AI Bookkeeping: How to Use AI in Your Accounting

AI bookkeeping means letting AI do part of the bookkeeping for you: reading receipts and invoices, suggesting account codes, reconciling payouts or answering questions about your numbers. It can save hours every month. But the business is still responsible for its books, and AI is only as good as the data it can see.

This guide covers the three ways to use AI in bookkeeping, what AI can and can't do today, and how to connect ChatGPT or Claude to Fortnox, the accounting system most Swedish businesses use.

Can AI do my bookkeeping? AI can do much of the work, but it can't take the responsibility. There are three ways to use AI in bookkeeping: AI built into your accounting software, an AI assistant like ChatGPT or Claude connected to your books, and automatic bookkeeping through integrations. For e-commerce, the third saves the most time. The other two help with what's left.

What is AI bookkeeping?

AI bookkeeping is bookkeeping where AI takes over part of the manual work. That can mean reading a receipt and suggesting the account and VAT, matching an incoming payment to the right customer invoice, or answering "which invoices are overdue?" straight from your books.

The difference from ordinary automation is that AI handles what doesn't follow a fixed rule: a receipt in a new format, a question in plain language, a discrepancy that needs explaining. Ordinary automation is better at what repeats exactly the same way every day. The best setups use both.

How does AI work in accounting?

In practice, in three steps:

1. The AI reads the document or the data: a receipt, an invoice, a payout or the books themselves.
2. The AI suggests or answers: an account, a VAT code, a match, or an answer to your question.
3. You review and approve before anything is booked.

The third step is the most important one, and it's what separates good AI bookkeeping from risky AI bookkeeping.

Can AI do the bookkeeping?

Yes, AI can do a large part of the work, but not all of it, and not the responsibility.

What AI does well today:

• Reading receipts and supplier invoices
• Suggesting account codes and VAT codes
• Matching incoming payments to customer invoices
• Reconciling payouts from payment providers against sales
• Summarizing and explaining numbers, for example why a month stands out
• Answering questions about customers, suppliers and invoices

What AI shouldn't do on its own:

• Book anything without someone reviewing it
• Make judgment calls, such as accruals, doubtful receivables or borderline VAT cases
• Change periods that are already closed
• Replace your accountant or auditor at year-end

Under Swedish bookkeeping law, as in most countries, the business is responsible for its books, whatever tool does the work.

Three ways to use AI in bookkeeping

The three ways solve different problems and work best together.

1. AI in your accounting software

Many accounting systems now include AI that reads receipts and supplier invoices and suggests how to book them, for example Spiris. It's the easiest way to start: it's already where you do your bookkeeping, and you approve each suggestion. The limit is that the AI only sees what's inside that software.

2. An AI assistant connected to your books

Here you use an AI app you may already have, like ChatGPT or Claude, and connect it to your accounting system. The connection uses MCP (Model Context Protocol), an open standard for how AI apps read data and take actions in other systems. Then you ask in plain language, and the answer comes straight from your books, not from a spreadsheet you pasted in.

This suits questions and analysis: what's overdue, how did the month go, does the payout add up. With the right permissions, the assistant can also carry out tasks, like registering a payment, after you approve it.

3. Automatic bookkeeping through integration

An integration connects your systems, for example your webshop, payment provider and Fortnox, and books every transaction by the same rules, every day. It isn't AI in itself, but it removes the most manual work for anyone with volume. And it's the foundation that gives an AI assistant correct data to answer from.

Automatic bookkeeping for e-commerce

For a webshop, the biggest time sink is rarely individual receipts. It's volume: hundreds of orders a week, payouts from Klarna, Svea or Walley with fees and returns already deducted, and VAT in several countries. No chatbot solves that. An integration that books everything by the same rules does.

What can be booked automatically:

• Sales, per order or as a daily summary, with the right account and VAT
• Payouts from payment providers, split into sales, fees and returns
• Returns and credit notes, linked to the original order
• OSS VAT per EU country, if you sell to consumers in other countries
• Customers and articles, so the same customer or article isn't created twice

Where AI and automation meet

The integration does the recurring work. AI helps when something doesn't add up: why doesn't the payout match the sales? Why did order 4512 never reach Fortnox? With an AI assistant connected to both your books and your integrations, you get the answer in a minute, with the evidence, instead of digging through logs.

