Email-to-Invoice Automation
The endgame of invoice automation: vendors email their bills to a dedicated address, and drafts appear in REDA Financials — vendor matched, lines coded, high-confidence ones already created — while your team reviews instead of types. Here's the pipeline and the setup.
Access you'll need. Working OCR items requires the Accounts Payable v2 permission set (AP clerks & managers) or Accounting Admin v2 (system administrators). The Accounting Controller v2, Accounts Receivable v2, and Read Only v2 sets can view items but not act on them.
The pipeline, end to end
An email arrives with PDF bills attached
Each PDF attachment becomes its own OCR Inbox item (Pending), with the sender, subject and file preserved. Non-PDF-bearing emails get a courteous bounce: "No PDF attachments were found in the email. Only PDF files are processed for invoice extraction." — and successful intake acknowledges: "2 PDF document(s) received and queued for processing."
The sender is resolved — even through forwards
A contact email match links the vendor immediately. Forwarded mail is handled with unusual care: the AI reads the email body to find the original sender, so your AP mailbox forwarding bills inward doesn't confuse vendor matching.
The 15-minute loop processes the queue
The OCR Document Processing loop (Operations tab) picks up Pending items, runs the two-phase extraction, and applies the confidence rule: at or above the threshold → draft PI auto-created; below → Awaiting Review for a human. Failures land in Errors with the reason.
Setting up the address (administrator, once)
The intake address is a Salesforce Email Service your admin creates on the package's OCR email handler — Setup generates a long service address, which most teams front with a friendly alias (ap-invoices@yourcompany.com forwarding to it). Two decisions worth making deliberately: whether vendors get the address directly or everything routes through your AP mailbox (forwarding works fine, per the original-sender handling), and who monitors the queue views daily.
The control loop for going hands-free
- Start conservative: a high auto-create threshold routes almost everything through review; lower it as spot-checks stay clean.
- Review auto-created drafts at posting — automation moved the typing, not the accountability.
- Watch the meter — every intake counts against the monthly document allowance; the Subscription tab shows consumption, and over-cap intake pauses with a clear message.
What each intake record preserves
Every PDF that arrives becomes a fully documented intake item — the queue is an audit trail, not just a to-do list. For each attachment, the OCR Inbox record captures:
| What's kept | Detail |
|---|---|
| The document itself | The PDF is stored as a file on the intake record, so reviewers always see exactly what the vendor sent |
| Sender Email | Who sent it — replaced by the original sender when the AI detects a forwarded bill |
| Sender Account & Vendor Match | When the sender's email matches a contact in your records, that contact's account is linked as both the sender and the suggested vendor — before extraction even starts |
| Email Subject & Body | Preserved with the record (long texts are truncated to fit — subjects to 255 characters, bodies to 32,768) |
| Source File | The attachment's filename, so multi-attachment emails stay distinguishable |
| Receiving Email | Which intake address received it — useful when you run more than one |
| Starting state | Status Pending, marked as an automated intake, document type Purchase Invoice |
What the sender hears back, verbatim
The channel answers every email it receives — vendors and forwarding colleagues are never left guessing:
| Situation | The reply |
|---|---|
| PDFs received and queued | "{n} PDF document(s) received and queued for processing." |
| No PDF attachments in the email | "No PDF attachments were found in the email. Only PDF files are processed for invoice extraction." |
| Something failed during intake | "An error occurred while processing your email attachments: {reason}" — and an Error-status intake record is still created with the reason captured, so nothing disappears silently |
A week on Meridian's intake address
Meridian Properties Group fronts its intake with ap-invoices@meridianproperties.com. A typical week:
| Intake result | After the 15-minute loop | |
|---|---|---|
| MetroPower bills the US entity — one PDF, sent directly by the vendor | One Pending item; MetroPower matched from the sender address immediately | High confidence → draft Purchase Invoice auto-created, coded from MetroPower's history |
| The UK office manager forwards a cleaning invoice from her own mailbox | One Pending item; the AI reads the forward and records the cleaning company as sender | Vendor matched despite the forward; above threshold → draft created |
| A vendor statement PDF and a promotional flyer PDF in one email | Two Pending items — one per PDF, filenames preserved | Neither is an invoice: both land in Errors with the reason, no invoices created |
| A scanned bill photographed as a JPG | Nothing queues; the sender gets the "No PDF attachments were found" reply | The AP team uploads the image on the New Purchase Invoice form instead |
Common mistakes to avoid
| Mistake | Why it hurts | Do this instead |
|---|---|---|
| Publishing the intake address on your website | Junk PDFs consume metered documents and clutter the Errors view | Share it with vendors deliberately, like bank details — or route everything through your AP mailbox |
| Vendor contacts missing email addresses in your records | Sender matching has nothing to match — every bill needs manual vendor identification at review | Keep a contact with the billing email on each vendor account; matching (and history-based coding) starts working immediately |
| Re-forwarding an email because nothing happened in five minutes | Each arrival creates new intake records and consumes the meter again — the first copy was just waiting for the loop | Check the queue first: Pending means "received, waiting for the next 15-minute cycle" |
| Nobody owns the queue | Below-threshold bills and errors age silently while due dates pass | Assign the Awaiting Review and Errors views to a named person's daily routine |
| Treating the bounce reply as spam | The replies are the channel's diagnostics — "No PDF attachments" tells you exactly why nothing queued | Read the reply; it names the problem and the fix |
Where to go next
- OCR Inbox & Document Queue — where the emails land.
- Creating an Invoice from OCR — the review half.
- Operations Scheduler — the loop's controls.
Frequently asked questions
Five OCR Inbox items, five extractions, up to five draft invoices — one per PDF attachment. Multi-page single-PDF bills process as one document.
Email is the unattended channel — PDF-only keeps the automated pipe to the format vendors actually send and screens out signature images and logos that ride along in emails. For a photographed receipt, use the form's upload, where a human is present.
Non-invoice PDFs are refused by the is-it-an-invoice check and sit in Errors/Rejected with the reason; no invoice is ever created from them. The address is still worth treating like any finance mailbox — shared deliberately, not printed on the website.
Yes — the meter counts intake, one per PDF, whether or not an invoice ultimately results. That's another reason to keep the address curated and vendors trained to send bills only. When the cap is reached, intake stops with "Monthly AI document processing limit has been reached. Please contact your administrator."
The address that sent the email. If your AP mailbox forwards bills inward, your own team sees the acknowledgements and bounce messages — vendors sending directly see them instead. Either way, someone always hears back.
Not today — the pipeline creates Purchase Invoices from vendor bills. Other document types are reserved for future phases; sending them now ends in the not-an-invoice refusal rather than a wrong document in your books.
Was this article helpful?