the same thing in different words. On QuickBooks: accounting data contains customers' and vendors' names, bank details and employee payroll (sensitive), so heavy de-identification is needed. On Xero: accounting data carries counterparties' names, bank and payroll data. On NetSuite: ERP data includes customers, vendors, pricing, payroll and employee records, plus counterparties' confidential terms.

The categories that come out of a set of books are much the same in every company.

Two of those rows are particular to finance records and catch holders out.

Lenders. A company that borrowed from a bank or took venture debt usually signed loan documents that watch what it does with its assets. The rights question buyers' counsel ask explains why: venture and bank loans commonly include a negative pledge on IP and an asset-sale covenant that lenders watch for disposals of material IP. The red flag is outstanding venture debt with a disposition covenant, or a lien on all assets including IP, and no consent. The good answer is that the loan documents were checked and the data licence is a permitted disposition or outbound licence, or the lender consented in writing. A lender's own figures inside the books (its interest rate, its covenant levels, its fees) are its confidential terms as well, which is a second reason to aggregate or remove the loan accounts.

Platform-derived fields. QuickBooks adds its own layer on top of what the company entered: suggested categories, bank rules it learned, benchmarks and reports it computes. Our QuickBooks record notes that the safer reading is that exported ledger data is the customer's Content, but derived reports, Intuit categorisation and benchmark fields may be Platform-derived. Keeping the entries the company made, and leaving out what the system generated, is the cautious line (section 3).

3. Exporting your books from QuickBooks Online

QuickBooks Online belongs to Intuit, and Intuit's terms answer the first question plainly. The Platform Terms say: "Your Content remains yours, which means that you retain any intellectual property rights that you have in your Content." The company's invoices, bills, entries and memos are its content.

The second question is harder. Among its prohibited uses, the same terms list scraping, accessing, or downloading content that doesn't belong to you, or training or developing artificial intelligence or machine learning models or related technology using information derived from the Platform without authorization. Our record calls this an ambiguous clause: it is unclear whether a customer's own exported books count as information derived from the Platform. The reading in our register is the cautious one set out in section 2: the ledger the company entered is its own content, and what Intuit's systems generated on top of it may not be. The record's advice is direct: have a lawyer read the clause before selling QuickBooks-derived data. The Intuit Developer Terms that govern the QuickBooks API could not be read when our record was made, so their AI clause is unverified, which is one more reason to take the books out through QuickBooks' own export rather than through a connected app.

How to export. Our record notes the route as Export to Excel (Settings, Tools, Export Data) during the subscription. An administrator can export lists (customers, suppliers, products, the chart of accounts) and reports, including the transaction-level reports that carry every entry with its date, account, amount and memo. For a buyer, the transaction detail is what matters, so a full export means the transaction reports for every year the books cover, plus the lists that give those transactions their meaning, rather than only the summary statements.

What happens after cancelling. Intuit's help says, in an excerpt our record holds, that "your data will be available in read-only access for 1 year from the date of cancellation" and that "as long as you're within the read-only access period of your cancelled account, you'll still be able to export your data." Our record notes that this applies for paid plans (90 days for trials). That is a longer window than most systems give, but it is a window, not a store, and a company closing down should export well before it ends.

The practical rule for QuickBooks is to export the company's own entries with its own administrator, keep a record of what was exported and when, leave out the suggested categories, benchmarks and computed fields, and have counsel read the AI clause before any licence is signed.

4. Exporting your books from Xero

Xero is the clearest case among the three, and the strictest. Our record of Xero's Developer Platform Terms (last updated 4 December 2025) holds the general rule, "You must not use API Data sourced from the Xero developer platform to train or fine tune any artificial intelligence models", and the clause that repeats it, clause 7(b)(i): "API Data cannot be used to train, fine tune, adapt, or enhance any AI Models." Clause 7(b)(ii) adds that "API Data shall not be passed to another third party without consent from the associated user." And clause 29 bars developers from: "Sell any user data (but you can charge users a fee for your app)."

Our record sums it up: of the accounting platforms we read, Xero's terms are the most restrictive, because data pulled through the Xero API may not be used to train or fine-tune any AI model at all, with no single-customer carve-out stated in the quoted text.