Junipeer connects webshops, payment providers and warehouses to Fortnox and other ERPs, and lets you ask ChatGPT or Claude about both your books and your integrations. Read more about Fortnox integrations and Workspace MCP.

How to connect ChatGPT or Claude to Fortnox

You connect an AI app to Fortnox with an MCP server, a service that sits between the AI app and Fortnox. With Junipeer it takes about five minutes and no code.

1. Add Fortnox MCP in Junipeer. Connect your Fortnox account in the Junipeer app. It costs €39/mo plus €2 per 1,000 requests, and you don't need any other integration.
2. Create a sign-in key and choose what the AI may do. Start with read-only access. You can give it more later.
3. Add Junipeer in your AI app. Paste the address https://api.junipeer.io/mcp as a custom connector in Claude or ChatGPT.
4. Sign in with Junipeer and approve. Then ask the way you'd ask a colleague.

Which plan do you need? In Claude, custom connectors work on every plan, including Free (one connector). In ChatGPT, full MCP support needs Business, Enterprise or Edu.

Using Claude? Here are all the steps with screenshots: Claude MCP. Read more about what the AI can do in Fortnox: Fortnox MCP.

What can you ask AI about your books?

Once the AI is connected to your books, you ask in plain language. The answer is calculated from your real numbers, not guessed.

Customer invoices and payments

• "Which customer invoices are more than 30 days overdue, and what's the total?"
• "Has Nordic Outdoor paid invoice 1042?"
• "Which customers pay late most often?"

Suppliers and costs

• "Which supplier invoices are due next week?"
• "Which costs have grown the most this year compared with last year?"

Follow-up

• "How did October compare with September? Explain the biggest differences."
• "What were sales per country last quarter?"

E-commerce and reconciliation

• "Does Klarna's payout on 9 October match the sales in Fortnox?"
• "Why did order 4512 never reach Fortnox?"

Which AI is best for bookkeeping?

It depends less on which AI model you choose and more on what it can access. An AI guessing from a pasted spreadsheet is worse than a simpler AI that reads straight from your books. Look at these six things:

1. Real data. Does the AI read straight from your accounting system, or only what you paste in?
2. Permissions. Can you let it read only, and give it more when you're ready?
3. Approval. Does it ask before it changes anything?
4. Traceability. Can you see afterwards what the AI changed and when?
5. Where your data goes. Is the service run in the EU, and what does your agreement with the AI provider say about your data?
6. Local knowledge. Does it understand your language, your chart of accounts and your VAT rules?

ChatGPT or Claude? Both work well for questions about your books. The difference is mostly in the plans: Claude allows custom connectors on every plan, while ChatGPT needs Business, Enterprise or Edu for full MCP support. Choose the one you already use.

AI and the law: responsibility, risks and tips

AI changes how bookkeeping is done, not who is responsible for it. Here's what to keep in mind.

The responsibility stays with the business

The business is responsible for correct bookkeeping, whether it's done by hand, by an accounting firm or with AI. If the AI books something wrong, it's the business's mistake. That's why someone should review what the AI suggests.

Records and archiving rules still apply

Every transaction needs its supporting document, and accounting records must be kept for the period the law requires. In Sweden, that's until the end of the seventh year after the calendar year in which the financial year ended. A summary made by AI doesn't replace the original documents.

Risks to watch for

• Guessing. An AI without access to your data can sound confident and still be wrong. Connect it to real data instead of pasting in numbers.
• Unintended changes. Give the AI read-only access until you know how it works.
• Closed periods. Changes to a month that's already been reported can cause problems with VAT.
• Personal data. Your books contain customer data. Check where the service runs and what your agreement with the AI provider says.

Five tips to get started

1. Start by asking, not booking. Let the AI answer questions for a while before it changes anything.
2. Give read-only access first. Raise permissions step by step.
3. Let the integration handle the volume. AI is best at what stands out, not what repeats.
4. Review the suggestions. Especially VAT, accruals and anything touching closed periods.
5. Talk to your accountant. Many accounting firms already use AI and can help you set up the right way of working.