Perspective · · 5 min read
Swedish accounting firms are leaving more on the table than they realise
If you run a Swedish firm and use Fortnox, Reda, Mobius or Rillion, you have the best set of tools the market has offered so far. It's also a good moment to ask what the next step looks like, because something new is coming.
This article isn't written to say you should be doing anything other than what you are doing today. It's written because you are probably doing more work than you need to, and the tools available so far have not offered you an alternative. That is starting to change.
Where the Swedish market is today
Swedish firms have an advantage few other markets share. Electronic invoicing is widespread thanks to Peppol and Svefaktura. Extracting basic fields is rarely a problem. Total amounts, suppliers, and invoice numbers arrive structured. The common tools build on top of that with suggestions for GL account and VAT code, and let the accountant review and approve.
This has worked well. Clicking has become faster year by year. Error rates have dropped. An accountant handles more clients than she did ten years ago. This is real progress, and it is worth stating clearly, because it often gets forgotten.
But it is also worth looking at what this progress has not changed. The accountant still opens every invoice. She reads the suggestions, fills in what is missing, and approves. Her time is still directly tied to the number of invoices the firm handles. If volume doubles, she either works twice as much or the firm hires someone else.
This is the model that every tool on the Swedish market supports today. It is an assisted model. A human is in the loop for every invoice, every time. The tools make her work faster, but they do not remove it.
What exists beyond the assisted model
In Finland and Norway, a different model has gone into production over the past few years. There, a significant share of invoices is posted without anyone opening them. The accountant only sees the invoices where the system is uncertain. For a firm processing thousands of invoices a month, this means a meaningful part of the daily workload disappears.
Tripletex, the leading Norwegian accounting platform, introduced autonomous invoice processing two years ago. It has not only made their firms more efficient. It has created a new revenue stream where the firm charges per posted invoice rather than per hour. The margin on the work moves from payroll to software. It has also changed what firms can offer clients, because the time freed from invoice handling can be used for advisory work.
This is the category of autonomous invoice processing. It is not a better version of the tools that exist in Sweden today. It is a different category. The difference is not how fast the accountant clicks. It is whether she clicks at all.
| AI copilot / assistant | Autonomous layer | |
|---|---|---|
| Who opens the invoice? | The accountant — every time, faster | No one, above a confidence threshold |
| Output | Better suggestions, faster review | Booked entry with audit trail, zero human touch |
| Automation stage | Human in the loop | Human out of the loop |
| Pricing model | Per seat or per plan | Per posted invoice — per actual outcome |
Read the right column as a layer above the left, not a replacement for it. If the firm's current tools produce good suggestions, that is training signal and workflow context for the autonomous layer. The autonomous layer adds one thing on top: a calibrated confidence decision about when the accountant is needed at all.
Half-automation makes the job faster. Full automation makes the job disappear for the easy 50–70% and leaves the hard 30–50% for the accountant, where her judgement actually matters. Both layers can run on the same firm at the same time.
Why Sweden has not seen it until now
A fair question is why no Swedish vendor has delivered this already. The answer is not that Swedish companies have missed something. It is that building a system that posts invoices autonomously in production is considerably harder than it sounds.
It is easy to get an AI model to suggest the right GL account on ninety percent of a sample of clean invoices. What is hard is handling ten-page invoices with internal coding systems, suppliers the system has never seen, partial credit notes against older invoices, and VAT rules that must be exactly right. This does not require a model. It requires a dozen different techniques working together, each calibrated against one customer's data, each continuously retuned as new models are released or new rules come into effect.
This kind of infrastructure is not built in six months. It needs millions of invoices from many customers, several years of production data to learn from, and a team whose only job is finding better combinations of techniques. This is why autonomous invoice processing has taken time to reach Sweden. It's not that no one has tried. It's that no one has had the platform to do it at scale.
What this means for your firm
If you are satisfied with how your firm runs today, that is a reasonable position. Most firms are, and the current tools do a good job within their category.
The question is more forward-looking. What would it mean if half your purchase invoices were posted without anyone opening them. How would it affect staffing. How would it affect the pricing model, if time is no longer the most important resource. Would it free up capacity for advisory work, which generates higher margin per hour. Could you take on more clients without hiring.
These are not questions that need answering today. But they are worth thinking about, because the market is moving. When autonomous invoice processing starts being offered in Sweden, the firms that move first will gain a competitive advantage before others have time to react. Not because it is a large technical change, but because it is a pricing-model change that is hard to match once a competitor has already made it.
There is no rush. But it's worth understanding what is possible, before your own assumptions about what automation means become outdated.
Frequently asked
What is autonomous invoice processing?
Autonomous invoice processing is end-to-end invoice handling where a meaningful share of purchase invoices is posted into accounting without a human opening them. The accountant only sees invoices the system is uncertain about. This is a different category from the assisted models common in Sweden today, where every invoice is reviewed and approved.
How does it differ from Fortnox, Reda, Mobius or Rillion?
Those tools have made the accountant faster. They suggest GL accounts and VAT codes on electronic invoices arriving via Peppol or Svefaktura. But the accountant still opens every invoice, reads the suggestion, and approves. Autonomous invoice processing removes the invoice from her workflow entirely when the system is confident enough. Her time stops being tied to the number of invoices the firm handles.
Why haven't Swedish firms seen this until now?
Building a system that posts invoices autonomously in production is considerably harder than it sounds. It needs a dozen techniques working together, calibrated against each customer's data, continuously retuned as models and VAT rules change. That requires millions of invoices from many customers, several years of live data, and a team whose only job is the stack. Sweden hasn't seen it yet because nobody has had that platform at scale — not because nobody has tried.
How is this different from an AI assistant that suggests the right GL account?
A suggestion tool makes the click faster. Autonomous processing removes the click. The change isn't the speed of the workflow — it's whether the workflow happens at all. It also changes the pricing model: firms can charge per posted invoice rather than per hour, because their time is no longer the binding constraint on capacity.
Does this replace the tools we already use?
No. Assisted tools like suggestion engines keep making the review faster. Autonomous processing is a layer above — it removes the review entirely for invoices the system is confident about, and sends only the exceptions to the accountant. Both can run on the same firm at the same time. The question isn't which tool to choose. It's whether to add the layer that removes work rather than speeds it up.
Commercial Director, FabricAI — part of Visma Solutions Oy.
See it in practice
Want to see outcome-based automation live?
Book a Partner Fit Workshop and we'll walk you through how we price the outcome — and what it takes to deliver it.