Reprocessing Failed Extractions
Some documents fail — a scan too blurry, a vendor that didn't exist yet, a transient service hiccup. Failures land in the OCR Inbox's Errors view with the reason recorded, and the Reprocess action gives every document a second chance after you've fixed the cause. Nothing is lost; nothing needs re-emailing.
Access you'll need. Reprocessing 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.
Triage by error message
| The message says… | It means… | Fix, then reprocess |
|---|---|---|
| "…does not appear to be an invoice." | The AI is protecting you from a statement, a flyer, a contract | If it truly isn't a bill, mark Rejected. If it is (unusual layout), reprocess — or create the invoice manually with the file attached |
| "AI service is temporarily rate-limited / overloaded…" | Transient service pressure | Just reprocess in a few minutes — nothing on your side is wrong |
| "Monthly AI document processing limit has been reached." | The org's document meter is spent | New month or bigger plan; the queue waits |
| "AI API key is not configured…" | Licence not activated (or lapsed) | Admin Console → Subscription → activate; see your administrator |
| A parsing/extraction failure | The document defeated the reader (rotated scan, image-of-an-image) | Get a cleaner copy if possible; otherwise manual entry with the file attached |
The full message reference
The table above is the triage view; this one is the complete catalogue, word for word as the messages appear on the record (or, for the intake replies, in the sender's mailbox). Placeholders like {0} are filled in with the specific detail at the moment of failure:
| The message | What it means / what to do |
|---|---|
| "AI service is temporarily rate-limited. Please try again in a few moments." | Too much at once — typically a burst of documents. Nothing to fix; reprocess after a short wait. |
| "AI service is temporarily overloaded. Please try again shortly." | The service itself is busy. Same remedy: wait briefly, reprocess. |
| "AI service returned an empty response. Please try again." | A one-off blank answer. Reprocess; escalate to support only if it repeats on the same document. |
| "AI service returned an error (HTTP {0}): {1}" | An unexpected service error — the placeholders carry the code and detail. Reprocess once; if it persists, share the exact message with your administrator. |
| "Failed to parse AI extraction response: {0}" | The AI's answer couldn't be read back — usually a document that defeated it (rotated scan, photo of a photo). Reprocess; if it repeats, get a cleaner copy. |
| "The uploaded document does not appear to be an invoice. Please upload a valid invoice document." | The first-pass check protecting you from statements, flyers, and contracts. If it truly isn't a bill, mark the record Rejected. |
| "The uploaded file could not be found. Please try uploading again." | The document file isn't on the record — typically an interrupted upload. Upload the file again on the New Purchase Invoice form. |
| "OCR Inbox record not found." | The record was deleted while someone was acting on it. Refresh; if the document still matters, re-submit it. |
| "Monthly AI document processing limit has been reached. Please contact your administrator." | The org-wide document meter for the month is spent. New month or a bigger plan; the queue waits. |
| "AI API key is not configured. Please activate your REDA Financials AI license." | The licence isn't active (or has lapsed). Your administrator activates it in the Admin Console's Subscription tab, then you reprocess. |
| "No PDF attachments were found in the email. Only PDF files are processed for invoice extraction." | Sent back to the sender by email — the email channel accepts PDFs only. Have the vendor resend as PDF, or upload the file yourself on the New Purchase Invoice form (which also takes JPG and PNG). |
| "An error occurred while processing your email attachments: {0}" | The sender's copy of a failure. An Error-status record was still created with the same reason captured — find it in the Errors view and work it from there. |
| "Unsupported file type. Please upload a PDF, JPG, JPEG, or PNG file." | Manual upload only accepts these four formats. Convert or re-export the document, then upload again. |
Why errors never retry themselves
The automatic processing loop — the "OCR Document Processing" card on the Admin Console's Operations tab, cycling every 15 minutes — only ever picks up records in Pending and Extracting status. A record marked Error is deliberately out of its reach: retrying a document whose vendor doesn't exist would fail identically every 15 minutes forever. Errors wait for a person, and Reprocess is how the person answers.
That design also protects the queue: when a document fails, only that one record is marked Error (with the reason in its Error Message field) and the loop continues with the next document. One unreadable scan never blocks the fifty good bills behind it. Your admin can watch the damage from a distance too — the Operations card shows Document Queue counters for Pending, Extracting, and Errors, so a rising Errors count is visible without opening the OCR Inbox tab.