Those terms govern developers and the apps they build, not the company exporting its own books from inside Xero. Our record has not reviewed Xero's customer-side terms on export: it says plainly customer-side export terms not reviewed, and notes that Xero supports report and data exports in-product. So the advice follows: a holder should use native exports rather than an API connector. That rules out a buyer's connector, a data-room tool that pulls the books through the API, or any third-party app the company authorises to read its Xero file for the purpose of a sale. It also means a holder should still expect a buyer's counsel to ask how the data left Xero: keep a short note of who ran which export, when, and from which screen.

A full Xero export follows the same shape as QuickBooks: the transaction-level reports for every year, the contacts, the chart of accounts, the items, and the bank reconciliation history, run by an adviser or administrator on the company's own subscription. Xero's terms on what happens to a cancelled organisation's data are not in our record, so the safe course is to keep the subscription paid until the export is complete and checked.

5. Exporting your records from NetSuite

NetSuite is where our record is thinnest, and we say so. Its status is unclear and its own note says: treat as unverified. The general NetSuite terms of service could not be retrieved (403/404) when read, and on the API and AI side the record holds not found: the current terms of service and the SuiteTalk and REST API terms could not be retrieved, so whether NetSuite terms restrict AI use was not established.

What the record does hold is Oracle's EU Data Act Addendum for NetSuite Cloud Services, in which Oracle will allow You to retrieve or delete Customer Data from the eligible Cloud Service (i) at any time up to the Switch Date, and in which the available self-service tools and processes to retrieve exportable data in a structured, machine-readable format from an eligible Cloud Service are described in its applicable User Guides. Our record reads this as support for the view that the customer can get its own data out. It does not settle what the customer may then do with it.

How to export. The record notes saved searches and CSV export, SuiteTalk and REST APIs during the subscription. For a sale, saved searches run by the company's own administrator and exported to CSV are the route closest to the books: one search per record type (transactions with their lines, customers, vendors, items, employees for the purpose of excluding them), with internal IDs kept so that the lines can be joined back into the sequences buyers pay for.

After termination, the record notes that post-termination access reportedly ends unless a paid read-only arrangement is bought (consultancy blog, unverified). Treat that as a warning rather than a fact: for NetSuite, the export belongs before the renewal date, not after it.

Because the general terms could not be read, a NetSuite holder has one more step than the others: have counsel read the company's own signed NetSuite agreement (the order form and the terms it incorporates) for any clause on the use of exported data, before a licence is signed.

6. Confidentiality owed to customers, suppliers and lenders

This is the section that makes finance records harder to license than most company records. Every entry in a set of books is the company's side of a deal with someone else, and many of those deals came with a promise to keep their terms quiet.

Customers. An invoice shows what a named customer bought, at what price, with what discount, on what payment terms. For a business that sells to other businesses, those prices were often negotiated and the contract made them confidential. The rights question buyers' counsel ask is whether the data belong to or describe your customers and whether the customer contracts allow any use beyond serving them. The good answer is that customer content is excluded, or each customer contract expressly permits this use, or the data is anonymised to the standard in the anonymisation item and the contract does not forbid that. Removing a customer's name is the start, not the end: a customer's distinctive order pattern, its unusual product mix or its price on a rare item can still point to it.

Suppliers and partners. A bill shows what a named supplier charged and on what terms, and a purchase order may attach the supplier's quote. Supplier agreements and NDAs usually limit use of the other side's information to the relationship. Buyers' counsel ask whether the data contain other companies' confidential information, such as partner documents, pricing, roadmaps or anything shared under an NDA, because NDAs usually limit use to a stated purpose and survive termination. The good answer is that partner material is excluded or reviewed, and you can list which counterparties appear.

The largest records sale on record shows that counterparties do object. In the Spirit Airlines sale to Google, five third parties have objected to the Spirit records sale over their own data inside it. The lesson our register draws from it is to find and exclude vendors' confidential data before the sale, not after a counterparty finds it.

Lenders. Section 2 covered the covenant that may require a lender's consent before the company licenses an asset at all. Inside the books, the lender's terms (rate, fees, covenant levels, security) are confidential information too. The cautious structure replaces loan accounts with a generic label and keeps only the pattern of entries (interest accrued monthly, a repayment each quarter) that a model can learn from.

