Document Ingestion and write back

Hi, this is an interesting project given the combination of intelligent document processing, financial data validation and direct Xero integration. The brief already provides a strong technical direction, so before I put together a proposal, I’d like to clarify a few implementation and delivery points:

1. Is there an existing application, prototype or codebase that the new platform needs to build upon, or is Phase 1 a greenfield development?
2. How many Xero organisations/companies would the initial MVP need to support, and would users connect their own Xero accounts through OAuth?
3. For matching documents to existing Xero transactions, what information will typically be available to establish the match, and should the system automatically match transactions or require user confirmation?
4. Are there particular invoice formats, supplier types or document qualities that the OCR/extraction system must support beyond standard digital PDFs?
5. Is the 90%+ extraction accuracy target based on specific fields, complete invoices, or an overall accuracy measurement across the document population?
6. Approximately how many documents are expected per day/month initially, and does the 100,000+ document requirement represent the expected first deployment or longer-term scalability?
7. Is the <60-second processing target intended for the complete workflow from upload through Xero creation, or specifically for OCR and data extraction?
8. Do you already have an Azure/Microsoft environment, including Entra ID, Azure SQL, Blob Storage/SharePoint and Power BI, or would setting up the cloud infrastructure be included in the engagement?
9. For approval workflows and exception handling, are the business rules already defined, or would the successful developer be expected to design these workflows with your finance team?
10. What level of Xero functionality is required in Phase 1 beyond supplier creation, bill creation and document attachment, particularly around tax rates, tracking categories, currencies and line-item handling?
11. Will there be access to representative real-world invoices and historical processing data so that extraction accuracy and automated-processing rates can be properly tested before production?
12. Who will be responsible for validating financial accuracy and approving the accounting workflows before the system is allowed to create live Xero records?
13. Is security/compliance assessment, penetration testing and formal production readiness included in the scope, or will these be handled by an internal IT/security team?
14. Are Phases 1–3 already funded and planned, or is Phase 1 being used to validate the platform before decisions are made about subsequent development?
15. What is the target timeframe and budget/milestone structure for Phase 1, and is the project approved and ready to commence?

How will documents enter the system initially — manual upload, email inbox, watched folder, API, or a combination?
Do you already have a preferred OCR/document-AI service, or should the implementation recommend the best option based on accuracy and cost?
For invoices that cannot be confidently matched to a supplier or transaction, should they go into a human-review queue before anything is written to Xero?
Do you have a Xero demo/sandbox organisation available for integration testing?
Should the original PDF only be stored in the document repository, or must it also be attached to the corresponding Xero transaction whenever possible?

Hi Scott,

What's the target go-live timeline for Phase 1 (MVP)? Roughly how many invoices/month do you expect to process initially, so I can plan for scale? Do you already have an Azure subscription and Microsoft Entra ID (M365) in place, or would these need to be set up as part of this engagement? Also, is Xero already configured with your suppliers, and is PO matching required from Phase 1, or can it be deferred to Phase 3 as outlined in the brief?"

Thank you
Vishal K