Every EPC finance team knows the Friday afternoon pile. Vendor invoice PDFs in Outlook. A spreadsheet someone stopped updating on Wednesday. A PM asking whether Package 14 invoices are reflected in the financial summary before Monday's review.
Someone has to open each attachment, read line items, and type values into a tracker. Scanned invoices break simple copy-paste. Field supers email photos of paperwork. Financial rollups in Plathera are only as current as the last person who entered data by hand.
The invoice is not missing. It is sitting in an inbox waiting to become a row leadership will treat as current on Monday.
Email is where vendors send invoices. It is rarely where invoices become structured register entries. Manual re-keying introduces lag, typos, and no defensible link between the PDF and the row in the financial summary.
Scanned attachments defeat basic text extraction. Routing the wrong document type to the wrong workflow creates cleanup work. When leadership asks which invoices hit a package this week, the answer often requires inbox archaeology instead of a query against the live program model.
Generic chat tools can summarize an attachment. They do not reliably produce a traceable invoice row with ingestion metadata tied to the source email and PDF.
Plathera accepts inbound email through Resend webhooks. Each project can expose dedicated addresses for different document types. The router classifies messages by mailbox local-part or subject keywords before any extraction runs.
invoice@ routes to the invoice pipeline. timesheets@, po@, and funding@ route to their respective pipelines using the same inbound system. Misrouted messages are a real failure mode; the routing rules exist to keep document types out of the wrong register.
A project controller can BCC the invoice address on vendor correspondence so capture happens without opening a parallel tracker.
Once a message is classified as an invoice, Plathera runs a fixed attachment pipeline. Each stage is logged. The outcome is a structured row in the invoices table, not a chat summary.
| Stage | What happens |
|---|---|
| Receive | Resend inbound webhook delivers the message to Plathera |
| Route | Classify by mailbox or subject into invoice, timesheet, PO, or funding pipeline |
| Attach | Save PDF to temp storage; process the first PDF when multiple are attached |
| Extract | Merge embedded PDF text, Tesseract OCR, and vision fallback for hard scans |
| Structure | Map fields to the invoice schema via a domain-specific processor prompt |
| Insert | Write a row to the invoices table with ingestion metadata |
| Confirm | Send optional confirmation email; deduplicate repeat webhook delivery |
Not every invoice arrives as clean digital text. Field photos, fax scans, and stamped copies break naive extraction. Plathera layers embedded text parsing, OCR, and vision transcription so supers can keep emailing paperwork without forcing finance back to manual entry for every scan.
That does not mean every format parses perfectly. Unusual layouts, poor scans, and ambiguous PO references may still need human review. The ingestion path produces a traceable row with extraction stats so finance can judge confidence before treating the register as final.
Structured fields include invoice number, PO reference, dollar value, and dates mapped into the invoices table. Ingestion metadata preserves which email and PDF produced the row: message ID, source address, and page-level extraction detail.
Those rows feed the same financial summary views the program already exposes: funding, purchase orders, invoices, and timesheets drillable by site, building, project, and package. The EPC Assistant can answer questions against that register instead of asking someone to rebuild state from Outlook.
After ingestion is configured, a PM or controller might ask the EPC Assistant:
Show me recent invoices that were created from inbound email ingestion.Or before turning ingestion on:
What happens when someone emails a scanned PDF invoice to invoice@?The answer should describe routing, extraction layers, the row written to the register, and where that row appears in the financial summary.
The invoice is not missing. It is sitting in someone's inbox waiting to be typed into a register that leadership will treat as current on Monday. Inbound ingestion should close that gap with a traceable row, not another manual step.
Worth a conversation if invoice capture on your program still runs through manual re-keying.
Forward or BCC vendor invoice PDFs to your project's inbound invoice address. Plathera receives the message via webhook, routes it to the invoice pipeline, and logs a row in the invoices register.
When multiple PDFs are attached, the pipeline processes the first PDF. Additional attachments may require separate messages or manual handling.
Yes. Ingested invoice rows feed the same financial summary views in the live program model. The EPC Assistant can query recent invoices created from inbound email ingestion.