Skip to content
AI Workshop

AI Agents in ERP 2026: SAP Joule with Claude, Dynamics 365, Oracle and Odoo Compared

ERP is where a company's money, stock, orders and people are recorded, which makes it the place where AI agents could save the most time and do the most damage. In 2026 all the big vendors moved: SAP made Anthropic's Claude a primary reasoning engine of Joule, Oracle launched 22 Fusion Agentic Applications and Microsoft put agents into Dynamics 365 and Business Central. Customers are more cautious. In the DSAG Investment Report 2026, only 3% of SAP users with AI use cases run them on SAP Business AI; 77% use other tools. This guide explains what ERP agents can do today, how SAP, Oracle, Microsoft and Odoo compare, which use cases pay back and how a Croatian or DACH company should start.

3% vs 77%
DACH SAP users with AI in production: on SAP Business AI vs on non-SAP tools (DSAG Investment Report 2026)
22
Fusion Agentic Applications Oracle released in March 2026 across finance, HR, supply chain and CX
9–10
Months to payback for data extraction, AP/AR automation and report creation (Gartner, Sep 2026)

From copilot to agent: what changed in 2026

The first wave of ERP AI, in 2023 and 2024, was mostly copilots: a chat window that explains a report, drafts an email to a supplier or finds a transaction. Useful, but the human still did the work. The 2026 wave is about agents: a model that reads ERP data, calls ERP functions and completes a multi-step task, such as matching 400 supplier invoices to purchase orders, preparing the reconciliation for the month-end close or rerouting an order when a supplier is late. The person checks and approves instead of typing.

Two technical shifts made this possible. Models got much better at long, tool-heavy tasks (our article on agent task horizons shows how fast), and the Model Context Protocol (MCP) gave vendors a standard way to expose ERP functions to any model. SAP now uses MCP to connect Claude with S/4HANA, SuccessFactors and Ariba.

The vendors: SAP, Oracle, Microsoft and Odoo

VendorAI layerModel partnersStatus, November 2026
SAP (S/4HANA, SuccessFactors, Ariba)Joule assistant, Joule agents, SAP Business AI PlatformAnthropic Claude as primary reasoning engine (announced 12 May 2026); Mistral and others via the generative AI hubSAP cites more than 200 agents and about 50 Joule Assistants; many are still on the way to general availability. Joule is tied to SAP's cloud offerings (RISE and GROW with SAP)
Oracle (Fusion Cloud ERP, HCM, SCM, CX)Fusion Agentic Applications, AI Agent StudioSeveral model providers through Oracle Cloud22 agentic applications released on 24 March 2026, 5 more for CX on 9 April; AI Agent Studio at no extra cost
Microsoft (Dynamics 365 Finance, Business Central)Copilot and agents in Dynamics 365; Finance agent in Microsoft 365 CopilotOpenAI models; Anthropic Claude is also available in Microsoft 365 CopilotBusiness Central 2026 release wave 2 (v29) from October 2026 with Payables Agent improvements; Account Reconciliation Agent in preview for D365 Finance
Odoo (popular with SMEs)Ask AI, AI agents and AI fields since Odoo 19 (September 2025)Connects to external models such as OpenAIAgents configured without code; strong fit for smaller companies

SAP and Anthropic is the most important deal of the year for ERP. SAP's CTO Philipp Herzig described Claude as a primary reasoning and agentic capability across SAP's portfolio, with examples such as a treasury manager asking Joule for a CFO briefing that takes minutes instead of hours, or closing the books at quarter-end. Forrester called SAP's "Autonomous Enterprise" vision credible but warned about concentration risk: if the reasoning layer of your ERP depends on one model provider, so do you. For the background on Claude, see our Claude model selection guide.

OpenAI reaches ERP mainly through Microsoft, whose Copilot is built on OpenAI models. Mistral plays the sovereignty card: its models are available in SAP's generative AI hub and can run in European data centres, which matters for public sector and regulated companies. General-purpose agents such as Manus are useful for research and documents, but they are no substitute for agents that work inside the ERP with its permissions, audit log and approval rules.

The reality check: customers build AI around SAP, not in it

The German-speaking SAP user group DSAG surveyed 198 member companies (73% in Germany, 12% in Switzerland, 10% in Austria) between December 2025 and January 2026. 43% have implemented AI use cases; 51% have not. But the AI runs mostly outside SAP: 77% of companies with use cases have them in production on non-SAP solutions, against only 3% on SAP Business AI. Even in proofs of concept, where SAP does best, non-SAP tools lead 62% to 26%.

Where DACH SAP Users Run Their AI Use Cases, by Stage

Source: DSAG Investment Report 2026 (press release, 26 Feb 2026). 198 SAP user companies in Germany, Austria and Switzerland, surveyed 8 Dec 2025 to 21 Jan 2026. Share of companies that have implemented AI use cases (43% of respondents); multiple answers possible.

Why? The top reasons DSAG members give for their SAP investment decisions are cost: 79% name the cost-effectiveness of SAP software and 79% the economic situation, followed by licence and contract design (70%). AI that requires a new cloud contract, new licences and a migration is a hard sell when budgets are under pressure. A Microsoft Copilot or Claude licence that works with exported data is quicker to buy.

