Post #004 – AI In Microsoft ERP
AP automation has been around for years. AI takes it further — and the 2026 version of what’s possible in D365 looks meaningfully different from what most teams are running today.
What AP Automation Has Always Promised
The promise of AP automation has been around for a decade or more: capture the invoice, extract the data, match it to the PO, route it for approval, pay it. Reduce the manual touchpoints. Get invoices processed faster. Capture early payment discounts. Reduce errors.
Most organizations have gotten partway there. They have electronic invoice submission portals. They have OCR to pull data off paper invoices. They have routing rules for approvals. But the “match it to the PO” step — and the exception handling when things don’t match cleanly — is where the manual work still lives. That’s also exactly where AI is making the biggest difference.
The AI-Powered AP Process in D365
Here’s what a modern AI-assisted AP workflow looks like in D365 today, from invoice receipt through payment:
- Invoice arrives (email, portal, EDI, or scanned document)
- Both F&O and BC support electronic document intake via the e-documents framework. Paper invoices can be processed through OCR; the data is extracted and brought into the system.
- AI reads the invoice and extracts key fields
- Vendor name, invoice number, line items, amounts, dates. AI handles variations in formatting — because vendors don’t all use the same invoice template. This is where AI outperforms rigid rules-based OCR.
- Vendor matching and account coding
- The AI matches the vendor to your master data (handling variations in vendor name formatting), and suggests the appropriate expense accounts based on the invoice description and your historical coding patterns.
- PO matching (three-way or two-way)
- Copilot matches the invoice to the open purchase order — including handling invoices for partial shipments, multiple line items, or quantity variances. This is the step that used to generate most of the manual work.
- Human review and approval
- Clean matches route to the appropriate approver based on your workflow rules. Exceptions — invoices that don’t match cleanly — are flagged for review. Human oversight stays in the process at the approval gate.
- Payment processing
- Approved invoices post and flow through the payment run process as normal. AI doesn’t change the payment controls — it changes the work that happens before the payment decision is made.
F&O vs. BC: Where Each Platform Is Today
| Capability | D365 Finance (F&O) | D365 Business Central |
|---|---|---|
| Electronic invoice intake | E-documents framework, vendor portals | E-documents framework (enhanced in Wave 1) |
| AI invoice data extraction | Via Copilot and third-party solutions | Payables Agent (Wave 1) extracts automatically |
| PO matching | Rules-based + Copilot for exception handling | Payables Agent handles matching before goods receipt |
| Autonomous agent processing | In development (Finance Agent via M365) | Payables Agent GA in Wave 1 |
| Human-in-the-loop | Required at approval | Required at approval — agent handles prep only |
| Exception flagging | Workflow-based routing | Agent escalates exceptions for review |
The honest summary: BC’s Payables Agent in Wave 1 is further along on the end-to-end autonomous AP story right now. F&O’s AP automation is more mature in terms of the underlying infrastructure and configurability, but the agentic layer is still catching up. For large enterprise organizations on F&O with complex multi-entity AP, the current Copilot features plus strong e-invoicing setup get you very far — but full autonomy isn’t there yet in the same way.
What “Human in the Loop” Actually Means
I want to be direct about this, because I think some of the AI AP messaging glosses over it. When Microsoft says the Payables Agent “automates AP end to end with human oversight,” what that means in practice is: the agent does the intake, extraction, matching, and preparation — and a human approves before anything posts. That’s an important distinction.
This is the right design for ERP, where posting errors have real financial consequences and auditors expect documented approvals. The value isn’t removing the human entirely. The value is removing the prep work — the 20 minutes an AP clerk spends hunting down which PO an invoice belongs to, manually coding the expense line, and formatting the approval email. That prep work is where AI makes the biggest time difference.
The Native vs. Third-Party Question
If you’re evaluating AP automation, you’re going to run into third-party solutions — there’s a mature market of vendors who’ve been doing AI-assisted invoice processing for years. The question of native D365 vs. third-party is a legitimate one, and the answer depends on your volume, complexity, and how far along the Microsoft native capabilities are for your specific scenario.
What I can say is that the native story is getting stronger with every release. For organizations already deep in the Microsoft stack, there are real advantages to staying native: tighter data integration, simpler licensing, and less to maintain. But for organizations with high invoice volumes, complex PO matching rules, or a need for supplier self-service portals, the third-party ecosystem still has depth that native functionality hasn’t fully matched.
This isn’t a “pick one” — it’s worth an honest assessment of where you are and where you’re going.
📚 Go Deeper — Microsoft Resources
- BC 2026 Wave 1 — Payables Agent details
- Vendor Invoice Automation in D365 Finance
- Copilot Capabilities in D365 Finance & Operations
Bottom Line
AI-powered AP in D365 is real, it’s available now, and it’s worth evaluating seriously — especially for BC customers on Wave 1 who have access to the Payables Agent. The key is going in with realistic expectations: the value is in eliminating the prep and matching work, not in removing approval controls. And like everything in ERP, clean data and well-configured processes will determine whether you get 80% straight-through processing or 40%.
Next up: Post 5 tackles one of the questions I get most often — can AI actually help with period close?
BB
Bobbi Bricker
ERP Capability Lead & D365 Functional Architect at Centric Consulting. Former controller. Practical by nature, curious by default. Writing about D365 F&O, Business Central, and now AI in ERP — because someone has to translate the tech into something finance teams can actually use.
Thank you for reading!
Interested in learning more? Below are a few of my latest posts:
- AI and ERP Security: What Copilot Means for Your D365 Security Roles and Internal Controls
- The Natural Language ERP: Stop Running Reports, Start Asking Questions
- AI Adoption in ERP: Why Change Management Is Your Most Critical AI Investment
- Agent 365: Microsoft’s Control Tower for All Your ERP Agents
- AI in D365 Supply Chain: From Demand Planning to Warehouse Intelligence
Leave a Reply