Building a VAT AI Tool; Need Feedback From Accountants
I’ve been building a small AI tool for UK MTD workflows and I think I may have misunderstood the actual problem.
Initially I thought the pain point was:
- extracting invoice data
- VAT classification
- exporting into Xero/Sage
So I built a prototype that:
- extracts supplier/date/totals/VAT from invoices
- classifies VAT automatically
- matches invoices against bank CSV exports
- exports MTD-ready CSVs
But after showing it to a few people, I’m starting to think OCR/extraction is becoming commodity infrastructure.
The real pain might actually be:
- chasing clients for missing documents
- review/approval workflows
- handling exceptions
- knowing what’s ready for filing
- audit trails
- communication loops
Basically operational workflow rather than data extraction itself.
Would really love honest feedback from accountants/bookkeepers here because I don’t want to build founder fantasy software.
A few questions:
- What’s actually the worst part of MTD compliance today?
- Are tools like Dext/Xero “good enough” already?
- Would you trust AI VAT classification if confidence scoring + human review existed?
- Is local-first/privacy-first meaningful at all for accounting firms?
- If you could remove ONE bottleneck from your workflow, what would it be?
Also curious whether smaller firms are underserved here, or whether this market is basically saturated already.
For context:
- solo founder
- built the prototype in a day
- rebuilt from Streamlit → FastAPI after hitting frontend limitations
- currently free/open source while validating
Not trying to pitch anything honestly; mostly trying to understand whether this is solving a real problem or just a technically interesting one.