What Drives SAP Investment Decisions in 2026 (share of DSAG members)

Source: DSAG Investment Report 2026, top challenges influencing SAP investment decisions for 2026. n = 198.

The migration problem: Joule needs the cloud, most customers are not there

The second obstacle is the installed base. Joule and most of SAP's new agents are tied to SAP's cloud offerings, and classic on-premise installations get no native Joule access. Among DSAG members, 56% run S/4HANA on-premise and 54% still run SAP ECC or older Business Suite releases, while only 17% use S/4HANA Private Cloud and 5% Public Cloud. The move is under way (ECC fell from 68% to 54% in two years) but most companies will not have SAP's agents in reach for a while.

SAP ERP Systems in Use at DSAG Member Companies, 2024 vs 2026

Source: DSAG Investment Reports 2024 and 2026. Multiple answers possible because companies run several systems in parallel during migration.

Budgets will not solve this quickly. 38% of DSAG members raise their total IT budget in 2026, but 24% cut it; 43% raise SAP spending and 28% reduce it. That is more polarised than in 2024, when only 18% cut IT budgets. The practical result is that many companies will add AI next to their ERP, through APIs, MCP connectors and data extracts, while the core migration continues.

IT and SAP Budget Plans for 2026 vs 2025 (share of DSAG members)

Source: DSAG Investment Report 2026. Remaining percentages: no answer.

Use cases that work today

Gartner's survey of 160 senior finance leaders (January to April 2026) is the best guide to where ERP AI pays off: data extraction, accounts payable and receivable automation and report creation generally return their cost within nine to ten months. Gartner also found that 63% of finance organisations said AI implementation in 2025 was slower than expected, so start with the proven cases:

  • Accounts payable. Reading invoices, matching them to orders and goods receipts, flagging differences. This is what Microsoft's Payables Agent does in Business Central, and it is the most common first project. Our guide to intelligent document processing covers the technology.
  • Reconciliation and month-end close. Matching bank statements and intercompany entries, explaining variances, preparing the close checklist. See also AI in accounting and audit.
  • Collections and receivables. Prioritising overdue invoices, drafting reminders in the customer's language, proposing payment plans.
  • Procurement. Comparing offers, checking contract terms, creating purchase requisitions from free text. Our article on AI procurement agents goes deeper.
  • Management reporting. Asking the ERP questions in plain language and getting a briefing with the numbers, instead of waiting for a controller to build an Excel file. Our text-to-SQL article shows how accurate this is.

Guardrails are not optional in ERP. An agent that books, pays or changes master data needs the same controls as a new employee: least-privilege permissions, approval for payments and above a value threshold, a full audit trail and separation of duties. Agents should never get a generic admin account. Our guides on agent identity and guardrails in production explain how to set this up, and the EU AI Act requires human oversight for high-risk uses such as HR decisions.

What this means for companies in Croatia and DACH

  • Do not wait for your ERP vendor. If you are on ECC, S/4HANA on-premise, an older Dynamics NAV or a local Croatian ERP, you can still automate invoice processing, reconciliation and reporting today with Claude or another model connected through APIs or MCP. That is what 77% of AI-active SAP users already do.
  • Use the migration as the moment to redesign. If an S/4HANA or cloud ERP project is planned anyway, put the agent use cases into the business case. Embedded agents cost less to maintain than custom integrations once you are on the new platform.
  • Choose by the processes, not the demo. Ask vendors which agents are generally available today, which licences they need and how they handle permissions and audit. Many of the announced agents are still on the roadmap. Our AI vendor selection guide has a checklist.
  • Plan for more than one model. SAP uses Claude, Microsoft mostly OpenAI, and Mistral offers European hosting. Keep your own integrations model-agnostic so you can switch, a lesson Forrester's concentration-risk warning makes clear.
  • Get the data right first. Agents are only as good as master data, account structures and document quality. Our article on AI-ready data shows what to fix before you start.
  • Measure value, not deployments. Gartner warns that counting pilots or tools is not proof of value. Track hours saved per closed month, days sales outstanding, error rates and cost per invoice. See our AI ROI guide.

The bottom line

2026 is the year ERP vendors turned AI assistants into agents, and SAP's choice of Claude shows that the best reasoning models are now part of the core business system. But adoption lags far behind the announcements: in the DACH region, AI around the ERP is more than twenty times more common in production than AI inside SAP. For most companies the sensible path is twofold: start now with high-payback finance processes using models connected to the ERP you already have, and design the embedded agents into the migration you are planning anyway, with permissions, approvals and audit from day one.

Bring AI agents into your ERP processes

AI Workshop connects Claude and other models to SAP, Dynamics, Odoo and local ERP systems, starting with invoice processing, reconciliation or reporting. Our Claude Certified Architects build agents with the right permissions, approvals and audit trail, and measure the result in hours and euros.

Talk to an AI consultant

Boris Agatić

Founder of AI Workshop and Claude Certified Architect. Over 10 years of building software, guiding companies from the first strategy call to production AI agents.

LinkedIn (opens in a new tab)