Chasing paper receipts, manually keying in supplier invoices and moving files between systems is work that neither creates value nor is fun. The good news in 2026 is that the technology finally lives up to its promise: AI interpretation is sufficiently accurate, the e-invoice network Peppol has become the de facto standard, and the integration platforms make it possible to connect the systems without a development department. The bad news is that most automation projects still get stuck – not on the technology, but on the process. Here's how it's actually done, step by step, and where it usually goes wrong.
Start with invoice processing – that's where the time is
For most companies, supplier invoices are the single biggest administrative time-stealer. It is also the area where automation has come the furthest. Modern financial systems such as Fortnox and Visma eEkonomi have built-in invoice interpretation: when an invoice is received, the supplier, date, OCR number, amount and VAT are read automatically, and the system suggests accounting accounts. What used to be manual input becomes a review and a button press.
The interpretation is no longer based only on classic OCR, but on machine learning that has been trained on millions of invoices and that gets better the more it sees your suppliers. In practice, this means that after a while recurring suppliers are invoiced almost completely automatically, while new or unusual invoices are highlighted for manual checking.
A real turning point came1 July 2024, when the Accounting Act was changed so that you no longer have to save the paper original of receipts and invoices after you digitize them. This means that the chain can be completely digital from receipt to completed verification – no folders, no double storage. The accounting information must still be archived in an orderly and reliable manner for at least seven years, but the format may be digital.
E-invoice and Peppol: build for the future
Since April 2019, e-invoicing is mandatory for everyone who supplies to the Swedish public sector, and the Peppol BIS Billing 3.0 format has become the norm. More and more large private companies today also require e-invoices from their B2B suppliers.
The big shift is ahead of us. Through the EU packageVAT in the Digital Age (ViDA), formally adopted on 11 March 2025, introduces digital reporting requirements (DRR) for cross-border B2B transactions from1 July 2030, and no later than January 1, 2035, all national systems must be harmonized against the EU standard EN 16931. The point for Swedish companies is simple: build your invoice flows for structured e-invoicing already now, and you will avoid a stressful migration later. A real e-invoice (a structured data document) can also be automated significantly longer than a scanned PDF.
Accounting documents: from inbox to verification
The next natural step is to automate the path from documentation to registered verification. Here, Swedish companies today combine AI interpretation with matching against bank transactions. Tools read receipts and invoices, calculate the correct VAT, match against the account statement and create the verification directly in the financial system.
A robust flow usually looks like this:
Collection:documents are captured automatically – via a dedicated email address to which receipts are forwarded, a mobile app that takes a photo of the receipt, or direct integration with the supplier's system.
Interpretation:AI extracts fields and classifies expenditure type and VAT rate.
Matching:the basis is paired with the correct bank transaction.
Posting:the system suggests accounts based on history.
Certificate and review:a human approves deviations and records above a certain threshold.
The defining design principle is that humans should review the exception, not the rule. Set clear rules for what is allowed to pass automatically (eg recurring suppliers under a certain amount) and what always requires a pair of eyes. Then the processing time drops dramatically without losing control.
Email and document flows: the next front line
Once the finance flow is in place, email and documents are the biggest remaining time wasters. This is where language models come in really handy, as they can understand unstructured text. Common, value-creating applications in 2026:
Triage of incoming emails:AI classifies and labels incoming mail – quote request, support case, invoice, spam – and routes to the right function or creates a case in the case management system.
Draft answer:the model suggests answers to common questions that a human reviews and submits.
Document Extraction:agreements, order confirmations and packing slips are read and key data is entered into CRM or business systems.
Document Routing:files are named, categorized and archived in the correct location automatically.
The glue between the systems is integration platforms – iPaaS tools like Make and Zapier, or custom integrations against APIs. They can listen for events (a new invoice, a new email) and trigger chains of actions across multiple systems. For many Swedish companies, this is where the real leverage lies: not in a single smart tool, but in the systems talking to each other without manual intermediate steps.
The pitfalls – and how to avoid them
1. To automate a broken process
The most common mistake is to put automation on top of a process that is actually bad. Then you scale up the chaos. Map and simplify the flowbeforeyou automate. Often it turns out that half of the steps can be removed completely.
2. Blindly trusting the AI interpretation
AI interpretation is accurate, but not error-free – especially on unusual formats, handwritten receipts or foreign invoices. Build in controls: automatic approval thresholds, random sampling and clear escalation in low confidence. Human review is also an explicit requirement in future regulations.
3. GDPR and the AI Regulation
As soon as the administration touches personal data – and invoices, emails and contracts almost always do – the GDPR applies in full, even when an AI is doing the work. The EU's AI Regulation (AI Act) is applied in parallel with the GDPR; the majority of the rules become applicable from August 2026 and Swedish supplementary legislation is planned to enter into force on 2 August 2026. The requirements for high-risk AI only come into effect on 2 December 2027. The sanctions are significant: up to 35 million euros or 7% of global turnover for prohibited AI uses, and up to 15 million euros or 3% for violations of the high-risk requirements. Control where your suppliers process data, have personal data processor agreements in place and document which decisions AI makes.
4. Getting stuck in supplier lock-in
Choose tools with open APIs and export capabilities. You want to be able to change a component without tearing the entire chain. Structured formats such as Peppol and EN 16931 are your friends precisely because they are supplier independent.
5. Forgetting change management
Automation is changing people's work. Without training and clear ownership, the new flows are not used, or bypassed. Appoint someone responsible, measure the results and iterate.
A realistic roadmap
Start narrow and measure. A proven order: 1) automate supplier invoices in the financial system, 2) digitize the receipt and expenditure flow, 3) connect financial systems with banking and CRM via an integration platform, 4) add AI triage of e-mail and documents. For each step: define what should be fully automatic, what requires review, and how errors are caught. Then you build an administrative machinery that is fast, controlled and ready for the regulatory requirements to come.
Want to get started?
At ZORC, we build automation flows and integrations that connect financial systems, e-mail and document management - with control, traceability and GDPR at the backbone. What a project lands on depends on the number of systems, the volumes and how tailored the integrations need to be. Get an estimate adapted to your flows this springquote calculator, or get in touch viacontactthen we look at your processes together.
