Three AI agents are now live in Business Central, with the Expense Agent entering public preview this month. Within a week of the May 8 release, half a dozen controllers I work with had asked the same question.
When do we turn them on?
It is a fair question, and the honest answer is more complicated than the marketing suggests. Microsoft has been clear about the framing: Business Central “accelerates the move to agentic ERP with enhancements to our AI-powered agents that automate sales and purchase scenarios in 2026 release wave 1” (Microsoft Dynamics 365 Blog, March 18, 2026). Three named agents are now available: the Payables Agent, the Sales Order Agent, and (in public preview) the Expense Agent. Agent Designer reached general availability this month for partners and customers who want to build their own.
That is real. So are the limits, the costs, and the workflows that have to change around them.
In 35 years as a controller and consultant, I have watched a lot of “this changes everything” software land in finance departments. Some of it did. Most of it needed a more honest conversation than the marketing offered. This post is that conversation for Business Central’s new agentic features. What is generally available. What each agent does and does not do. What it costs to run. And a seven-step checklist a finance leader can use before any of it goes live.
What “Agentic ERP” in Business Central Means
The phrase matters because it sets the expectation. Here is how Microsoft defines it in the Wave 1 release plan:
“We’re moving Business Central toward AI-driven ERP by embedding AI and automation into everyday business processes. We’re helping organizations move from manual tasks to autonomous, insight-driven operations. Business Central increasingly acts as an active partner in daily work, helping you analyze data, take action, and make decisions with greater confidence and efficiency.”
(Microsoft Learn, 2026 Wave 1 plan)
Strip the marketing register and what you have is software that operates inside Business Central under the same security model and permissions as a human user. It watches for work (an inbound email, a customer request), proposes an action (a vendor invoice draft, a sales quote), and routes the result to a person for review. Microsoft frames the architecture as layered: BC agents live inside the product as the execution layer, with cross-system orchestration handled by Copilot Studio above them (Microsoft Learn, Agent Designer release plan).
One constraint to put on the table before we go further: agents are online-only. The Wave 1 overview states that “Copilot and agents are available only to Business Central online customers.” If your organization is still on Business Central on-premises or partner-hosted, none of what follows applies until you migrate to SaaS.
The Payables Agent: What it Does (and Doesn’t)
The Payables Agent is the one most finance teams will ask about first. The pitch is direct: end-to-end processing of vendor invoices. Watch a mailbox, read the PDFs (using Azure Document Intelligence under the hood), match vendors and accounts, draft the invoice, surface to a supervisor.
What it actually handles end-to-end is narrower than the pitch. Read the documentation page and the limitations are listed right there:
- Only processes emails with PDF attachments
- Skips emails with more than 10 attachments
- Won’t process PDFs larger than 5 MB
- Won’t process PDFs with more than 10 pages
- Caps at 100 emails per day
- Caps at 500 invoices per day
The same page lists what the Payables Agent does not yet support: “Purchase order matching / Approval flows / Anomaly detection.” (Verify against the live Microsoft docs page at publication — PO matching was reportedly entering public preview in early 2026 and this list may have been updated.)
If you run a three-way-match AP process, read that last list twice. No PO matching means the Payables Agent today is a drafting assistant, not an AP automation platform. It will get the invoice into Business Central faster and probably more accurately than manual entry. It will not match that invoice to a purchase order. It will not route it through your approval hierarchy. It will not flag a duplicate or an unusual amount. Those workflows still belong to your existing process and your existing people.
The volume caps are worth checking against your actual numbers. A $20M services firm processing 30 vendor invoices a day will never see them. A $200M distributor pushing 800 invoices a day across business units is designing around the 500-invoice limit from day one.
One more constraint: “This Copilot feature is validated and supported in English only. While it can be used in other languages, it might not function as intended.” If your AP team handles invoices in French, Spanish, or German, that needs to be in your evaluation, not discovered in production.
The Payables Agent is generally available in all countries and regions per the Copilot and Agents availability table, with the English-only validation caveat above.
The Sales Order Agent: Where CTP Integration Earns Its Place
The Sales Order Agent is the one I find more interesting from a margin and promise-date perspective. The agent reads inbound customer emails, locates the customer in Business Central, holds multi-turn conversations to clarify missing details, checks item availability, and drafts a sales quote.
The detail that earns this agent its seat at the table, especially for manufacturers and distributors, is the integration with Capable-to-Promise (CTP). From the docs:
“The Sales Order Agent includes an option to use Business Central’s capable-to-promise (CTP) functionality that lets the agent calculate the earliest possible delivery date for items not currently in stock. CTP evaluates production capacity, procurement timelines, and supply chain constraints to determine when an item can realistically be delivered.”