Running a reprocess
Fix the cause first
The classic: vendor didn't exist → create the vendor account. Entity mismatch → check the bill was addressed to an entity you've configured. Transient errors need no fix at all.
Click Reprocess on the record
The extraction resets and re-runs both phases against the stored document ("Reprocessing document..."), then opens the pre-filled invoice form exactly like a fresh Create Invoice. Reprocessing is also the right move on successful-but-wrong extractions — after you fix the master data that misled it.
Queue hygiene
- Errors view daily — most entries are one-click retries; the rest are master-data to-dos that improve every future extraction.
- Watch for patterns — the same vendor failing repeatedly means their invoice layout needs a look (or a territory guidance record, via your admin) rather than per-document retries.
- Reprocess consumes no new document credit for the metered month when re-running a stored document — the read was already counted at intake.
A Friday sweep of Meridian's Errors view
Meridian Properties Group's UK entity clears its REDA: Errors view every Friday. This week's haul shows how different the fixes can be for the same-looking queue:
| Record | Error Message says | What actually happened | The fix |
|---|---|---|---|
| OCR #00000198 | "AI service is temporarily rate-limited. Please try again in a few moments." | A facilities vendor emailed twelve invoices in one burst on Tuesday; two hit the limit | No fix needed — Reprocess each; both extract cleanly on the retry |
| OCR #00000201 | "Failed to parse AI extraction response: …" | A photographed, sideways receipt from a one-off contractor | Ask the contractor for a proper PDF; reprocess when it can't be had and enter manually if the retry fails too |
| OCR #00000203 | "The uploaded document does not appear to be an invoice. Please upload a valid invoice document." | A tenant's lease summary forwarded to the intake address by mistake | Nothing to reprocess — mark Rejected, tell the forwarder where lease documents actually go |
| OCR #00000205 | "AI API key is not configured. Please activate your REDA Financials AI license." | The licence had lapsed overnight before the renewal key was applied | The admin activates the new key in the Admin Console's Subscription tab; then reprocess |
Total time: about ten minutes — and note that only one of the four (the sideways receipt) was a problem with a document. The Errors view is mostly a retry list, occasionally a to-do list, and only rarely a bad-news list.
Common mistakes to avoid
| Mistake | Why it hurts | Do this instead |
|---|---|---|
| Reprocessing instantly without reading the Error Message | Master-data failures (missing vendor, unmatched entity) fail identically on the retry — you just burn time | Read the message first; transient service errors are the only ones that need no fix |
| Re-emailing the document instead of reprocessing | Each email intake creates a new record and counts a new document against the monthly allowance — reprocessing the existing record costs nothing extra | Reprocess — the file is already stored on the record |
| Deleting Error records to tidy the view | You lose the error message, the document, and the proof of what arrived — and the month's usage count stops matching the records behind it | Mark dead ends Rejected; they cost nothing to keep |
| Waiting for the loop to retry errors | It never will — the loop only processes Pending and Extracting records | The Errors view is human territory; sweep it on a schedule |
| Reprocessing to "fix" an invoice that already exists | Blocked by design: "Cannot reprocess — a Purchase Invoice has already been created from this record." | Edit the draft Purchase Invoice directly — it's an ordinary draft |
Where to go next
- OCR Inbox & Document Queue — the Errors view's home.
- Creating an Invoice from OCR — where a successful reprocess lands.
- Operations Scheduler — the loop that processes the queue.
Frequently asked questions
It replaces the record's extraction result with the fresh read — the document and the intake history stay. Since no invoice exists yet (the guard ensures it), nothing in your books is touched.
Reprocess is per-record (it opens the review form), so a service-blip burst is worked by opening each item from the Errors view. Avoid re-forwarding the same emails as a shortcut: every arriving PDF creates a new intake record and counts against the monthly document meter — reprocessing the existing record doesn't.
Once a Purchase Invoice has been created from the record: "Cannot reprocess — a Purchase Invoice has already been created from this record." At that point the invoice is the source of truth — correct it directly rather than re-reading the document underneath it.
The meter counts documents as they enter the queue — each intake record increments the month's usage once. Reprocessing re-reads a document that has already been counted, on its existing record.
Both work. Create Invoice lets you hand-pick the vendor in the form right now; Reprocess re-runs the full extraction so the vendor match (and that vendor's coding history) lands automatically. If you expect more bills from this vendor, reprocess once — it starts the learning loop properly.
Was this article helpful?