Introduction
Most mid-market leaders have now sat through the Copilot demo. Someone from finance asks the obvious question afterwards and rarely gets a straight answer: what changes on Monday morning?
That gap is where most Dynamics 365 AI conversations stall. The material describes a transformation; the demo shows a polished workflow on clean sample data. Neither tells a CFO at a $40 million distributor whether their three-day close becomes a two-day close.
This guide covers what AI in ERP looks like inside Dynamics 365 Business Central, as of 2026 release wave 1, version 28 (Microsoft release plan. Most published guidance on this subject is written for Dynamics 365 Finance & Operations, Microsoft’s enterprise platform. This is written for the mid-market.
What is the difference between Copilot and an AI agent?
Dynamics 365 Copilot is assistive. You prompt it, it responds, and a person reviews the output before anything happens. Agents are autonomous within boundaries you set: they monitor data, respond to events, and complete work unprompted. Microsoft describes the current wave as expanding agents into complex, end-to-end processes rather than single-step help.
Copilot Studio agents are the third category. When a process is specific to your business and nothing built in covers it, you build one on the Power Platform against your Business Central data. This is where most industry-specific value lives.
Which AI agents are available in Business Central today?
Three matter for mid-market firms. The Sales Order Agent reads inbound customer orders from email and drafts them as sales orders. The Payables Agent automates accounts payable end to end, reading invoices, matching vendors and accounts, and preparing them for approval with human oversight. The Expense Agent handles expense capture and categorisation. Availability moves between monthly updates, so confirm current status before planning around it.
Business Central AI vs Finance & Operations AI
Finance & Operations is Microsoft’s enterprise ERP for multi-entity global organisations. Business Central is the mid-market platform, and capabilities often reach one before the other. If you are evaluating AI ERP software for a company between $10 million and $250 million in revenue, guidance written for Finance & Operations will overstate both capability and cost.
Copilot vs AI agents at a glance
| COPILOT | AI AGENTS | |
|---|---|---|
| What it does | Assists a user with a task | Executes a task independently |
| Who starts it | The user, via a prompt | An event, schedule, or data change |
| Oversight | Human reviews every output | Human sets bounds, reviews exceptions |
| Licensing | Included in Essentials and Premium | Consumption-based Copilot Credits |
| Typical use | Summarising, drafting, analysing | Order entry, payables, expenses |
DYNAMICS 365 AI IN FINANCE
The Payables Agent, from invoice to approval
The most consequential finance capability in the current wave. Microsoft describes the agent as automating accounts payable end to end: interpreting the invoice, matching it to the right vendor and ledger accounts, and preparing it for approval with human oversight retained.
The phrase to hold on to is human oversight. The agent prepares; a person approves. The work removed is the keying and the coding, not the control.
Bank reconciliation and payment matching
Reconciliation assist matches bank statement lines to open ledger entries, including the awkward ones rule-based matching misses: partial payments, combined remittances, unhelpful references. It flags what it cannot resolve rather than guessing.
Cash flow forecasting and collections
Business Central projects cash flow from open receivables, payables and posted history. AI improves the forecast by weighting actual customer payment behaviour rather than assuming every invoice is paid on terms. On collections, Copilot drafts follow-up correspondence with the invoice context already populated. More on the CFO view in Is Copilot for Business Central worth it for CFOs?
DYNAMICS 365 AI IN SUPPLY CHAIN
How does the Sales Order Agent work?
A customer emails an order, often as a PDF attachment or loose text. The agent reads it, identifies the customer, matches line items to your catalogue, drafts the order and routes it for review. For distributors still keying emailed orders by hand, this removes a whole category of work.
Demand forecasting and reorder point suggestions
Forecasting projects demand from posted sales history and proposes reorder quantities and timings. It earns its keep on slow-moving and seasonal SKUs, where intuition does not scale across thousands of items. Monitoring also surfaces late purchase orders and stock imbalances before a customer complains. See also improving demand forecasting accuracy with AI and ERP.
Where does AI forecasting break down?
On short or dirty history. A forecast built on twelve months that include a migration or a one-off bulk order will be confidently wrong. Eighteen to twenty-four months of clean posted history is the realistic threshold, and a company that migrated from QuickBooks last year does not have it.
DYNAMICS 365 AI IN OPERATIONS
Asking your ERP questions in plain English
An operations manager can ask which customers ordered a given item last quarter without knowing where that report lives. The effect is wider than convenience: it removes the bottleneck where three people are the only ones who can get anything useful out of the system.