Licensed data. Some books carry data the company bought: a supplier price list loaded into items, credit reports attached to customers, industry benchmarks imported into reports. Licences for bought data usually forbid resale or sublicensing, so the good answer is that bought and licensed material is identified and removed before packaging, and the provenance file shows what remains is the company's own.

Controlled information. A company that sells to defence or aerospace customers may have item descriptions or project names in its books that point to controlled technical work. Buyers' counsel ask whether the data include controlled technical information, such as semiconductor designs, aerospace or defence engineering, or advanced encryption source code, because releasing it to a foreign national can be a deemed export under the EAR. In finance records the risk usually sits in item names, project codes and attachments, which is one more reason to replace item descriptions with categories.

The general point, which the customer data guide sets out in full, is that de-identification and these promises are separate questions. Removing a counterparty's name does not cure confidentiality owed to it if the price or the term itself is what the contract protects. The safest structure for most companies is to license the shape of the work (the sequence of documents and entries, the timing, the corrections) with amounts transformed or banded and every counterparty replaced by a stable pseudonym, and to exclude outright any customer or supplier whose contract forbids even that.

7. Personal data: payroll, staff and individual customers

Finance records hold personal data in three places, and each needs its own answer.

Payroll. Payroll journals, employee records, salary changes, bonuses, reimbursements and expense claims are personal data about staff, and among the most sensitive a company holds. Our QuickBooks record lists employee payroll (sensitive) among the reasons accounting data needs heavy de-identification. Payroll rarely adds anything a buyer needs that cannot be kept in aggregate: a model can learn how a payroll journal is posted from a total per department per month, without knowing what any person earned. The usual answer is to exclude employee-level payroll altogether and keep only aggregated journals.

Individual customers. A business that sells to consumers (a clinic, a shop, a trade business, a subscription service) has invoices addressed to people, with their names, addresses and what they bought. Where the company told those people in its privacy notice what it would and would not do with their information, that promise governs. Buyers' counsel ask about earlier privacy promises, and the customer data guide explains how notices, state laws and the data broker laws apply. For licensing to an AI buyer, the usual answer is the same as for staff: replace each person with a pseudonym, remove addresses and contact details, and keep the transaction pattern.

Business contacts. A business-to-business company's invoices are addressed to companies, but its records still name people: the accounts payable contact, the person who approved a purchase order, the email that received an invoice. In California, the state's privacy agency counts contacts at business customers and vendors as consumers, while most other state laws exclude people acting in a commercial or employment context. A holder with California contacts should treat those names as personal data and remove them.

The de-identification standard. For US consumer or staff data, the question buyers' counsel ask is whether the de-identification meets the California definition, which needs three things from the business holding it: reasonable technical measures, a public no-re-identification commitment, and contracts making recipients comply. The red flag is a 'PII removed' claim with no public commitment and no contractual flow-down.

Bulk financial data and foreign buyers. Finance records are one of the categories the US Department of Justice's Data Security Program counts. Buyers' counsel ask whether the buyer, or anyone it may pass the data to, is a foreign person, and whether the dataset contains US sensitive personal data above the bulk thresholds, which include health or financial data over 10,000 US persons. The good answer is that the buyer was screened for ownership by or location in China, Russia, Iran, North Korea, Cuba or Venezuela, the data were checked against the thresholds, and the licence carries an onward-transfer clause in the licence for foreign buyers. A consumer business with tens of thousands of customers' payment histories is squarely inside this question; well de-identified, business-to-business books usually are not.

One category this guide does not cover: tax returns and tax workpapers prepared for other people, which carry rules of their own that our records do not hold. An accounting firm holding its customers' tax files should take advice on those before anything else.

8. De-identifying and aggregating finance records

Finance records are structured, which makes them easier to de-identify than free text in one way and harder in another. Easier, because names, addresses, bank details and tax identifiers sit in known fields. Harder, because amounts, dates and patterns identify too: a single customer that pays a distinctive sum on the same day every month is recognisable to anyone who knows the business, with or without its name.

A sound treatment for books works in layers.

Remove what nobody needs. Bank account numbers, tax identifiers, addresses, phone numbers, emails, employee identifiers, and all attachments (scanned receipts, signed contracts, supplier PDFs). These add risk and nothing a model needs to learn the workflow.

