← All posts

May 21, 2026 · 11 min read

The end of ERP as we know it? McKinsey's thesis, through Soberan's lens

Soberan product screens showing ERP, CRM, and agent execution workflows
Soberan's view: ERP remains the record, while agents become the governed execution layer across ERP, CRM, and Contact Center.

What McKinsey got right

The strongest part of the McKinsey thesis is that ERP disruption is architectural, not cosmetic. Adding a conversational assistant to an old screen is not the shift. The real change is that AI agents can sit above the system foundation, coordinate work across domains, and turn intent into governed action.

McKinsey also separates two futures that buyers should not confuse. One version says agents replace ERP entirely. The more durable version says ERP becomes more headless: the clean core, data model, controls, and audit trail still matter, but users interact through semantic and agentic layers instead of through transaction-heavy screens.

That is close to how Soberan thinks about the category. ERP is not just a database or a finance suite. It is the operational memory of the business. AI makes that memory useful when agents can read it, respect policy, and move the next step forward.

ERP is not dying. The screen-first operating model is.

For decades, ERP value has been trapped behind screens, menus, roles, forms, and implementation logic. The system could store the truth, but a human still had to know where to click, which report to pull, which exception mattered, and which downstream team needed the update.

Agentic AI changes the interface from navigation to delegation. The user should not have to open five screens to answer whether an order can ship, whether a customer can be extended credit, or whether a supplier delay will create a stockout. The agent should assemble the context, apply the operating policy, propose or execute the action, and leave the record behind.

That is why the future ERP conversation is not only about better UI. It is about changing the locus of control. Humans set goals, policies, approvals, and exceptions. Agents perform the repeatable work. ERP remains the controlled record that proves what happened.

The missing piece: operations do not stop at ERP

One place where the ERP debate often stays too narrow is the boundary around ERP itself. Real operating work rarely lives inside one module. A blocked order is not only an order-management issue. It may involve inventory, credit, sales promise, customer service, logistics, finance, and a WhatsApp conversation with the customer.

A renewal risk is not only a CRM issue. It may involve open invoices, delayed shipments, low product availability, unresolved support cases, and a sales rep who has not followed up. A collections case is not only an AR issue. It may involve contact center channels, invoice context, dispute routing, payment plan policy, and customer history.

This is the Soberan view: the agentic layer cannot be trapped inside ERP alone. It has to connect ERP, CRM, and Contact Center because that is where the operating loop actually lives. The work starts in one function and finishes in another.

Implementation economics will change, but not because AI is magic

McKinsey argues that AI can materially reduce ERP implementation effort and duration. We agree with the direction. Agents can accelerate process discovery, configuration, testing, documentation, training preparation, migration checks, and support workflows.

But there is a trap in treating implementation speed as the whole story. Faster implementation only matters if the operating model after go-live is better. Otherwise companies get a cheaper version of the same old problem: a system that records work but still requires people to chase status, reconcile exceptions, and move decisions manually.

The real gain comes when implementation and operations converge. The same process map that configures the system should also become the policy map that agents use to run daily work. The same tests that validate go-live should become ongoing controls. The same training content should become in-system guidance and exception handling.

Build versus buy will split by layer

McKinsey's fifth disruption is that ERP value creation shifts from build to buy. We think the more precise answer is that companies will split the stack. They will buy commodity execution patterns and governance primitives, while keeping their differentiating policy, operating judgment, customer strategy, and business ontology.

A manufacturer should not need to build from scratch an agent that follows up on supplier confirmations, checks purchase orders, validates inventory risk, or routes a payment dispute. Those patterns are repeatable. But the manufacturer should absolutely control which suppliers matter, which customers receive priority, which exceptions need approval, and what tradeoffs are acceptable.

The winners in AI-native ERP will not be the vendors with the flashiest assistant. They will be the vendors that make governed execution simple: clear permissions, readable audit trails, policy-aware workflows, human escalation, and measurable operating outcomes.

What buyers should ask now

  • Can the system show a full workflow from signal to action to record, not only a chat answer?
  • Can the agent operate across ERP, CRM, Contact Center, and external systems without losing context?
  • Where does the system draw the line between automatic execution, prepared action, and human approval?
  • Can every agent action be audited: source data, policy used, output generated, approval path, writeback, and exception reason?
  • Does the vendor measure operating outcomes such as cycle time, cash collected, stockouts avoided, orders released, tickets resolved, or approvals cleared?
  • Can the company start with one workflow and expand without creating a mess of disconnected agents?
  • Does the architecture protect the clean core while still allowing agents to execute real work?

One operating layer

Soberan is built around the idea that ERP, CRM, Contact Center, and the agent should not be separate islands. The customer promise, the inventory record, the invoice, the service case, and the approval policy belong in one operating loop.

That is why Soberan Agent is not positioned as a chatbot on top of ERP. It is an execution layer. It can help qualify leads, follow up with customers, work collections, route support, prepare purchase actions, detect inventory risk, move order exceptions, and escalate the work humans should decide.

The important part is not that the agent speaks. The important part is that it acts with context and control. It reads the same operating memory your team uses, follows policy, writes back to the right record, and exposes the handoff when judgment is needed.

The practical takeaway

McKinsey's article is a useful signal because it moves the ERP conversation beyond simple automation. The next era is not ERP plus AI features. It is ERP as part of an AI-governed operating system for the business.

For leaders, the immediate move is not to launch dozens of pilots. It is to choose one workflow where the record, policy, action, and outcome are clear. Then prove the loop: detect the signal, gather context, execute or prepare the action, route exceptions, write back, and measure the result.

The end of ERP as we know it is not the end of ERP. It is the end of ERP as a place where humans spend their day clicking through transactions. The new question is how much of the daily operating work your system can run with policy, proof, and human control.

Sources and context