Business Central Sales Order Agent vs Lleverage

Oct 6, 2026·11 min read
Forklift carrying a pallet of stacked sacks across a yard of wrapped pallets at sunset

The Business Central Sales Order Agent is Microsoft’s built-in AI agent that reads customer emails, creates a sales quote and turns it into a sales order once confirmed. We think it suits short orders from known contacts, with a person approving every reply. Lleverage fits orders in files, past 15 lines, or from senders not on file.

What is the Business Central Sales Order Agent?

The Sales Order Agent is an AI agent built into Dynamics 365 Business Central. It reads customer emails in one Microsoft 365 mailbox, finds the customer and the requested items, checks availability and answers with a sales quote or a sales order as a PDF. Microsoft lists it under Generally available on the Copilot & agent capabilities page and charges for it in Copilot Credits.

It works through the same pages a person uses. From a Role Center it opens pages, fills in fields and runs actions, guided by instructions in natural language and by the captions and tooltips on each page. It runs as its own user, SALES ORDER AGENT followed by the company name, with the SOA - EDIT permission set, so every record it creates carries its user ID. One company can run several instances, each with its own mailbox or folder and its own review settings.

Setting it up: mailbox, permissions, billing and languages

A partner or admin needs 4 things in place before switching it on. Whoever creates the agent needs the Configure All Agents permission (ID 9665), which the AGENT - ADMIN permission set grants, and delegated administrators cannot activate it. The mailbox is a Microsoft 365 user or shared mailbox, and whoever sets up monitoring needs Full Access and Send As on it; Exchange can take a few hours to pass those rights on. Billing for agents is set per environment in the Business Central admin center. And every customer contact needs the right address in its E-Mail field, because that address is how the agent knows who wrote.

Copilot Credits are bought in advance as credit packs or paid as you go through an Azure subscription linked to a Power Platform environment. Microsoft’s billing page prices each step: 2 credits to analyse an incoming email, 5 for an attachment with sales data, 5 to check availability, 5 to create or update a quote or order, and 2 to write the reply. Its own example, 100 requests a month with a PDF on half of them, comes to 1,650 credits.

On languages, Microsoft’s pages differ. The set-up page lists Danish, English, French, German, Italian, Portuguese (Brazil) and Spanish. The language table lists Dutch (nl-NL and nl-BE), Finnish, Norwegian and Swedish too, and leaves out Portuguese. Mixed-language emails can lower the quality, and in a multilingual country such as Belgium item search depends on the text search language you set. The agent is offered in every Business Central country, though where Azure OpenAI is not available an admin has to allow data to move across geographies.

What the agent does with an order, and where it stops

Each email becomes a task with a timeline. A relevance model drops messages that are no request, such as offers to sell or attempts to reach internal data, and a second check skips attachments like terms and conditions. Items are searched in 3 layers: exact identifiers (item number, vendor item number, GTIN, item reference), then attributes such as colour or size, then a fuzzy search across the catalogue, item translations and extended text included. With capable-to-promise switched on, it quotes the earliest shipment date for items that are out of stock.

It always creates a sales quote first, because a quote does not touch planning, reservations or availability. Two switches decide what follows. With Send quotes for confirmation on, the customer sees the quote and confirms by email. With it off and Make orders from quotes on, the quote becomes a sales order straight away and the customer only sees the order. In both cases a Business Central user approves every outgoing email.

Microsoft’s FAQ sets out the limits. The agent handles up to 15 item lines per document and only lines of type Item, so charges, resources and fixed assets stay with a person. It creates no new items, contacts or customers, and passes price and discount requests to a user. It can change a quote when the customer asks, but leaves a sales order alone once it exists. Attachments must be PDF, PNG or JPG, a PDF at most 50 pages and 20 MB, with about 150,000 characters across all attachments in one email. Its instructions sit in Microsoft’s code: partners can read the prompts on GitHub, and adjusting or extending them is not supported.

One order email, followed through both

Take an invented case. Lisa, a buyer at Noordkamp Lagers B.V., emails a PDF purchase order for 6 bearing types with a delivery date and her PO number. Her address is on a contact card linked to the customer.

