Most of the AI tools your finance team is being pitched right now have one thing in common: they only work with one ERP.
Some are built by ERP vendors themselves — extensions that only work inside their platform. Others are third-party tools whose entire architecture assumes a specific ERP's data model. Either way, the result is the same: you sign a contract for AI today, and three years from now when you want to migrate ERPs, you discover that your AI tooling has to be ripped out and replaced too.
This is the silent lock-in problem of single-stack AI. And it's getting worse, not better, as AI becomes core to finance operations.
The pattern that creates lock-in
Single-stack AI tools tend to share a few characteristics:
- The data model is hard-coded to one ERP's standard tables and fields. Switch ERPs, the AI breaks.
- The integration layer relies on one ERP's specific APIs or middleware. Switch ERPs, you start over.
- The vendor's go-to-market and roadmap are tied to one ERP's customer base. They have no incentive to support others.
You don't notice any of this when you're picking a tool. You notice it when you're trying to leave one — or when you grow into a multi-ERP environment via acquisition and the AI doesn't follow.
What ERP-agnostic actually means
True ERP-agnostic AI has three properties.
First, it abstracts the data layer. The AI's understanding of your invoice, your KPI, your user role doesn't depend on whether that data came from SAP, Oracle, or Microsoft Dynamics. It's been mapped to a normalised model that the AI reasons against.
Second, it has multiple integration adapters. Not just one mature one and a 'we'll get there' promise — actual production adapters for the ERPs that matter to mid-market finance teams.
Third, the vendor's commercial incentives are aligned with multi-ERP support. They sell to customers across ERP stacks. They have engineers paid to maintain each adapter. They're not just servicing the one ERP whose conference they're sponsoring this year.
What to ask vendors
When you evaluate AI tools for finance, ask:
- Which ERPs do you support in production today, and which do you have on your roadmap?
- If we migrate from ERP A to ERP B in two years, what happens to our configurations, training data, and historical context?
- How does your data model abstract from the underlying ERP — can I see the mapping?
- What percentage of your customer base is on each ERP?
A vendor with one production adapter and a roadmap of 'coming soon' for everything else is functionally single-stack. A vendor with adapters in multiple production environments and a clear architecture for adding more is genuinely ERP-agnostic.
The strategic case
Beyond the lock-in argument, ERP-agnostic AI has a strategic upside: it doesn't constrain your future ERP decisions.
Mid-market companies often inherit a multi-ERP environment via acquisition. If your AI works on the acquirer's ERP but not the target's, integration projects become AI-replacement projects too. Conversely, if your AI works across both, the acquired company can keep running on their ERP while still benefiting from your platform — and the eventual migration is decoupled from the AI question entirely.
For finance leaders thinking five years out, ERP-agnostic AI is just better optionality. You can switch ERPs, acquire companies on different ERPs, or run multiple ERPs in parallel — and the AI doesn't get in the way.
Bottom line
The AI you pick today will outlive your current ERP relationship. Choose accordingly.