For a manufacturer, that is the part that pays. A salesperson who can promise an accurate delivery date on a low-stock item without a 30-minute back-and-forth between sales, planning, and purchasing is the difference between winning the order and losing it to a competitor who answered first. The agent putting a defensible date in front of a customer in 90 seconds rather than 90 minutes is real operational value.
There is a governance affordance worth naming explicitly: “The agent always involves designated Business Central users to review and approve all outgoing messages before they’re sent to the customers.”
Every outbound message goes through a human gate. That is by design, and it is the right design. It also means the agent does not “automate sales orders end-to-end” in the sense some marketing copy implies. It drafts. You approve. It sends.
Language support is broader than the Payables Agent: Danish, Dutch, English (multiple regional variants), Finnish, French, German, Italian, Norwegian Bokmål, Spanish, and Swedish are all on the availability table.
Expense Agent: a New Cost Model Worth Noting
The Expense Agent is the newest of the named agents. Microsoft’s Aleksandar Totovic announced it on April 27, 2026, with public preview starting in May 2026. It captures receipts, categorizes expenses, handles mileage and per diem, routes project expenses to the right ledgers, and reinvoices billable expenses automatically.
Two things make the Expense Agent worth a separate paragraph.
First, it is the first BC agent designed for people who do not have a Business Central license. Microsoft states it plainly: “No Business Central licenses required.” Field technicians, drivers, traveling salespeople, project consultants. The agent is positioned for users who never log into BC. They submit receipts through a dedicated web app, Outlook, or Microsoft 365 Copilot Chat, and the expenses arrive in BC for accounting to handle.
For organizations whose field workforce historically used external expense tools (Concur, Expensify, Zoho Expense), this changes the renewal conversation.
Second, usage is metered through Copilot Credits, not per-seat BC subscriptions. That is a meaningful business model shift. You can extend expense capture to a population of non-ERP users without paying for seats. You are also now metering usage on a different financial axis, which means a new line in your budget that behaves differently from per-user licensing. More on that below.
A constraint to flag: at preview launch, the Expense Agent is US-only and English-only. Microsoft has indicated additional languages and regions starting July 2026. If you operate internationally today, the Expense Agent is a US pilot conversation, not a global rollout.
Agent Designer: Custom Agents and the Build-or-Wait Decision
Agent Designer reached general availability in May 2026 after a public preview that began in early February. Microsoft’s framing of the audience is broader than just developers:
“This approach empowers partner personas like consultants, product owners, and domain experts, as well as developers and even power users, to explore new scenarios, validate ideas, and build confidence in their agents before moving to production.”
Translation: Agent Designer is meant to let people who are not full-stack AL developers prototype custom agents using natural-language instructions. Custom agents extend the same runtime as the built-in Payables and Sales Order agents, with the same security and permission model.
Microsoft MVP Stefano Demiliani has documented that BC update 28.1, rolling out in late May and June 2026, adds the production-deployment path for custom agents: “Dynamics 365 Business Central update 28.1 will allow you to deploy your own Business Central agents to production environments” (Demiliani, May 10, 2026).
If you are tempted to start designing a custom agent right now, the controller-grade question is the same one I have asked about every customization in 35 years: what does this replace, what does it cost to maintain, and who owns it when the consultant goes home?
A custom Payables-style agent that handles a specific category of vendor invoices the built-in agent cannot read is exactly the right scope for a first build. A custom agent that orchestrates your entire month-end close is, today, ambitious.
For partners and customers thinking about quality assurance, Microsoft also shipped BC-Bench in Wave 1. It is a benchmarking framework for evaluating AI coding agents against real Business Central AL development tasks. Developer-facing tool. The fact that it exists at all is a useful signal: Microsoft is building the infrastructure to measure whether AI-built agents actually meet the quality bar before they reach production.
Copilot Credits: What This Actually Costs to Run
Here is the angle most BC partner blogs are not covering, and it is the one a CFO will ask about first. The built-in agents and any custom agents you build with Agent Designer all consume Copilot Credits. Not flat per-user licenses. Not per-invoice fees. Credits. From the Agent Designer documentation:
“As with the built-in Sales Order and Payables agents, custom agents consume Copilot credits as they perform their steps. Credits can be provisioned through prepaid or pay-as-you-go models, providing flexible cost management as your AI usage scales.”
That phrasing has implications a controller should think through before signing the procurement order.
Consumption is variable, not flat. Demiliani has documented that credit consumption “depends on the feature used and the actions that your agent performs” (Demiliani, March 24, 2026). A simple AP draft and a multi-turn sales conversation with CTP lookups do not consume the same credits. Your monthly cost depends on transaction mix and volume. Which means it is variable. Which means you are building a new cost model alongside your existing per-seat licensing.