Replace parties with stable pseudonyms. Every customer, supplier, employee and lender becomes a consistent token (one customer becomes "Customer A" in every entry, one supplier "Supplier B"), so that the sequence survives: the same customer's invoices, payments and credit notes still link to each other, which is the pattern buyers pay for, while the name is gone.

Transform amounts. Scale every amount in the books by one factor that is not disclosed, or band amounts into ranges, so that the arithmetic still balances (invoices still sum to receivables, payments still clear them) but no real price or salary survives. For a buyer training workflows, balanced books matter more than real figures.

Coarsen dates where needed. Keep the order of events and the gaps between them, but shift the whole ledger by a fixed offset or round to the week, so that a transaction cannot be matched to a public event.

Clean free text. Memo fields, invoice line descriptions and notes carry names, project codes and the occasional complaint about a customer. They need the same treatment as any free text: detection of names and identifiers, replacement with the same pseudonyms used in the structured fields, and a person's review of a random sample.

Aggregate what does not need to be individual. Payroll by department and month, small customers below a threshold rolled into one, rare items into categories. Aggregation is often the cleanest answer for the most sensitive parts of the books.

Who does this work varies by buyer. Some buyers do it themselves: Nyne's terms say Nyne anonymises the data you provide itself. That means the raw books leave the company first, which puts every customer's price and every employee's salary in a buyer's hands before de-identification. Many holders prefer to de-identify before anything leaves, either in-house or through a provider they choose. Our de-identification standard sets out how we measure that work and runs the detection engine in the holder's own browser, so the books do not have to be uploaded to be checked.

De-identification has limits. It does not cure confidentiality owed to a customer whose contract protects the price itself, it does not make a loan agreement permit a disposal it forbids, and it does not cure a licence on bought data. So the order of work is exclusion first, then de-identification and aggregation, then a person's review of a random sample.

9. What finance records fetch as part of a company's records

No buyer we have read publishes a price per invoice, per transaction or per year of books. They price the company's whole working record, in which the finance system is one of the systems, and the public record holds a few prices that name accounting or finance explicitly.

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 ranges sit above what closes, partly because sales are disclosed more often for small closed companies; the prices guide explains the gap.

Four things about those rows matter for a holder of finance records.

Read each figure for what it is. Nyne's figures are estimator outputs for stated answers, and its current company programme is for companies with at least 20 full-time US W-2 employees. Grepped's own site configuration calls its referrer calculator a deliberately generous range. Red Deer Data's accounting deal is self-reported by a firm whose site went live about September 2026, with no counterparty, filing or press confirming it. None of these is a realised sale price, which is why the medians above, computed from many records, are the better anchor.

Books alone are priced lower than books with the systems around them. The estimators that name accounting price it together with customer and scheduling records, and the lever our register draws from buyers' own statements is to bring every connected system, especially documents and email. micro1's revenue chief put it from the other side: without Slack, CRM or email records the rest of a corpus is worth less. Finance records are often the spine that dates and values everything else: they tell a buyer which emails, which tickets and which documents belonged to which paid piece of work.

Size and years set the band. Nyne's estimator rises from $170,000 to $299,000 for the smallest band it publishes to $777,500 to $2,217,000 for the largest, and it prices the same company lower outside the US: $64,000 to $138,000 for 50 to 100 employees with eight to fifteen years, outside the US, Canada and the EU.

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 most useful single step a holder can take. Our valuation puts a range on your own company from these records before you talk to any buyer, with no upload.

