Good results start before recognition
IDP accuracy depends on the documents you receive and how your setup is configured. Well-designed documents, clean images, clear rules and complete validation fields all reduce errors.
Introduce IDP gradually: start with a small set of representative documents, test them, and expand once results are stable.
Design documents for automatic reading
If you control the template – such as job sheets, forms or timesheets – design it with automatic reading in mind:
- Avoid grey or coloured bands across rows; they reduce contrast and confuse cell detection.
- Use solid black grid lines on a white background, thick enough to be clear.
- Leave enough space in each cell so handwriting stays inside it.
- Give columns short, clear headings such as Date, Resource, Hours or Site.
Capture good images
- Scan PDFs at around 300 DPI.
- When photographing, capture the whole page, keep the phone parallel to it, avoid shadows and don’t fold or crop the edges.
- Keep documents the right way up – only enable auto-rotation if they often arrive rotated.
- Ask operators to write clearly, especially numbers such as 1, 7, 4, 9, 0 and 6, and to use consistent terms.
Configure in the right layer
| Put it in… | When the rule… |
|---|---|
| General rules | Applies to every document the assistant handles. |
| Document type rules | Applies to a category, such as all delivery notes. |
| Layout rules | Applies to a specific page or template. |
| Area rules | Applies to one field or section. |
Avoid repeating the same rule in several places – it becomes hard to maintain and can lead to contradictions. Create separate layouts for pages with different structures, and use crosspage_data to carry details such as a document number across pages.
Make validation part of your data model
Set up validation fields for every value your system needs as a code – resources, job sites, work phases, units, hour types, vehicles and payment methods. Include the code, the full name and the aliases people actually use.
Plan quality control
- Decide which documents can be imported automatically – for example high confidence and all required fields validated.
- Send documents with low confidence or missing essential data for manual review.
- Keep the original document, the extracted JSON and any corrected version together.
- After go-live, review corrections and unrecognised cases regularly and update your rules, layouts and validators.
Production readiness checklist
Before you go live, check that you have…
- Real sample documents – clean, average and difficult.
- Clear document types with shared rules in one place.
- Separate layouts for different page structures, with carefully drawn areas.
- Specific, non-contradictory AI rules.
- Validators with codes and aliases for recurring values.
- Only authorised WhatsApp numbers registered, with each operator’s language set.
- Tested everything with the IDP processing test.
- Agreed thresholds, error handling and manual review for your integration.
Prefer a guided walkthrough? Book a training session and an Echo7 specialist will help your team set it up.
Book a training sessionWant to go deeper?
This guide covers the essentials. For detailed reference on every setting and feature, explore the full Echo7 Wiki.
Echo7 Wiki