InvoiceFlow - 100+ invoices were being processed by hand every month. InvoiceFlow automated the entire process, from the email arriving to the invoice being filed and entered into the ledger.

100+Invoices / month

0Routine manual touches

7Automated stages

100%Traceable flow

The problem

The invoices weren’t complicated. Processing them one by one was.

A construction firm was handling more than 100 invoices every month through the same repetitive manual loop. Someone had to open the email, download each attachment, decide whether it was actually an invoice, read the document, copy the relevant fields into a spreadsheet, normalize the vendor information, check for duplicates, rename the file, move it into the correct Drive folder, and mark the email so everyone knew it had been handled. None of those steps was particularly difficult. Together, they created a process that consumed attention every single month and depended on people getting every small handoff right. InvoiceFlow replaced that loop with one automated pipeline. The goal wasn’t to make invoice processing faster. It was to make routine invoice processing disappear.

How it runs · 07 steps

Email ArrivesInvoiceFlow watches the designated Gmail workflow and processes each attachment independently.

trigger

Document Is ClassifiedGemini checks whether the attachment contains the required invoice information. Statements, receipts, quotes, and unsupported documents are routed out before extraction.

Fields Are ExtractedStructured invoice data is extracted from valid documents, with separate handling for construction-specific AIA payment applications.

Vendor Is ResolvedRaw vendor names are normalized and mapped to the accounting code used by the business.

process

Duplicates Are CheckedA composite invoice key is checked against existing records before the new row is written.

process

File Is OrganizedThe attachment is renamed consistently and moved into the correct Google Drive location.

storage

Record & Status Are SyncedA clean row is written to Google Sheets and the source email is labeled based on the outcome: processed, needs review, or no invoice.

output

Engineering detail · 04

AI where judgment is neededClassification and extraction use Gemini because the documents vary in structure and format. The rest of the workflow stays deterministic wherever possible — lookups, duplicate checks, file naming, routing, status updates, and storage all follow explicit business rules. The system uses AI for ambiguity, not for everything.

Different documents, different rulesInvoiceFlow detects construction-specific AIA G703/G732 payment applications and switches extraction logic automatically rather than forcing every document through the same schema. Standard invoices follow the normal extraction path while specialized documents get their own field mapping.

Dirty inputs become clean recordsVendor names arriving on invoices rarely match accounting records perfectly. InvoiceFlow normalizes the raw vendor string, resolves it against the business’s vendor-code mapping, and automatically creates review stubs for vendors it cannot confidently match. Duplicate detection runs before the write using a composite vendor, project, and invoice-number key.

Exceptions stay visibleAutomation should not make unusual cases disappear. Documents that fail classification, unknown vendors, possible duplicates, and other exceptions remain visible to the AP team through explicit routing and Gmail labels. Every attachment has a destination and every processed email shows its final workflow state.

Result

InvoiceFlow now handles the routine path for more than 100 invoices every month without the AP team manually moving information between Gmail, documents, spreadsheets, and Drive. Incoming files are classified, structured data is extracted, vendors are normalized, duplicates are flagged, documents are filed, accounting rows are created, and the original email is labeled automatically. The team still owns the decisions that need human judgment. The repetitive movement of information around those decisions no longer needs them. Email arrives. The system does the rest.

”

“He quickly understood our business requirements and translated our objectives into a clean, scalable solution. What stood out most was his ability to think beyond just "making it work."”

Peyton Vitter · Upwork

NextPromptheus →Teams were relying on AI without knowing which prompts actually worked. Promptheus gave them one place to test, improve, share, and reuse what gets better results.

Bring me a problem like this →