10. Preparing and delivering finance records

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

1. Secure the administrator and the subscription. Confirm that a current officer holds the primary administrator role in QuickBooks, an adviser or administrator role in Xero, or an administrator role in NetSuite, and that the subscription stays paid until the export is complete and checked. Turing's intake accepts records only from someone with admin authority, so the person who can run the export should also be the one who starts any buyer conversation.
2. Count before you export. Transactions per year, invoices and bills per year, how far back the transaction detail reaches, how many entries carry a memo, how many customers and suppliers appear, which other systems link to the ledger. Polyshares starts from counts per system, read from the admin console, and Turing says estimates are fine at that stage.
3. Read the agreements. Collect the loan documents and check for a negative pledge or a disposal covenant ( lender consent); collect customer contract templates and the largest customers' signed contracts for confidentiality clauses ( customer contracts); list the suppliers and partners whose terms were shared under an NDA ( counterparty confidentiality); collect every privacy notice in force over the life of the books ( earlier privacy promises).
4. Have counsel read the system's terms. For QuickBooks, the AI clause in Intuit's terms; for NetSuite, the company's own signed agreement; for Xero, confirm that the route is a native export and not an app.
5. Run the export yourself. Transaction-level reports for every year, lists, chart of accounts and reconciliation history, exported by the company's own administrator through the system's own tools. No connector or third-party app authorised to pull the books for a buyer. Record the date, the tool, the person and what was exported.
6. Check the export is whole. Compare totals in the export with the trial balance for each year: the books should balance in the export as they do in the system. Keep internal IDs so that lines join back to their documents.
7. Exclude, then de-identify and aggregate. Remove payroll detail, attachments, bank details, loan accounts, bought data and any customer or supplier whose contract forbids the use, in a working copy; then pseudonymise parties, transform amounts, coarsen dates and clean memo fields (section 8). Draw any sample from the de-identified copy, not from the raw export, and do not show a sample before the rights review, since a sample is itself a disclosure.
8. Write down how you did it. Who exported what, when, what was excluded and why, the agreements read, who de-identified and how it was checked. This is your rights file and your proof of title: buyers have been offered records by people who did not have the rights to sell this data.
9. Agree delivery and acceptance in the licence. The format, who de-identifies and to which standard, objective acceptance criteria and a review window, the no-re-identification clause and its flow-down, and the onward-transfer clause where the buyer is foreign. Keep finance records and samples out of email and chat: DataVendor tells holders plainly not to send data or code by chat or email.

Keep two things apart: what you describe to buyers, and what you deliver. Buyers price from the description first: years, transaction counts, systems, the share of entries with a memo, 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. Watch the payment terms as closely as the price: micro1's licence lets the holder invoice only after micro1 has accepted the data in writing, which makes the acceptance criteria the real payment trigger.

11. A closed company's books

A closed company is a common holder of a full set of books, and the one with the least time. Three questions decide whether its finance records can still be licensed: do the records still exist, who can sign, and who can run the export.

Do the records still exist? For QuickBooks, Intuit's help says data stays available in read-only access for 1 year from the date of cancellation for paid plans, and that export remains possible within that period. For NetSuite, access after termination reportedly ends unless a paid read-only arrangement is bought, an unverified report that is reason enough to export before the renewal date. For Xero, our record does not hold what happens to a cancelled organisation. In every case the safe course is the same: keep the subscription paid, or the read-only window open, until the export is done and checked. Records also lose value with time in a softer sense: the people who could explain the chart of accounts and the odd entries leave, and the systems that linked to the books shut down.

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 written approval, and, for a wind-down, the board's approval and any investor consents. In Delaware, where many US startups are incorporated, a dissolved corporation continues for three years to wind up its affairs, including disposing of its property, and it adopts 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. Buyers ask that the right officeholder signs.

Lenders in a wind-down. A closing company often still owes a lender, and the lender's security may reach the company's assets, its data included. The lender-consent question applies with more force here: if a lien covers all assets including IP, the lender's written consent, or a payoff, comes before any licence.

The bankruptcy rule on personal information. In bankruptcy, a sale of personally identifiable information is restricted where the debtor had told individuals of a policy prohibiting the transfer of personally identifiable information about individuals to persons that are not affiliated with the debtor, and the court may appoint a consumer privacy ombudsman under section 332. The definition is narrow: under section 101(41A), it covers information an individual gave in connection with obtaining a product or service primarily for personal, family, or household purposes. A consumer business's invoices to individuals fall inside it; a business-to-business ledger mostly does not. The bankruptcy guide walks through the rule and the precedents. For books the practical answer is the same as outside bankruptcy: pseudonymise every party, aggregate payroll, and license the shape of the work.

Who can run the export? The QuickBooks primary administrator, the Xero adviser or the NetSuite administrator may have left. Transfer the role to an officer, the trustee or the company's accountant while the subscription is live. If nobody with access remains, the vendor's support is the route, and the company will need to show it still exists and who acts for it. Our estates page sets out the path for trustees, assignees and dissolved companies.

The value is still real: the market pays for a closed or closing company's records about 0.65 times what it pays for the same company operating, so a closed company's books, exported in time and cleaned well, still fetch a real price.

