Getting the Most Out of AI SmartDoc

SmartDoc works best on clean, single-document PDFs with predictable layouts. Good scans, native PDFs (over image-based PDFs), tight cropping, and consistent document type names all improve accuracy. Use SmartDoc for routine, high-volume document intake; use manual reminder creation for one-off items where typing is faster than reviewing extraction. Watch your monthly token budget — heavy users can hit the cap.

When SmartDoc is the right tool

Use SmartDoc when:

You're processing 5+ documents in a session.
Documents follow predictable formats (recurring renewals of similar items).
The document has clear, typed text with a visible expiry date.
The original is a native PDF or a clean scan.

Skip SmartDoc when:

It's a single one-off document and typing the fields would take 30 seconds.
The document is heavily handwritten or low-resolution.
Multiple unrelated documents are bundled into one PDF (split them first).
Your data policy forbids third-party AI processing for this content.
You're nearing your monthly token cap and have routine documents you could process manually.

Tips that materially improve accuracy

1. Use native PDFs over scanned PDFs

A PDF where you can highlight and copy the text ("native PDF") extracts much better than a PDF that's just images ("scanned PDF"). When you have a choice — downloading from a portal, for example — pick the native version.

2. Scan at 200+ DPI when scanning paper

Phone photos and 72-DPI scans force SmartDoc to work harder. 200–300 DPI scans, with good lighting and minimal skew, extract reliably.

3. One document per upload

Don't bundle. SmartDoc treats a multi-document PDF as a single document and produces confused, mixed-up extraction. Split before uploading.

4. Crop tightly

Remove cover pages, blank pages, unrelated content, and over-large margins. Less noise = better extraction (and lower token usage).

5. Use consistent document types

SmartDoc tends to propose new document types when it doesn't recognise an existing match. If you call something "Trade License" on one item and "License - Trade" on another, you're going to end up with duplicate type entries. Stick to a single naming convention.

When SmartDoc prompts to "Create and select a new document type…", pause: is there an existing type that fits? If yes, switch to it. Only create truly new types.

6. Use a sensible custom-field schema

If your workspace defines custom fields, give them clear, consistent names. SmartDoc has a better chance of matching "Policy Number" reliably than two fields named "Policy No" and "Policy #" inconsistently.

7. Review extraction before saving — always

The form is editable for a reason. Even at 95% confidence, individual fields can be wrong, and the wrong field most likely to break things is the expiry date.

Managing your monthly token budget

SmartDoc usage is capped per user per month by token count:

Plan ID 1 (entry plan): 20,000 tokens per user per month.
Other plans: 300,000 tokens per user per month.

What that means in practice:

A short clean document (a one-page certificate) uses relatively few tokens.
A long, dense, multi-page PDF uses many more.
When you hit the cap, scans are blocked until the cap resets next month.

If you're processing high volumes:

Don't waste budget on bundled or unsplit PDFs — they consume tokens for poor extraction.
Don't re-scan the same document repeatedly to "see if it gets better" — each scan consumes tokens.
Crop tight scans — pages of blank or irrelevant content still consume tokens.
Spread heavy intake across multiple users rather than one user blowing through the cap.

If your team consistently hits the cap, the answer is a plan upgrade — talk to support.

Workflows at scale

If you're onboarding many existing documents into Remindax:

Option A — SmartDoc per document. Best for accuracy. Upload one, review carefully, save. Slow but each item gets full attention.

Option B — Bulk import + SmartDoc for new arrivals. Use the reminder import template to onboard everything you already have records for (you've already typed the data). Use SmartDoc going forward for new documents as they arrive. This combines speed (import) with quality (SmartDoc on novel documents) without burning your token budget on bulk historical data.

Option C — Pre-filter then SmartDoc. Sort your inbox — routine document types go through SmartDoc; one-offs or unusual items get manual entry. The 80/20 rule applies.

How to handle low-confidence extractions

When SmartDoc returns a low confidence percentage or obviously wrong fields:

Don't save and figure it out later. Fix it now — once saved, you're editing the reminder anyway.
Check the original document. Sometimes the issue is the document itself (no clear expiry date, multiple ambiguous dates), not SmartDoc.
Note the document type. If SmartDoc consistently struggles with one type (e.g. handwritten work orders), note it and skip SmartDoc for that type going forward.
Consider whether re-scanning helps. It usually doesn't — and it costs tokens. Edit the form manually instead.

What SmartDoc isn't for

Legal contract analysis — SmartDoc extracts identifying fields, not legal content.
Bulk OCR of large libraries — for thousands at once, talk to support.
Renewals on existing reminders — SmartDoc always creates new reminders. Manually attach renewed documents.
Highly sensitive content under strict data classification — see SmartDoc overview.
Replacement for review. A wrong expiry on a SmartDoc reminder is worse than no reminder at all.

Edge cases & gotchas

Documents in multiple languages vary in accuracy. Non-Latin scripts may extract poorly.
Tables and forms with labeled fields extract well. Free-form contracts vary.
Documents with multiple "expiry" candidates (issue date, renewal date, policy end date) confuse SmartDoc. Manually verify.
Watermarks or stamps over key fields can mask the text.
Heavy graphics behind text can reduce accuracy.
DOCX vs PDF/image processing differs slightly — DOCX is read locally first, then text sent to OpenAI; PDFs and images go directly.