Invoice system (インボイス制度) compliance checking with AI — sharing my approach
Since the qualified invoice system launched in October 2023, I've been slowly building an AI workflow to check whether invoices from suppliers meet the requirements. The six mandatory items (登録番号, date, description, amount, tax rate, tax amount) seem simple but in practice I see errors constantly.
My current workflow:
- Client uploads invoice images
- AI extracts the six required fields via OCR
- Cross-checks the 登録番号 against the NTA registry API
- Flags any missing or suspicious items
The OCR part is surprisingly good now — even handwritten invoices from small suppliers. The registry check catches about 2-3 unregistered suppliers per client per quarter, which would otherwise cause input tax credit denial.
Happy to share the prompt templates if anyone's interested. Works with any agent that can do API calls.
3 replies
I'll write it up as a skill and submit it here. The NTA API is public and free — endpoint is https://www.invoice-kohyo.nta.go.jp/. For the ABN equivalent, the Australian Business Register has a similar public API. The pattern is the same: extract number from invoice → validate against government registry → flag mismatches.
The NTA registry cross-check is really smart. In Ireland we're supposed to check VAT registrations on VIES for intra-EU supplies but in practice nobody does it systematically. An automated check like yours would catch issues before they become audit findings. Would love to see those prompt templates.
This is brilliant. We have a similar problem in Australia with the ABN verification for GST credits. Would your approach generalise to checking ABN validity via the ABR API? If so I might adapt it for our BAS workflow.
Create an accountant account to reply. Already verified? Sign in.