REDA One Help Center
EN
Language
English EN Español ES Português PT Italiano coming soon Français FR Deutsch coming soon العربية coming soon
reda.one ↗
Home / REDA Financials AI / Bulk Operations / Batch Processing / OCR Processing Batch

OCR Processing Batch

The conveyor belt behind document intake. The card: "Automatically processes incoming documents through AI extraction and creates purchase invoices. Runs every 15 minutes when active."

Getting there. Open the App Launcher (the grid icon at the top-left of any page) → REDA Financials Admin → the Financial Admin Console tab → Operations, then find the OCR Document Processing card to Start or Stop the queue processor.
Access you'll need. Starting and stopping batch jobs requires the Accounting Admin v2 permission set. The Accounting Controller v2 and Read Only v2 sets can open the console to view outcomes; the Accounts Payable v2 and Accounts Receivable v2 sets cannot see it, though they work the draft invoices the loop produces from the OCR Inbox.

The essentials

  • A queue processor, not a nightly job: this card works differently — you Start it and it cycles every 15 minutes, draining the OCR Inbox; Stop pauses new cycles (work in flight finishes).
  • What each cycle does: takes pending inbox documents through AI extraction (vendor, dates, amounts, lines), matches them to your master data, and creates draft purchase invoices for review.
  • Queue visibility: the card shows the pipeline — pending, extracting, errored — so a glance tells you whether intake is flowing.
  • Uses AI document credits: each processed document draws on your plan’s monthly allowance (visible on the Subscription tab).
The expanded OCR Document Processing card. Unlike the scheduled batches it carries a DOCUMENT QUEUE strip with three counters — PENDING 0 in blue, EXTRACTING 0 in amber and ERRORS 0 in red — and instead of a Run Time it offers a Start button that begins a processing loop which reschedules itself every 15 minutes, beside a Manual Run panel whose Run Now button executes a single cycle without starting the loop.
One of the two cards that breaks the pattern. There is no run time here because OCR is a loop, not an appointment: Start sets it going and it re-schedules itself every 15 minutes. The queue counters — PENDING, EXTRACTING, ERRORS — are how you read its health at a glance.

What the loop moves — the document lifecycle

Every document the loop touches lives on an OCR Inbox record whose Status says exactly where it is. The system’s own definitions:

StatusMeaningWho acts
PendingReceived (usually via email), awaiting the next 15-minute cycleNobody — the loop picks it up
Processing / ExtractingAI at work — first read done, full extraction underwayNobody
ExtractedData ready; confidence below the auto-create bar — awaiting a humanAP reviews via Create Invoice
CreatedA draft Purchase Invoice exists, linked on the recordNormal AP flow takes over
ErrorProcessing failed; the reason is stored on the recordFix the cause, Reprocess
Reviewed / RejectedBookkeeping marks for human-verified and deliberately discarded itemsYour team’s convention

The loop works the queue oldest-first, one document at a time — a document that fails is marked Error with its reason, and the loop carries on with the next one. One bad scan never jams the queue.

Why some documents become invoices by themselves. A territory setting, OCR Auto-Create Threshold, decides the hand-off: "Minimum AI confidence (0-100%) to auto-create invoices from OCR. Below this threshold, documents require manual review. Default: 80%." At or above the bar, a draft Purchase Invoice is created automatically and the record goes straight to Created; below it, the record stops at Extracted — exactly what the Awaiting Review list view collects. Either way, everything the AI creates is a draft: the normal posting gates stand between extraction and your ledger.

Reading the card’s queue counters

The card’s Document Queue counters — pending | extracting | errored — are the queue’s dashboard-at-a-distance: an administrator sees work waiting and failures accumulating without opening the OCR Inbox tab. Pending steadily draining is healthy. Stopping is safe, and the button says so: "Stop the OCR processing loop. Pending documents will remain in the queue until processing is restarted." — nothing is lost; the queue just waits.

