Key takeaways
- Autonomous AI agents in Business Central act independently within defined guardrails — and only request human approval when needed.
- The Sales Order Agent creates sales orders from incoming emails; the Payables Agent processes supplier invoices via OCR and AI-based account assignment.
- The Agent Designer lets companies and partners create custom AI agents without coding.
- All agents require Business Central Online (cloud) — on-premises is not supported.
- Good results require well-maintained master data, a Microsoft 365 mailbox and activated Copilot billing.
What are AI agents in Business Central?
The crucial difference to Copilot: an agent doesn't wait for an instruction. It acts independently within defined guardrails — and only asks for human approval when it needs it. For mid-sized companies working with lean resources in sales and accounting, this is a significant development.
Microsoft calls the concept 'agentic ERP'. It doesn't mean the AI makes decisions — it means the AI autonomously handles routine work and leaves the review and sign-off to the specialists.
Sales Order Agent — sales orders from emails
The Sales Order Agent monitors a designated email mailbox — for example orders@company.com. When a customer inquiry arrives, it analyses the content, identifies the customer in Business Central master data, extracts items, quantities and delivery requirements and creates a sales order draft.
The agent automatically checks inventory availability, proactively sends clarification emails when information is missing and — depending on configuration — creates a quote or an order draft. Final approval stays with the sales representative.
Payables Agent — processing supplier invoices
The Payables Agent processes incoming supplier invoices. It retrieves PDFs from an email mailbox, reads invoice number, date, amounts and line items using OCR and AI, and suggests the appropriate G/L accounts based on posting history.
It recognises documents already processed in shared mailboxes, creates purchase invoice drafts for review and learns from the existing account assignment history. The actual posting remains with the finance team.
Agent Designer — building your own AI agents
With Agent Designer, companies and partners can create their own AI agents without coding for the first time. In a graphical interface you define in natural language what the agent should do, which data it may read and write, and when it must request human approval.
For more complex requirements the prototype can subsequently be converted into an AL extension. Setup takes place in a sandbox environment. You need the AGENT-ADMIN permission for designing and AGENT-DIAGNOSTICS for monitoring and troubleshooting.
Sales Order Agent vs. Payables Agent — comparison
Both agents follow the same principle — autonomous processing with human oversight — but they serve different processes and have different strengths.
- Sales Order Agent: processes emails with order requests, identifies customer, items and quantities, creates sales quotes or orders, sends clarification emails when information is missing.
- Payables Agent: processes emails with PDF invoices, identifies supplier, invoice data and line items, creates purchase invoice drafts, flags ambiguities for manual resolution.
- Both require well-maintained master data — the Sales Order Agent needs correct customer email addresses, the Payables Agent needs consistent posting history for automatic account assignment.
Requirements for deployment
AI agents require Business Central Online (cloud) — they are not available for on-premises. You need a Microsoft 365 mailbox with 'Read and Manage' permission, activated Copilot billing in the admin centre and the permissions AGENT-ADMIN for setup and AGENT-DIAGNOSTICS for monitoring.
On the data side, agents are only as good as your master data: customer records with maintained email addresses, item records with unambiguous descriptions and posting history for automatic account assignment.
Regarding data protection: the agents use the Azure OpenAI Service. Depending on the tenant's hosting region, data may be transferred to other Azure regions. The administrator must explicitly consent. For EU customers we recommend setting data residency to the EU region.
Our assessment: where agents help and where they don't
AI agents pay off most at high repetition frequencies: companies processing dozens of similar orders or invoices daily benefit the most. Standardised inputs and a lean headcount in inside sales or accounting are the best prerequisites.
Agents are less suited to complex inquiries with multi-tier discounts, customer-specific configurations or volume pricing. With poorly maintained master data the agent creates more rework than relief. And anyone still running on-premises cannot use them — yet another reason to evaluate cloud migration.
Frequently asked questions
- No. Copilot features and autonomous agents are exclusively available in Business Central Online (cloud). On-premises installations cannot use these features.
- The agents use so-called Copilot credits, billed via the Business Central Admin Centre. The exact consumption depends on usage volume. The Business Central licence itself already includes a base allowance.
- No. In the default configuration the agents create drafts — quotes, orders or invoice drafts. Final approval and posting is always performed by an authorised employee.
- The agents use the Azure OpenAI Service and support all common business languages, including German and English. The tenant's configuration language is automatically applied.
- The technical setup — connecting the mailbox, assigning permissions, activating the agent — takes just a few hours. The more time-consuming part is preparation: cleaning up master data, running tests in the sandbox and coordinating with the business team.