Cost visibility is permissioned, not default. Demiliani also notes that in the current implementation, only the SUPER and AGENT-DIAGNOSTICS permission sets can view billing details inside BC. The standard AGENT-ADMIN permission set is insufficient. Practically: if you want your AP supervisor to monitor what the Payables Agent is costing per week, you have a permission-set design decision to make before turning the agent on, not after.
Prepaid vs. pay-as-you-go is a real budgeting question. Microsoft offers both. Prepaid gives you predictable budget control and a service-pause failure mode when credits run out. Pay-as-you-go gives you elastic capacity and an unbounded invoice. Neither is wrong. The right answer depends on whether your finance organization is more uncomfortable with running out of credits mid-month or with not knowing what next month’s invoice will be.
If your CFO asks for a per-invoice cost number, today there is not a published one. Microsoft’s documentation describes the model, not the unit economics. The honest answer is “we have to pilot it on a representative sample of our actual invoice volume and measure.”
What the Numbers Say
Microsoft published a Forrester Total Economic Impact study in March 2026 with figures partners are now citing in sales decks. The headline numbers: over 200% ROI over three years, $460K net present value, six-month payback, up to 30% reduction in monthly close time, up to 50% time savings for AP, AR, and billing.
Two things to read twice on a study like this.
First, it is a Microsoft-commissioned Forrester Consulting study, not independent Forrester research. That is disclosed in the study itself and it does not make the numbers wrong, but a controller weighing investment decisions should know where the analysis came from. Vendor-commissioned TEI studies are useful directional benchmarks, not substitutes for your own pilot data.
Second, the methodology is a composite-organization model based on four customer interviews. The composite is described as $50M annual revenue, 300 employees, 15 core finance/accounting users, 100 light users, deployed in the cloud. If your organization is materially different in scale, vertical, or business model, the gap between the composite and your reality is where the projections become aspirational rather than predictive.
The steel-manning version of the citation, if you have a CFO conversation coming where these numbers will land in front of you: “Forrester’s Microsoft-commissioned study, based on four customer interviews aggregated into a composite mid-market firm, projects greater than 200% ROI on a three-year horizon. Our pilot should measure whether our specific workflow and volume produce a comparable result.” That framing keeps the directional benefit on the table without overcommitting on unit economics that have not been tested in your environment.
The Controller’s Pre-flight Checklist
If you are responsible for the decision and you want a structured starting point, here is the seven-step pre-flight I would use before turning any of this on, drawing on what Microsoft itself documents about the current state of these agents.
1. Confirm your deployment topology. Copilot and agents are online-only. If you are still on-premise or leveraging partner-hosted BC, the question is a migration question first, not an agent question. Check our notes on BC hosting models for the broader context.
2. Measure your actual volume against the documented limits. Pull last quarter’s invoice count per day, sales order count per day, average attachment count, and PDF page count. Compare to: 100 emails/day, 500 invoices/day, 10-page PDFs, 5 MB max, 10 attachments per email for Payables. Will the agent cap on a normal Tuesday or only on a peak day? The right answer changes which agent you start with.
3. Map your three-way-match dependency. If you require purchase order matching, approval routing, or anomaly detection on AP, the Payables Agent today does not provide any of these. Decide whether the agent reduces enough manual work upstream to be worth deploying alongside your existing AP workflow, or whether you wait for those features.
4. Decide your Copilot Credit model before the pilot, not after. Run a one-week pilot on a controlled subset of invoices or sales orders, measure credit consumption, project monthly cost at full volume. Then choose prepaid or pay-as-you-go based on which failure mode your finance team is more comfortable with. Either model is defensible. Picking one in the dark is not.
5. Configure billing visibility for the people who need it. Set up AGENT-DIAGNOSTICS permissions for whoever in finance will monitor cost. Today the standard AGENT-ADMIN permission set is not enough to see billing detail. This is a permission-design decision that has to happen before the agent goes live, not after the first invoice surprises someone.
6. Plan the human-review workflow before go-live. Both the Payables Agent and the Sales Order Agent route output to designated reviewers. Decide who those reviewers are, what their queue capacity is, and how exceptions escalate. The agent does not remove this role; it changes what the role looks like. A senior AP analyst reviewing 50 drafted invoices a day is a different job than that same analyst keying 50 invoices a day.
7. Treat Wave 1 as the starting line, not the finish. What is generally available today is meaningfully different from what Microsoft will ship in Wave 2. PO matching, approval flows, and anomaly detection are not on the Payables Agent today; some of them will be. Your decision should account for the trajectory, not just the current state.
