n8n for back-office automation works well when the steps are fixed and every system in the chain has an API: moving records between cloud apps, alerts, scheduled reports and simple approvals. It starts to break when the work begins as a PDF or an email, ends in an ERP without a built-in node, and needs a person who knows the customer.
At Lleverage we think n8n is a good tool for a company with an engineer who wants to own its automations. The order desk or accounts payable team of a manufacturer or wholesaler has a different problem. Most of their effort sits outside the canvas: reading the document, writing into the ERP, handling the exceptions and keeping it all running a year later. That is the work our AI agents for order management and accounts payable take on, inside the systems you already run. To see it on your own documents, book a demo.
This post is for operations and finance leads who have an n8n build, or are about to start one. It sets out which parts of the back office n8n carries and which it leaves to you. For the wider market, see our list of 15 n8n alternatives. For the canvases side by side, see our n8n, Make and Zapier comparison.
What is n8n, and how does it handle back-office work?
n8n is a workflow automation tool. You build a workflow on a canvas out of nodes: a trigger that starts it, then steps that fetch, change and send data. n8n calls its integrations nodes and offers hundreds of them, from Gmail and Slack to Xero and QuickBooks Online. When a node lacks an operation, or a service has no node, the HTTP Request node calls the service's API directly. Community-built nodes can be installed on top.
You can run n8n in 2 ways. n8n Cloud is hosted by n8n on paid plans. The Community edition is free and runs on your own servers, and installing and configuring it takes technical skill. For back-office work that difference matters. The person who installs n8n also patches it, backs it up and gets the call when a workflow stops at 07:00 on a Monday.
n8n has added AI as well. Its AI Agent node can call tools. A human review step can pause the workflow before a tool runs, so a person approves or denies it.
Where does n8n for back-office automation work well?
n8n does well on work with a clear shape, between systems that already have an API:
- Moving data between cloud apps, where the mapping is fixed: a won deal in HubSpot creates a customer in Xero, or a new row in a Google Sheet becomes a task in Asana.
- Alerts and routing, such as a Slack message when a payment run fails. An error workflow can also alert a team when another workflow fails.
- Scheduled reports, for example a nightly export from a database, cleaned up and sent as a spreadsheet.
- Approvals with a yes or no answer. With the Slack node's send and wait operation, an approver approves or declines a step inside Slack before the workflow goes on.
- Prototypes. An engineer can show in an afternoon that a process can be automated at all.
If an engineer at your company likes this work, and the systems are cloud apps with good APIs, n8n is a reasonable choice. It is often a cheap one too.
Where does n8n break in back-office work?
The trouble starts with the work most back offices in manufacturing, wholesale and logistics spend their day on. That means orders, supplier invoices, purchase order confirmations and master data. This work has 4 parts that n8n leaves to you.
The document at the start
A customer order arrives as a PDF, an Excel file or a few lines in an email. Reading text is also the most common way European companies use AI: according to Eurostat's 2025 survey, 19.95% of EU enterprises with 10 or more employees used AI, and analysis of written language was the most used kind, at 11.75% of enterprises.
n8n's Extract From File node turns a PDF or a spreadsheet into data the workflow can read. It does not know which part of that text is the delivery date, the customer's article code or the shipping address. You add an AI node for that and write its instructions. Then you keep them up to date as each customer's layout changes. With 25 customers who each order their own way, someone maintains 25 sets of rules.
The ERP at the end
The work is only done when it is in the ERP. n8n has built-in nodes for Odoo, QuickBooks Online and Xero, and a Microsoft Dynamics CRM node that covers accounts. We found no built-in node for SAP, Business Central, Exact Online or AFAS in n8n's node list. For those, an engineer builds the calls with the HTTP Request node or installs a community node. That engineer then owns the result: sign-in, error handling, field mapping per company and every ERP update. An on-premise ERP adds a further question: how a cloud workflow reaches a server that should stay closed to the internet.
The exceptions in the middle
In order entry, the hard cases decide whether automation pays off. Think of an unknown article code, an order that could belong to 2 sites of the same customer, or a price that does not match the agreement. n8n's human review shows a person which tool the AI wants to use, with which values, and the person approves or denies it. An order desk usually needs more. The person corrects a line, picks the right customer or adds a missing reference, and the work goes on with that answer. In n8n you design that screen, the queue and who sees what.
The person who keeps it running
The free Community edition leaves out the sharing of workflows and credentials: only the instance owner and the person who built a workflow can open it. It also leaves out environments, version control with Git, SSO and log streaming. Those come with paid licences. A process that books money or stock should not depend on a build only 1 person can open.
What does n8n cost for back-office volumes?
n8n prices its plans in executions. One execution is one run of a whole workflow, however many steps it has. On 2 November 2026, Starter cost €20 a month billed yearly for 2,500 executions and Pro €50 for 10,000. Business, which is self-hosted, cost €667 a month for 40,000, with SSO, environments and Git version control. Enterprise is quoted.
For an order desk, 150 orders a week comes to about 650 runs a month, well inside Starter's 2,500. Most of the cost sits in engineering time: building the document reading, the ERP calls and the exception screens, then maintaining them. A self-hosted build adds the server, the monitoring and the updates.
How does an AI agent built for the back office handle the same work?
Lleverage starts from the job, such as order intake or invoice matching. An AI agent runs that job inside the systems you already use. Compared with a canvas you build yourself, the ERP connection, the exceptions, the rules and the upkeep all work differently.
