Add AI to Your ERP or Build a Native AI ERP

Every operator who has lived through an ERP cutover has an opinion about the next one. The last decade's answer was to move off on-premises systems and onto cloud suites: NetSuite, SAP S/4HANA, Dynamics 365, or Sage Intacct. The 2026 question is different. The incumbents are shipping agentic features at a pace most portfolios cannot absorb, and a well-funded group of challengers is pitching the opposite path: throw out the general ledger and start again on a system built around large language models from the first line of code.

For a portfolio CFO sitting on three to ten operating companies, this is not an academic debate. It is a budget line, a data-model decision, and a people decision. The wrong call is expensive in two directions. A premature replacement risks a failed migration. A delayed one risks running operations on a system that is, functionally, the slowest tool in the shop.

The honest framing is that neither path is obviously right. It depends on the asset, the data, and the holding period.

The Incumbent AI Rollout Is Already Underway

Before considering a rip-and-replace, operators should take stock of what the current ERP vendor is already shipping. The pace has accelerated sharply. Oracle has released over 600 AI agents across its Fusion Cloud suite, embedded into ERP, EPM, HCM, SCM, and CX at no additional licensing cost, with the vendor claiming up to 96% of transactions can be automated inside Oracle Cloud ERP. Microsoft meters its Copilot agents through Copilot Credits, with pay-as-you-go pricing at $0.01 per credit plus a $30-per-user Copilot add-on on top of the base Dynamics or M365 seat. NetSuite announced that beginning July 6, 2026, NetSuite AI Units will be included with eligible user licenses at no additional cost.

Gartner research cited across vendor comparisons projects that by 2026 over 80% of ERP vendors will embed generative AI capabilities into their suites. This matters for a practical reason: in many portfolios, agentic features are quietly turning on inside systems that were procured years before anyone used the word "agent." Security and finance teams are discovering, after the fact, that AI features embedded in existing ERP, CRM, and developer platforms may already exceed what has been inventoried.

The implication is not that incumbents have won. It is that the first question an operator should ask is which agents are already paid for, which are consumption-metered, and which are not yet switched on. That audit often removes 40% of the perceived "AI gap" before any migration conversation starts.

What AI-Native ERPs Actually Change

The challengers (Rillet, DualEntry, Campfire, Light, Doss, Everest, and a handful of others) are not simply NetSuite with a chatbot. The architectural claim is that legacy mid-market ERPs were architected as systems of record, not systems of action, and that bolting AI onto a relational schema designed in the 1990s puts a ceiling on what agents can do. The data-model difference is the point. Native systems treat the ledger as a stream of events the model can reason over, rather than a destination for human-entered journal entries.

Venture capital has committed, in the aggregate, approaching half a billion dollars to the category in roughly eighteen months. The sales pitch is explicit: "switch off NetSuite." Early traction skews toward services businesses and tech-native companies rather than heavy industrial operators. One useful observation from a 2026 A/B test on European mid-market CFOs: the "keep your GL, add the layer" framing drew roughly twice the reply rate of "replace your ERP". That is not an argument against native systems. It is an argument that the installed base is skeptical, and skepticism is a cost the challenger has to overcome deal by deal.

For a holding company running an industrial services roll-up on NetSuite OneWorld across six subsidiaries, the data gravity alone is likely decisive. For a newly acquired professional services firm still closing books in QuickBooks and spreadsheets, a native platform is now a credible first ERP rather than a staging post to a larger suite.

The Cost Picture Is Not What the Vendors Say

Budgeting conversations tend to anchor on sticker prices and skip the operational drag. Three numbers are worth internalizing before any procurement memo gets drafted.

First, the baseline cost of an AI augmentation project on an existing ERP. Integrating basic AI add-ons into an existing ERP typically costs $25,000 to $100,000, with full enterprise AI ERP deployments reaching $350,000 to over $1 million across multiple modules. These are implementation and integration fees, not the recurring consumption charges on top.

