Is Your ERP Ready for AI? The Criteria That Matter in 2026

The executive committee has just approved a budget for AI agents. Three floors down, the same company has been postponing the cleanup of its supplier master data in the ERP for two years. Nobody has connected the two decisions yet.

McKinsey has now put a number on that blind spot, and the name it gives it sums up 2026: the great divide between AI agents and the ERP.

“Only about 40 percent of companies report any enterprise-level EBIT impact from their AI initiatives, and most attribute less than 5 percent.” Source: McKinsey, Bridging the great AI agent and ERP divide, January 2026

Why is your ERP not ready for AI despite growing budgets?

Because the money is migrating to generative AI while investment in the technology foundation declines: McKinsey finds that almost half of IT organizations plan to invest in generative AI, with investment dropping significantly for core capabilities such as infrastructure and architecture. The firm calls this imbalance the great divide.

The consequence has a name too: pilot purgatory. Use cases multiply without the end-to-end processes, data and systems that would let them scale.

The finance function shows the pattern clearly. According to the State of AI in Finance 2026 report by CFO Connect, 56% of finance teams now use AI, but 45% remain in pilot mode and only 17% have embedded it in core workflows.

What does an AI-ready ERP look like, concretely?

An AI-ready ERP shows four signs: reliable master data, processes documented end to end, up-to-date approval rules and data accessible through APIs. Nothing spectacular, and that is the point: AI runs on what the ERP already structures.

McKinsey reminds us that the ERP holds the company’s operating DNA: the process knowledge, data structures and business logic accumulated over years. AI agents do not replace that foundation, they extend it.

And they are arriving, ready or not. Gartner predicts that 40% of enterprise applications will integrate task-specific AI agents by the end of 2026, up from less than 5% in 2025. We detailed what SAP, Microsoft and Oracle are already shipping inside their ERPs.

How do you measure the gap between your ERP and your AI ambitions?

By comparing your situation to published benchmarks rather than vendor demos. The 2026 numbers describe a market accelerating on the surface and stalling underneath:

• only about 40% of companies report any EBIT impact from AI, and most put it under 5% (McKinsey, January 2026);
• 40% of enterprise applications will integrate task-specific AI agents by the end of 2026, up from less than 5% in 2025 (Gartner, August 2025);
• over 40% of agentic AI projects will be canceled by the end of 2027, for lack of clear value or sufficient risk controls (Gartner, June 2025);
• 43% of SAP customers have AI use cases in production, but 77% of those scenarios run on non-SAP solutions (DSAG, February 2026);
• in Canada, 19.2% of businesses used AI in the second quarter of 2026, and the rate reaches 40.4% in finance and insurance (Statistics Canada, June 2026);
• in Quebec, 12.7% of businesses used AI in production in the second quarter of 2025, and expected use for the following 12 months (13.1%) is growing more slowly than in Ontario (16.5%), according to the Institut de la statistique du Québec (November 2025).

A note on method: these figures come from public surveys and analyst firms, with scopes that vary from one study to the next. None of them describes your specific situation.

The full Canadian picture is in our analysis of AI adoption in business in Canada.

The 2026 bottleneck is not model power. It is the state of the system that serves as their memory.

Where do you start preparing your ERP for AI without rebuilding everything?

One domain at a time: a bounded process, its master data validated, its governance rules written, then a measured pilot before any rollout. McKinsey observes that companies getting 5% or more EBIT impact from AI transform a full domain rather than stacking isolated use cases.

In the ERP engagements PlanAxion leads with B2B distributors and mid-sized organizations in Quebec, the same diagnosis returns: customer and supplier master data has not been reviewed since the implementation. Duplicates, outdated payment terms, approval hierarchies that no longer match the org chart.

One process is especially well suited to the role of first domain: cash application. Bounded, measurable, fed by the ERP’s history. That is the pool the cash application solutions PlanAxion implements target first, and the same reasoning guides the choice of the right automation layer.

To turn that finding into a roadmap, that is the approach of our AI workshop: 4 weeks, five steps (prepare, identify, prioritize, validate the data, decide and deliver), one process at a time. The investment varies with scope and is confirmed during a short exploratory call.

Should you pause your AI projects until the ERP is modernized?

No. The two workstreams advance together, provided you pick one domain and finish it. McKinsey notes that AI agents themselves become a lever for modernizing the ERP, notably to document processes and clean up data. The right reflex is not choosing between AI and the foundation: it is refusing any agent project that cannot name the data it depends on.

Frequently asked questions about AI-ready ERP

What is an AI-ready ERP?

An AI-ready ERP gives agents reliable master data, documented processes, up-to-date approval rules and data accessible through APIs. AI does not require a brand-new ERP: it requires a foundation whose data and business logic can be trusted, because agents extend the ERP rather than replace it.

Do you need to replace your ERP to adopt AI agents?

No. The major vendors ship their agents inside existing suites, and a replacement is justified by other criteria: obsolescence, maintenance costs, structural functional limits. Preparation runs through master data, process documentation and governance, not through a full reimplementation.

Why do so many AI agent projects fail?

Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls. An agent wired to stale data accelerates errors instead of fixing them, which is why the foundation comes first.

How long does it take to prepare an ERP for AI?

It depends on the scope. A bounded domain, such as supplier master data or cash application, can be prepared in a few weeks to a few months. A 4-week prioritization workshop is often enough to pick the first domain, measure the data gap and sequence the work.