Keep Save and send manual while you learn a parser, then enable auto-send for trusted destinations without skipping duplicate or packet checks.
Create your first parser in minutes. No code, no setup calls.
Authorize Parsedit from Zapier, map Upload Document or New Parsed Document to a parser, and manage connections from the Integrations page.
Manual approval means a human clicks Save & send (or saves without sending). Auto-send means extracted results go to enabled destinations when processing finishes, without that click.
Both are valid. Mixing them up on day one is how test vendors appear in production QuickBooks.
Auto-send policy differs for Sheets staging and ledgers.
Manual review is the default. Auto-send is a parser destination setting you turn on after you trust extraction and destinations are connected. There is no confidence-threshold auto-approval path. Docs: Automation.
| Destination | Early policy | Later policy |
|---|---|---|
| Google Sheets staging tab | Manual, then auto-send | Auto-send common |
| Slack / Zapier notify | Often auto | Auto |
| Webhook to staging API | Manual until idempotent | Auto |
| QuickBooks / Xero | Manual | Manual longer; auto only when dull |
When auto-sending packets or multipage files, confirm preferred send mode on manual sends first: Main document only, Each document separately, or Everything together. Packets, Send modes explained.
ledger_status = not_postedHuman review philosophy: Why human review still wins.
Parsedit gives you the switch and the review queue. Soft CTA: leave ledger destinations manual through your first month of real volume.
Week 1–2: manual Save & send only. Track edit rate and destination errors in a simple sheet.
Week 3: auto-send to a non-ledger destination (Sheets staging or Slack). Keep QuickBooks/Xero manual.
Week 4: if edit rate is low on top vendors and duplicate rules behave, consider auto-send for low-risk parsers only.
Roll back immediately if vendor-name creates spike or if packet children post incorrectly. The goal is boredom, not bravado.
A trusted parser is not “the model scored high once.” It is:
If any item fails, stay manual. Revisit after you fix the template or intake filters.
Put these in an internal runbook so every new teammate does not guess:
Revisit quarterly. Volume and vendor mix change.
Practice once: enable auto-send on a staging parser, send a bad sample on purpose, turn auto-send off, confirm new docs sit in Ready to review, and fix the bad row downstream. A team that has never rolled back will freeze when something noisy hits production.
Parser destinations: Sheets Active + QuickBooks Active. Auto-send on for the parser after week 3.
Reality check: confirm product behavior for your build. If auto-send delivers to all active destinations, do not enable auto-send while QuickBooks is Active unless you intend ledger posts. In that case keep QBO disconnected or inactive until you are ready, or keep auto-send off and manually select Sheets only.
The safe operational pattern many teams use: auto-send only on parsers whose sole active destination is Sheets/Slack/webhook-staging. Keep a separate step for ledger sends.
Name which destinations can auto-send early and which should stay manual. Vague “use AI responsibly” sections help no one.
Can we auto-send only some fields? Shape the template; destinations receive the extracted document according to their connectors.
How do we pause auto-send quickly? Turn it off in destination settings; documents will wait as ready to review.
Does auto-send bypass duplicate rules? Intake dedupe and accounting document-number rules still apply as documented.
Run two quiet weeks manually, then auto-send Sheets only. Start free.