The pattern under all seven steps is the same. Agentic ERP capabilities in Business Central are real. They are useful. They are constrained in ways that are documented but rarely discussed in vendor marketing. Honest pilots produce honest decisions.
Where Tigunia Fits In
If you are working through a Wave 1 evaluation and you want a partner who treats the limitations as carefully as the capabilities, that is the conversation we are built for. Tigunia’s Business Central practice combines the implementation experience to set up a controlled pilot with the operational accounting experience to translate Copilot Credit consumption into your budget cycle.
For the full feature walkthrough of every agent, including architecture, setup overview, and what is coming in Update 28.1, see The Complete Guide to Business Central AI Agents.
For broader context on where AI fits across the Microsoft stack, our Keys to AI Success With Microsoft Copilot guide covers the strategic frame. For grounding in how Business Central itself is structured, the Complete Guide to Business Central Modules lays out the platform the agents operate on top of.
When you are ready to pilot the Payables Agent or Sales Order Agent against your real invoice volume, or to design a custom agent that solves something the built-in pair does not, let’s talk. The honest version of this conversation starts with your data, not the demo.
Frequently Asked Questions
What is agentic ERP in Business Central?
Microsoft’s 2026 Release Wave 1 framing describes a shift from manual tasks to “autonomous, insight-driven operations” through AI agents that operate inside BC using the same security and permission model as human users (Microsoft Learn, Wave 1 plan). As of May 2026, the generally available agents are the Payables Agent (GA since 2025 Wave 2) and the Sales Order Agent (GA with 2026 Wave 1), plus Agent Designer for custom agents (GA May 2026), with the Expense Agent in public preview.
What does the Business Central Payables Agent do?
It monitors mailboxes for vendor invoices with PDF attachments, uses AI to read invoice content, locates the vendor and accounts in BC, and prepares invoice drafts for human review and approval. Microsoft documents these limits: 100 emails per day, 500 invoices per day, 10-page PDF maximum, 5 MB file size maximum. It does not currently support purchase order matching, approval flows, or anomaly detection (Microsoft Learn, Payables Agent).
How much does the Business Central Payables Agent cost to run?
Built-in BC agents and custom agents built with Agent Designer consume Copilot Credits per task. Credits are available as prepaid or pay-as-you-go. Microsoft does not publish a flat per-invoice price; consumption varies by agent feature and actions performed (Demiliani, March 24, 2026).
What is the Sales Order Agent in Business Central?
The Sales Order Agent reads inbound customer emails, locates the customer in Business Central, holds multi-turn email conversations to clarify missing details, checks item availability (with optional Capable-to-Promise integration for delivery-date calculation), and drafts a sales quote. All outbound messages from the agent are reviewed and approved by designated BC users before being sent (Microsoft Learn, Sales Order Agent).
What is Business Central Agent Designer?
Agent Designer is a Business Central feature that reached general availability in May 2026. It lets consultants, product owners, domain experts, and power users prototype custom AI agents using natural-language instructions, without writing code. Custom agents use the same runtime, security, and permission model as Microsoft’s built-in Payables and Sales Order agents. Update 28.1 (rolling out May–June 2026) adds the production-deployment path for custom agents (Microsoft Learn, Agent Designer release plan; Demiliani, May 10, 2026).
Are Business Central AI agents available on-premises?
No. Microsoft’s Wave 1 overview states that Copilot and agents are available only to Business Central online (SaaS) customers. On-premises and partner-hosted deployments do not have access to these capabilities (Microsoft Learn, Wave 1 plan).
When did Business Central 2026 Wave 1 (Update 28.0) release?
Update 28.0 became available to existing customers with environments scheduled to update beginning May 8, 2026. Administrators can schedule the 28.0 update to any date within the 5-month update period, which ends August 31, 2026 (Microsoft Learn, BC 28.0 what’s new).
Our Fact Checking Process
We prioritize accuracy and integrity in our content. Here's how we maintain high standards:
- Expert Review: All articles are reviewed by subject matter experts.
- Source Validation: Information is backed by credible, up-to-date sources.
- Transparency: We clearly cite references and disclose potential conflicts.
Your trust is important. Learn more about our Fact Checking process and editorial policy.
Our Review Board
Our content is carefully reviewed by experienced professionals to ensure accuracy and relevance.
- Qualified Experts: Each article is assessed by specialists with field-specific knowledge.
- Up-to-date Insights: We incorporate the latest research, trends, and standards.
- Commitment to Quality: Reviewers ensure clarity, correctness, and completeness.
Look for the expert-reviewed label to read content you can trust.