Automating expenses and the rest of the admin
Expense handling is the classic low-value, high-friction process in a mid-market firm, and the Expense Agent captures and categorises it. Beyond that, not every process fits a standard ERP workflow. Power Automate handles the deterministic ones. Copilot Studio handles those needing judgement, as custom agents grounded in your Business Central data and governed by your approval rules. Because Copilot also reaches into Teams, Outlook and Excel, people who avoid the ERP still get its data, which matters more than any single feature where adoption is the real constraint.
WHAT THIS DELIVERS, BY ROLE
The business case looks different depending on who is asking.
| ROLE | WHAT DYNAMICS 365 AI CHANGES |
|---|---|
| CFO | Faster close, automated payables coding and matching, better cash flow visibility, approval control retained |
| COO | Capacity to absorb volume growth without proportional hiring; fewer transcription errors reaching customers |
| IT / CIO | AI deployed inside an existing Microsoft tenant, under existing identity, permissions and data-residency controls |
| Operations manager | Self-service answers from ERP data; less waiting on reports from someone else's queue |
WHAT DYNAMICS 365 AI COSTS
What is included in your Business Central license
Essentials lists at $80 per user per month and Premium at $110, both billed annually, following Microsoft’s November 2025 increase. Copilot assistance is included in both tiers at no additional license cost, so most organisations can pilot it without raising a purchase order.
How do Copilot Credits work?
Agent activity is billed separately. Credits cost $0.01 each pay-as-you-go, billed in arrears against an Azure subscription, with prepaid capacity available at a discount. Each agent action consumes a few credits, so the bill tracks how often an agent runs and how much work each run involves.
To make that concrete: a Sales Order Agent handling around 100 order emails a month lands near 1,650 credits, roughly $16.50. An illustration rather than a quote, but it corrects the assumption that agent automation carries enterprise pricing. Cost scales with volume and agent design, not agent count.
WHAT DYNAMICS 365 AI WILL NOT DO
It will not fix bad master data
This is the one that sinks projects. Duplicate vendors, inconsistent item codes, customers entered three different ways. AI does not detect that your data is unreliable; it produces confident output built on it. Data cleanup is not preparation for the AI project. It is the AI project.
It will not replace judgement or accountability
Agents act within boundaries a person defines, and Microsoft’s own description of the Payables Agent keeps approval with a human. Your controller does not disappear; the routine matching in front of them does.
It will not run safely without governance
Permissions come first: which roles approve what, which data an agent may read, what happens outside its rules. Data residency is part of this. Since 1 July 2026, environments on version 28 or later in some countries may process Copilot and agent requests in a different Azure geography, manageable on the Copilot and agent capabilities page. For regulated firms that is a pre-go-live conversation.
It will not rescue a heavily customised legacy system
If you are running Dynamics NAV or GP with years of custom code, none of this is practically available. The AI lives in the cloud platform. Migration comes first, and using AI as a reason to delay that decision usually costs more than making it.
HOW MID-MARKET TEAMS SHOULD GET STARTED
- Audit your master data. Duplicate vendors and inconsistent item codes are the failure points, and they are cheaper to fix before agents rely on them.
- Pick the process where your team spends most time on repetitive work: usually supplier invoices, bank reconciliation, or emailed order entry.
- Record the baseline first: hours per month, error rate, or days to close.
- Set permissions and approval boundaries, and confirm your data-residency position if regulated.
- Pilot with a small group, then measure the same metric again. Expand only from proof.
Volt Technologies helps mid-market teams identify which processes are worth automating and deploy Copilot and agents with governance in place from the start.
Ready to find out what AI would actually change in your business?
Frequently Asked QuestionsÂ
Dynamics 365 AI is the set of AI capabilities built into Microsoft's business applications: Copilot assistance, autonomous agents such as the Sales Order and Payables Agents, and custom agents built on the Power Platform. In Business Central it supports finance, supply chain and operations directly inside the ERP.
Copilot assists a user who prompts it and reviews the output. An agent executes work independently when triggered by an event or a data change, within boundaries the business defines. Copilot supports people; agents complete tasks.
Only if that data is clean and consistent. Duplicate records and conflicting units of measure produce confidently wrong output. Most organisations need a master data cleanup first, and forecasting needs eighteen to twenty-four months of clean posted history.
Finance & Operations is Microsoft's enterprise ERP for large multi-entity organisations; Business Central serves the mid-market. Capabilities arrive at different times, and guidance written for Finance & Operations overstates both availability and cost for a smaller company.
A single well-scoped process such as payables or emailed order entry can show a measurable difference within weeks of go-live, provided the data is clean. Organisations needing significant data remediation should plan a longer runway.