Troubleshooting the loop

SymptomLikely cause → action
Queue growing, nothing processingThe processor is stopped — press Start. If started and stuck, check the last audit entry for the error.
Everything erroring at onceUsually the document cap: the monthly AI allowance is exhausted — "Monthly AI document processing limit has been reached. Please contact your administrator." Confirm on the Subscription tab.
Errors like "AI service is temporarily rate-limited. Please try again in a few moments."Transient conditions on the AI service — nothing is wrong with your documents. Reprocess after a short wait.
"AI API key is not configured. Please activate your REDA Financials AI license."The org has no active AI licence — an administrator activates the key in the Subscription tab.
Every document counts against the monthly meter — junk included. Counting happens at intake, so a non-invoice PDF emailed to the address consumes a metered document even though extraction will refuse it. Keep the intake address for vendor bills, watch the Document Consumption card on the Subscription tab ("Used This Month" / Remaining / "Monthly Cap"), and know that when the cap is reached intake pauses with a clear message rather than degrading quietly — processing resumes when the new month opens.
A morning at Meridian’s AP desk. Meridian Properties Group receives vendor bills by email, so the queue fills itself overnight. Priya opens the Awaiting Review view first — a utility bill whose scan quality left confidence below the bar gets Create Invoice and a sanity-check; a bill from a brand-new vendor gets its vendor account created, then Reprocess so the match resolves; a statement forwarded by mistake is marked Rejected, kept as evidence. Then a one-minute glance at Errors: one transient rate-limit message, reprocessed after coffee. Queue clear by 9:15.

Common mistakes to avoid

MistakeWhy it hurtsDo this instead
Waiting for Pending records to move while the loop is stoppedNothing processes itself — the queue grows silently and vendors start chasingCheck the card; press Start — it re-schedules itself every 15 minutes
Lowering the auto-create threshold to shrink the review queueLow-confidence extractions become draft invoices unreviewed — errors move downstream into APKeep the review bar meaningful; Awaiting Review exists so doubtful reads get human eyes
Ignoring the errored counter for weeksMost errors are one-click retries when fresh; stale ones become archaeology while real bills sit unpaidGlance at the Errors view (or the card’s counter) daily — see Reprocessing Failed Extractions
Routing statements and receipts through the intake addressThey consume monthly allowance and are refused at extraction — only vendor bills travel the pipeline todaySend vendor bills; keep general correspondence out of the intake mailbox
Treating extraction as postingNobody reviews, errors reach the ledger late, trust in the feature erodesEvery created invoice is a Draft on purpose — verify against the document before posting

Operating it

Most orgs start it and leave it running — 15-minute freshness means an emailed invoice becomes a reviewable draft before the coffee’s done. Errored documents stay visible in the inbox with reasons, ready for reprocessing after fixes. Run Now forces an immediate cycle.

Where to go next

  1. OCR Overview — the pipeline end to end.
  2. Creating an Invoice from OCR — the reviewer’s workflow.
  3. Operations Scheduler Overview — the console.

Frequently asked questions

The processing loop isn’t running. It’s a Start/Stop loop, not a daily-time job — press Start on the card and the queue drains on the 15-minute cycle. A Pending item is normal for a few minutes; a pile of them means the loop is stopped.

No — the button’s own text explains: "Stop the OCR processing loop. Pending documents will remain in the queue until processing is restarted." Work in flight finishes, nothing is lost, and Start resumes exactly where it left off.

Never — created invoices are always drafts, and the normal posting gates (periods, approvals, your click) stand between extraction and the ledger. The AI types; you decide.

No — it runs as a system context, consuming no seat. What the AI plan meters is document volume: each processed document counts against the org-wide monthly allowance, visible on the Subscription tab.

Its extraction confidence met the territory’s OCR Auto-Create Threshold (default 80%), so a draft Purchase Invoice was created automatically and the record went straight to Created. The invoice is still a draft — review it there before posting.

Vendor bills → Purchase Invoices, today. The intake record reserves other types (Expense Receipt, Bank Statement) for future phases — treat those as roadmap, not features. Non-invoices are refused at extraction: "The uploaded document does not appear to be an invoice."

Was this article helpful?