Business Central is a managed connection that our team sets up with yours. The agent can look up customers and items, create sales orders and sales quotes, and create purchase invoices. The permissions of the connected Business Central account decide what it can see and change. Lleverage also works in SAP, Dynamics 365 Finance and Operations, Exact, AFAS, Ridder iQ and the other systems on our ERP integrations page. A system with no standard connector gets a custom integration, with Windows (NTLM) sign-in for older on-premise systems. Where there is no API at all, the agent works the screen with the access a person has.
When a workflow needs a decision or missing information, it pauses and sends a request to a named person or role: confirm which customer an order belongs to, add a missing reference number, or choose between 2 products. The run continues once they answer. The agent can also hand part of a job to a colleague, such as an approval for a discount. It scores how sure it is of each piece of work. Below the line a person decides, above it the work goes through on its own, and only you move the line.
Customer-specific rules are kept as Skills. When people rate the agent's work, it drafts an improvement to the Skill, and a person approves the change before it goes live. You can also show the agent a process on a screen-share call, and it writes down each step, screen and exception.
A failed run explains in plain language what happened and who can act next. Project admins and builders are notified of failures in production. Our forward deployed engineers build the first process with your team, and it typically runs supervised within a few weeks. Roles, single sign-on and an audit trail come with it, as our control page sets out.
Microtechniek, a 130-person engineering firm in Velsen-Noord, shows where a general workflow tool stops. Workflow tools and Power BI had already replaced much of its manual work. Anything that had to read from or write to its on-premise Ridder iQ ERP stayed manual. In August 2026, Microtechniek wrote between 45 and 71 customer purchase orders a week into Ridder iQ without anyone typing them. Each order used to take Microtechniek about 10 minutes by hand, and its admin desk now has more than 500 hours a year back. Order entry runs through evenings and weekends, so a Friday-afternoon order is in the ERP before anyone is back at a desk.
Xpol, a fresh flower wholesaler on Business Central, has the agent prepare each order as a draft that a person signs off, with unclear fields left blank. Xpol saves about 20 minutes per large order, across 150 orders a week.
"It’s a matter of building trust in the organisation with these kinds of initiatives. You can’t just throw something like this over the fence." Cees Maaskant, General Manager, Xpol
How does n8n compare with an AI agent for the back office?
| Question | n8n | Lleverage |
|---|---|---|
| Who builds it | Your engineer, on a canvas | Our engineers with your team |
| Who runs it day to day | Whoever built it | Your operations team, with our engineers behind them |
| Reading PDF and email orders | Extract From File plus AI nodes you configure | The agent reads the document and checks it against the ERP |
| SAP, Business Central, Exact, AFAS | HTTP Request or community nodes | Connections set up with our team |
| On-premise ERP | You arrange access | Reached without opening it to the internet |
| Exceptions | Approve or deny steps, other screens you build | Requests to a named person, and the work goes on |
| Sharing and version control | Paid licence | Roles, single sign-on and audit history |
| Pricing | Free Community edition, or paid plans counted in executions | One monthly price per agent, set after we have seen the work |
When should you keep n8n, and when should you look at n8n alternatives?
Keep n8n for what it does well. If an engineer maintains it and the job is app-to-app plumbing, alerts or reports, there is little reason to move it.
Look at n8n alternatives when 3 or more of these are true for a process:
- The work starts as a document or an email in a format you do not control.
- It has to end up in an ERP with no built-in n8n node, or one that runs on premise.
- More than a handful of cases a day need a person to correct something, beyond a yes or no.
- Only 1 person understands the build.
- Mistakes cost money: a wrong price on an order, an invoice booked against the wrong supplier.
The 2 can run side by side: you can keep n8n for internal plumbing and hand the order desk or invoice processing to an agent. Our post on connecting an AI agent to your ERP explains what that connection takes. Build vs buy for back-office AI agents weighs the 2 routes.
Frequently asked questions
Can n8n connect to SAP or Business Central?
n8n has no built-in node for SAP, Business Central, Exact Online or AFAS in its node list. It reaches them through its HTTP Request node or a community node, using the ERP's API. That works, but an engineer has to build the calls, map the fields for each company and keep them working after every ERP update.
Is n8n free for business use?
The self-hosted Community edition is free and covers most of the product. It leaves out sharing of workflows and credentials, environments, Git version control and SSO, which need a paid licence. n8n Cloud runs on paid plans counted in executions. Self-hosting also brings server, update and monitoring costs that the licence price does not show.
Can n8n read PDF orders and invoices?
In part. The Extract From File node pulls the text out of a PDF or a spreadsheet. Working out which value is the article code, the delivery date or the supplier's VAT number takes an AI node with instructions you write and maintain. You also build your own checks against the ERP before anything is created.
Does Lleverage replace n8n?
It takes over the back-office jobs where the work starts in a document and ends in the ERP, such as order intake and invoice matching. You can keep n8n next to it for app-to-app automations between cloud tools. Lleverage also has a REST API, so other systems, n8n included, can start its workflows and follow their runs.
Where to start with your own back office
Pick the back-office process that costs your team the most hours, usually order entry or supplier invoices. Count how many documents a week it handles and how many of them need a person today. Then hold that process against the 5 points above. If 3 or more apply, an n8n build will spend most of its life on exceptions and ERP calls.
We map that process with you, set up the agent in your ERP and run it supervised on your real documents before anything goes through on its own. Book a demo and bring a week of orders or invoices. To see how the agent creates sales orders step by step, look at our order management page.