12. How a sale through AIData.Broker works

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

The valuation comes first, with no upload. The valuation asks about your systems (which accounting system, how many years, how many people, what else the company ran) and returns a range from the public records in section 9. If you add files, they are measured on your own computer. The valuation binds no one.

You keep the records and deliver them yourself. We do not hold your export, show any part of it to a buyer, or license it in our name. You, or a de-identification provider you choose, deliver directly to the buyer, and the buyer pays you directly. That keeps the export on the route the systems' terms point to: your own administrator's export, not a connector.

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 accounting data for AI training? Yes, if the company exports its own books through its system's own tools, takes out what it owes to customers, suppliers, lenders and staff, de-identifies and aggregates the rest, and the right person signs for the company. Buyers name finance systems among the sources they want ( Nyne, Handshake, Ooak Data), and labs pay specialists to build accounting workflows into practice tasks ( Terac).

Do QuickBooks' terms allow it? Intuit's terms say "Your Content remains yours", but bar training AI using information derived from the Platform without authorization. Whether your own exported books count as derived from the Platform is unclear; the cautious reading keeps the entries you made and leaves out Intuit's categorisation, benchmarks and computed reports. Have counsel read the clause.

Do Xero's terms allow it? Xero's developer terms say "API Data cannot be used to train, fine tune, adapt, or enhance any AI Models", so no app or connector may pull your books for training. Our record has not reviewed Xero's customer-side export terms; the route is Xero's own in-product exports, run by your own adviser or administrator.

Do NetSuite's terms allow it? We could not read the general NetSuite terms, so we do not know. Oracle documents self-service tools and processes to retrieve exportable data in a structured, machine-readable format, which supports getting your own data out. Have counsel read your signed agreement for any clause on use.

Do we need our customers' or suppliers' consent? Not where their names are removed, their prices transformed, and their contracts do not forbid the use. Where a contract makes the terms themselves confidential, exclude that customer or supplier, or ask. Buyers' counsel ask about customer contracts and counterparty confidentiality.

Does our lender have a say? Possibly. Venture and bank loans commonly include a negative pledge on IP and an asset-sale covenant. Check the loan documents; the good answer is that the licence is a permitted disposition or the lender consented in writing.

Can we include payroll? Employee-level payroll is sensitive personal data and adds little a buyer needs. The usual answer is to exclude it and keep aggregated journals by department and month, de-identified to the California standard ( reasonable measures, a public commitment, binding contracts).

Does it matter if the buyer is foreign? Yes, for books with personal data. The DOJ rules count health or financial data over 10,000 US persons as bulk data, so the buyer has to be screened and the licence needs an onward-transfer clause.

How much are accounting records worth? Buyers price the whole working record rather than the books alone. Nyne says an established US business's accounting and customer records commonly fetch $100K or more; 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.

Who removes the personal data? It depends on the buyer. Some do it themselves ( Nyne anonymises the data you provide itself), which means the raw books leave first. Many holders prefer to de-identify before anything leaves; our de-identification standard explains how.

Our company has closed. Can we still sell its books? Often yes, if the records still exist and the right person signs. QuickBooks keeps a cancelled paid account in read-only access for 1 year; in Delaware a dissolved corporation continues for three years to wind up, and in bankruptcy the trustee acts. See the estates page.

What this means for you

Your books are the most exact record your company keeps of how it does its work: every sale, purchase, payment and correction, dated, valued and checked against the rest. That is what buyers want from them, and labs are paying specialists to recreate exactly that work in practice tasks. They are also the records most full of other people's confidential terms and personal data: every customer's price, every supplier's terms, every lender's covenants and every employee's pay.

So do four things now, before any buyer conversation. Make sure a current officer holds the administrator role and that the subscription stays paid. Count your transactions by year and see how far back the detail reaches. Pull out the loan documents, the largest customer and supplier contracts and the privacy notices in force over the life of the books, and read what each says about confidentiality and disposals. 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 the books yourself, leave payroll detail and bank details behind, pseudonymise every party and transform the amounts, and put a description before every buyer that fits at once. The CRM guide covers the sales history that usually sits beside your ledger, the email archives guide covers the mail around both, and the rest of the What sells section shows the other records buyers pay for.