With the Sales Order Agent, on default settings:

  1. The email lands in the monitored folder and the agent opens a task. It is the first message in the conversation, so a user confirms the incoming request.
  2. The agent matches Lisa’s address to her contact, finds the 6 items, checks availability for the delivery date and creates a sales quote.
  3. A user reviews the reply with the quote PDF, confirms it, and the agent sends it.
  4. Lisa replies that the quote is fine. A user confirms her message, the agent converts the quote into a sales order, and a user approves the order confirmation before it goes out.

That is 4 confirmations by a person, as in Microsoft’s general flow. On Microsoft’s price list the quote round costs about 19 Copilot Credits and the confirmation round about 9, our sum. Change one detail and a person takes over more of the work: with 22 lines Lisa is past the 15 that Microsoft supports, an Excel purchase order is skipped as an unsupported file, and if a colleague forwards her email a user has to pick her contact by hand with Use another contact once.

With Lleverage:

  1. The agent picks the email up from the shared Outlook mailbox and reads the PDF.
  2. It looks Noordkamp up in Business Central from what the order says, finds each item, and checks availability and price before anything is written.
  3. It creates the sales order in the right Business Central company and adds the 6 lines through the API. Above your confidence line the confirmation goes back without anyone approving it.
  4. Below the line, for example on an article code it does not know, the colleague who owns the customer gets the question in Teams with the reason. The agent takes the answer and finishes the order.

The 22-line Excel file and the forwarded email go through the same steps, because the agent looks the customer up from what the order says, whoever sent it.

Business Central Sales Order Agent vs Lleverage, side by side

We read Microsoft Learn’s pages on the Sales Order Agent (overview, set-up, processing, FAQ, billing and the language table) and the Lleverage docs on 12 November 2026. The first row is where Microsoft’s agent is ahead. How the connection itself works is on our order management page for Business Central.

Sales Order Agent (Microsoft)Lleverage
Getting startedAhead here. Built into Business Central and generally available: turn it on, pick a mailbox and folder, add users. Try it out tests it on your own items with no mailbox connected.Forward deployed engineers set up the Business Central connection with your team and build the first flow with your order desk. The first process typically runs supervised within a few weeks.
Where orders come inOne Microsoft 365 mailbox folder per instance, read through Exchange. Microsoft says other ways to receive email are not supported.yes: Outlook and Gmail mailboxes, shared ones included, and the same work can start from an API call or a schedule.
Order filesEmail text plus PDF, PNG and JPG. A PDF up to 50 pages and 20 MB, about 150,000 characters across all attachments.yes: email text, PDF, Word, Excel and CSV files, and portal exports, as Xpol’s customers send them.
Knowing the customerOnly by matching the sender’s address to the E-Mail field on a contact card. A forwarded email needs a user to pick the contact by hand.yes: looks the customer up in Business Central from what the order says, whoever sent it, as at Topa with about 700 customers.
Document it createsAlways a sales quote first, turned into a sales order after the customer confirms or straight away, depending on 2 switches. Changes quotes, not orders.yes: creates the sales order, or a sales quote when that is the flow, in the Business Central company you choose, through the API.
Long and mixed ordersUp to 15 lines per document, all of type Item. Charges, resources and fixed assets stay with a person.yes: reads the whole order and adds each line to the sales order. At Xpol a large order saves about 20 minutes.
Who signs offA user approves every outgoing email. By default a user also confirms the first message from known senders and every message from unknown ones.yes: your confidence line decides. Above it the order goes through on its own; below it a named colleague decides in Teams, Slack or the inbox and the agent carries on.
Your own rulesInstructions sit in Microsoft’s code. Partners can read the prompts but cannot adjust or extend them; custom fields work with results that vary.yes: Skills built from your examples and corrections, each change approved by a person. Xpol runs 25 customer rulesets.
Where it runsBusiness Central online. On-premises Business Central and Navision are not described in Microsoft’s pages for the agent.yes: Business Central online and Navision on your own servers, reached without opening them to the internet.
How it is billedCopilot Credits per step: 2 to analyse an email, 5 per attachment with sales data, 5 to check availability, 5 per quote or order, 2 per reply.One monthly price per agent, with integration and AI usage included. No seats and no usage bills.

