AI-native ERP: what it is and what happens to your legacy history when you switch

What is an AI-native ERP?

An AI-native ERP is a general ledger and operations system designed so that automation does routine finance work inside the workflow: coding transactions, matching bank and card activity, flagging anomalies, preparing reconciliations and answering questions in plain language. The contrast is with established ERPs that add AI assistants to a data model built before these tools existed.

The category is young. For a CFO, the label matters less than two practical questions: will the automation shorten the close and slow finance headcount growth for this company, and what happens to the years of history the current system holds?

How AI-native, AI-added and legacy ERPs compare

What happens to legacy history when you switch?

Most ERP migrations, AI-native or not, load opening balances, open items and a limited window of detailed transactions, and leave everything older in the outgoing system. That choice is easy to leave until late in the project, when it gets made by whoever scopes the data load, and it decides whether a decade of transactions stays usable or becomes a backup nobody reads.

The guide on how many years of history to load into FP&A software covers the planning side. For the archive itself, there are five realistic homes:

The guide to legacy ERP data archiving compares these archive options in more depth.

The six-question history decision

Run it before anyone signs off the data migration scope.

1. What do we have? Years of history per module (general ledger, AR, AP, inventory, projects, payroll), entities, attachments, custom fields and the systems that feed the ERP.
2. What does the new system need? Opening balances, open items and enough detail for comparatives, budgets and the next audit.
3. Where does the rest live? Pick a home from the table above, with a named owner and a budget line.
4. What must we keep? Retention periods depend on the entity, the state and any open disputes or audits, so get written answers from the tax adviser and counsel before any record is purged.
5. What can the new vendor do with our data? Read the data-use section of the contract, including any right to use customer data to train or improve the vendor's own models. That matters if the company might later grant an exclusive AI-training license on its records.
6. Is the history worth more than its storage bill? Before the old system goes read-only, ask whether the history, joined with CRM, support and project records, could be licensed.

Why older ERP history can be worth more than it looks

AI systems are increasingly built to complete finance and operations tasks, not just answer questions about them. Teaching and testing an agent to do that work takes records of real multi-step processes with their outcomes: requisition, purchase order, goods receipt, vendor bill, approval and payment, with the exceptions, credits, disputes and close adjustments that happen along the way. That material lives inside companies, not on the public web.

Researchers at Epoch AI estimated in 2024 that, if current trends continue, language models could fully use the stock of public human-generated text sometime between 2026 and 2032. It is a forecast with wide uncertainty, but it helps explain why permissioned, non-public records attract interest.

ERP history is most useful when it joins to other systems. The joins behind a customer profitability analysis, linking orders, invoices, support effort and margin by customer, are the same ones that turn a ledger into a record of how the business actually worked. The irony is plain: the project that brings AI into the finance stack can be the one that leaves years of finance history behind.

What this means for a fractional CFO as a referral partner

You often scope the vendor selection and the migration yourself, so you see the history decision before anyone else. Raise data licensing at one of four moments: when the vendor shortlist is set, when the data migration is scoped, when the cutover plan is approved, or when the old system's decommission date is booked.

The introduction itself is short:

1. Check basic fit with the owner or CEO: the business is US-based, reached 50+ full-time employees at peak (contractors excluded), has kept documented records over several years of operation, created those records itself and has someone with authority to sign.
2. Use the company fit checker for a first, non-binding read.
3. Introduce the company through your referral link or the referral form. You pass on basic fit information only, never exports or descriptions of records.
4. SourceX qualifies the company, which then completes a data inventory, ideally before the legacy system goes read-only.
5. The company agrees price and terms before AI labs and data buyers review the opportunity; once a company is deal-ready, buyers typically respond within about two weeks.
6. After signing and delivery under the agreed redaction rules, the company is paid, and your reward follows only once SourceX has received its fee.

What to say at the steering meeting:

Partners earn 25% of the eligible platform fees SourceX actually collects from the referred company's licensing deals, capped at $100,000 per referred company. Rewards become payable only after the buyer pays and SourceX receives its fee; a lead, meeting or signed agreement alone does not trigger payment. Nothing is taken from the company's price to fund the reward. If you are a licensed CPA, check your state board's rules before accepting any referral payment. More on how introductions work for your role is on the page about referral opportunities for fractional CFOs.

Limits and open questions

• The category is young. Vendor longevity, auditor comfort with AI-prepared entries and module depth all vary; ask for references from companies of similar size and complexity.
• Not every company's history is licensable. Records held on behalf of clients, mainly consumer personal data, mainly protected health information, or data already licensed for AI training usually rule a company out.
• A license does not change retention duties. Keep what tax and legal obligations require either way.
• The Epoch projection is about public text in general; it says nothing about demand for any particular company's records.
• Vendor data-use terms can change at renewal, so recheck them each cycle.
• If the switch is part of a wider AI build, the guide to the R&D credit for AI development projects covers which project records support a claim.

Next step

Put the six-question history decision on the agenda for the next steering meeting. If the company fits and the owner is open to it, register as a partner and introduce the company before the old system goes read-only. The CEO can also apply directly at sourcex.si/apply through your referral link, which keeps your credit attached.