Second, the baseline cost of a full ERP replacement. Panorama Consulting's 2026 median ERP project lands at roughly $450,000 over 15.5 months, though their sample skews toward companies above $200M in revenue. Lower-middle-market deployments can land below that, but the time dimension rarely does. Fifteen months of distracted finance leadership is itself a cost.

Third, the failure rate. Gartner has estimated that ERP implementation failure rates can exceed 75% when the definition includes projects that were abandoned, delivered over budget, finished late, or failed to deliver expected business benefits. That is the number that should sit at the top of every steering committee deck. The upside of a successful rollout is real: AI-driven automation in ERP has been credited with 20–40% operational cost reductions, 70% invoice processing labor reduction, and ROI of 330–400% within 24 months. The downside is a two-year distraction with a 3-in-4 chance of disappointing the board.

Decision Criteria for a Portfolio Operator

The question is rarely "which ERP is best." It is "which path fits this specific company, now." A few criteria do most of the work.

• Transaction complexity. Multi-entity, multi-currency, inter-company eliminations, complex revenue recognition, and heavy inventory consumption favor an incumbent that has decades of edge cases coded in. A single-entity services business with clean billing favors a native platform.
• Data cleanliness. AI agents amplify whatever is in the ledger. Companies with heavy historical tech debt get less from a copilot layered on top. A native migration forces a cleanup, which is often the real value.
• Holding period. A business the holding company intends to own for twenty years can absorb a longer, more disruptive replacement. A platform acquisition heading toward an add-on sprint should not be in the middle of an ERP cutover during integration.
• Operator bench. Native ERPs push more of the configuration work onto the finance team itself. An operator without a controller who can own the system is a bad fit for either path, but a worse fit for a replacement.
• Integration surface. Count the systems that currently write to and read from the ERP. Every one of them is a migration task.

Most portfolio companies in the $2M–$30M EBITDA band that already run NetSuite, Dynamics, or Intacct will be better served by sequencing the incumbent's agents in, cleaning data in parallel, and revisiting the replacement question in 24 months. Companies on QuickBooks or spreadsheets have the easier call: a native platform is now a reasonable first ERP, with the fallback that ERP.io and similar advisors can run the implementation without the overhead of a traditional systems integrator.

Sequencing the Work

Whatever path a portfolio chooses, the sequencing matters more than the vendor. A workable order for most lower-middle-market operators:

1. Inventory what is already licensed. Turn on the agents that come with the current contract before paying for new ones. This is almost always the highest-return week of work in the project.
2. Fix the data model. Chart of accounts, item master, customer master, vendor master. Agents trained on bad data produce bad outputs faster.
3. Pick one process to automate end-to-end. Invoice processing and AP are the most forgiving starting points because the economics are measurable and the failure modes are visible.
4. Decide the replacement question with evidence. After six months of running agents on the current ERP, the gap between what the incumbent can do and what the business needs becomes concrete rather than hypothetical.
5. If replacing, plan the parallel run. No native platform should carry month-end close alone until it has run alongside the incumbent for at least two full cycles.

This sequencing applies equally to agentic work outside the ERP. Portfolio-level automation (agentic AI consulting, custom LLM deployment, replacing outsourced functions with agents) rests on the same data foundation. Related reading on how this plays out operationally: replacing outsourced VAs with AI phone systems and building institutional memory across a portfolio.

The Honest Answer

The answer to the title question is almost always "add first, then decide." The incumbents are shipping fast enough that the differential closes every quarter. McKinsey found that 88% of surveyed organizations regularly used AI in at least one business function in 2025, up from 78% a year earlier, while roughly two-thirds had still not begun scaling it across the enterprise. The portfolios that will do well in this cycle are the ones that treat ERP modernization as a sequence of small, reversible decisions rather than one large, irreversible one. The ERP replacement that pays for itself is the one that happens after the operator already knows, from running agents on the current system, exactly what the new one needs to do.