When Microsoft’s agent is enough

We think the built-in agent is the right start when most orders are short emails or PDFs from contacts whose address is on file, under 15 item lines, in a language Microsoft validated, and your desk is happy to approve each reply inside Business Central. It needs no project, the task timeline shows the agent’s reasoning and citations, and Business Central approval workflows can add a check, for example before a quote is released. Its Try it out section runs a task on a sample message, or on a sample PDF purchase order built from your own items, before any mailbox is connected.

It also works as a split. Microsoft describes a dedicated folder fed by Outlook rules for approved senders, so one instance takes trusted customers and converts quotes into orders straight away, while everything else stays with your team.

When Lleverage fits better

We think an AI agent like Lleverage fits when your order inbox looks like this:

  • Orders arrive as Excel, CSV or portal exports next to PDFs and plain email.
  • Orders run past 15 lines, or mix items with charges.
  • Buyers, assistants and shared addresses order for the customer, so the sender is not the contact on file.
  • Customers have habits your desk applies by hand, such as unit conversions, label text or depot splits.
  • Confirmations should go out without a person approving each email, with only the exceptions reaching your team.
  • Part of the group still runs Navision on its own servers.

Topa took this route for about 700 customers who order by plain text email, PDF, CSV and Excel: over 90% of incoming orders now go into Business Central with no manual input, and customers get their confirmation within 30 seconds. At Xpol the agent applies 25 customer rulesets and prepares each order in Business Central for review. If you would rather build your own agent on Microsoft’s stack, our Copilot Studio comparison covers that route.

What changes at 1,000 orders a month

On Microsoft’s per-step prices, 1,000 email orders a month with a PDF on half of them come to about 16,500 Copilot Credits for the first round alone, and more when customers confirm by email (our calculation, scaled from Microsoft’s example). The default daily limit is 100 incoming emails, and anything over it waits for the next day. When credits run out the agent stops reading new email, and once they return it works through the backlog, which Microsoft warns can use a large part of the new credits.

The bigger cost is time on your desk. Outgoing emails always need a person, so even with the quote step hidden from customers that is at least 1,000 approvals a month, plus every order the agent cannot finish. The number to track is the share of orders that need nobody, which our article on the touchless rate explains how to measure.

Frequently asked questions

How many Copilot Credits does one order use?

On Microsoft’s billing table, an email with a PDF purchase order costs about 19 credits to reach a quote: analysis, attachment, availability, quote and reply. A customer confirming by email adds about 9 for the sales order and its confirmation. Microsoft’s own example puts 100 requests a month at 1,650 credits. Lleverage charges one monthly price per agent instead.

Does the Sales Order Agent work in Dutch?

Microsoft’s language table lists Dutch for the Netherlands and Belgium, while the set-up page gives a shorter list without it, so test on your own emails first. Mixed-language emails can lower quality, and Belgian companies should set the text search language for item search. Lleverage handles the orders of Topa’s customers across the Netherlands.

What does a partner set up before activating the agent?

The Configure All Agents permission for whoever creates it, a Microsoft 365 shared mailbox with Full Access and Send As, billing for the environment in the Business Central admin center, and clean E-Mail fields on customer contacts. Exchange can take hours to apply mailbox rights, and a delegated administrator cannot activate the agent, so plan the go-live day around both.

Can a partner change how the Sales Order Agent behaves?

Only around the edges. Partners can copy the SOA - EDIT permission set, adjust the agent’s profile, add instances, route mail with Outlook rules and read the prompts on GitHub. Changing the instructions or extending the agent is not supported. In Lleverage your team changes a Skill through examples and corrections, and a person approves each change.

What happens when the Copilot Credits run out?

The agent stays active but stops reading new emails, while replies it already prepared still go out. When credits return it works through everything that piled up, which Microsoft warns can use a large share of the new credits. Microsoft suggests a daily email limit, 100 by default, and switching the agent off when it is not needed.

Can both run on the same Business Central?

Yes, as long as each reads its own mailbox or folder, because the Sales Order Agent processes every email in the folder it watches. An Outlook rule can send trusted customers’ short orders to the agent’s folder and the rest to Lleverage. Both then create documents in the same company, each under its own user.

Give your back office an AI workforce.