Document Insights

Document Insights is the part of the AI Document Agent that evaluates the supporting documents a user or business uploads during verification — for example a proof of address, utility bill, bank statement, or business registration document. For each document it produces an overall decision (APPROVED, REVIEW, or DECLINED) together with the specific warnings, extracted data, and risk indicators behind that decision.

Product Overview

Supporting documents are one of the easiest things for a bad actor to manipulate — a genuine file can be edited, recreated, screenshotted, or flattened into an image to hide changes. At the same time, the details on a document need to actually match what the applicant told you.

Document Insights automates both checks. It inspects each uploaded document for signs of manipulation and reads its contents to confirm the document is what it claims to be and that its data lines up with the supplied information — returning a single, explainable decision your team can act on.

How It Works

Every document is assessed along two dimensions:

• Is the document authentic? An automated document-fraud check inspects the file's structure, metadata, content-version differences, and rendering signatures for signs that the document was tampered with, recreated, screen-captured, or converted to an image — i.e. that it is no longer a genuine, unaltered original.
• Does its content check out? An extraction and validation step reads the document's data, confirms the document type, applies any validation rules configured on your template, and compares the extracted fields (name, address, dates, IDs, and so on) against the data supplied for the verification.

What Document Insights Returns

• An overall document result (APPROVED, REVIEW, or DECLINED) reflecting the most severe signal found.
• Warning codes describing each issue detected, split into document-authenticity signals and data & validation signals.
• Extracted data — the fields read from the document.
• Indicators that explain the authenticity assessment, in three separate arrays.

Indicators

The response carries three indicator arrays, not one:

A document can carry entries in all three at once. Read them together rather than treating any one array as the verdict, and use the overall result for the decision.

Each entry looks like this:

{
  "id": "sandbox_trust_indicator",
  "category": "document_structure",
  "title": "Document structure is consistent",
  "description": "No signs of tampering detected in the document structure."
}

indicator_attributes and metadata may also be present on an entry. Fields with no value are omitted rather than returned as null, so read defensively.

id and category come from the underlying document-fraud provider and are not a fixed AiPrise set, so treat them as opaque values. Display title and description, and avoid branching on id unless you have confirmed the specific value you depend on.

Document Insights and Proof of Address

Four names appear around this feature. They relate like this:

Proof of Address is a use case: what the user uploads, and what a template asks for. Document Insights is the engine that evaluates it. They are not alternatives, and in the response neither one contains the other.

They are sibling shapes, not nested

A proof-of-address upload comes back as one of two shapes. They share a common core and each adds its own fields.

The twelve shared fields are the document-authenticity core, which is why the two shapes look alike. The address-specific half, name and address matching and the dates behind recency, exists only on ADDRESS_VERIFICATION.

📘

If you are migrating from older documentation

DocumentInsights used to contain an address_verification_info object. That was accurate until May 2024, when address verification was separated into its own section. The field has not been returned since, and the API reference has been corrected. If your integration still reads it, it has been receiving nothing.

Which Response Shape You Get

Both shapes arrive in the same place: the additional_info array on the verification response. Each entry carries what you sent, which shape came back, and the data itself.

{
  "additional_info": [
    {
      "additional_info_type": "ADDRESS_PROOF_DOCUMENT",
      "additional_info_response_type": "ADDRESS_VERIFICATION",
      "data": { "...": "shape depends on the line above" }
    }
  ]
}

additional_info_response_type has four values:

Branch on additional_info_response_type, not on what you uploaded. The same ADDRESS_PROOF_DOCUMENT upload can return either ADDRESS_VERIFICATION or DOCUMENT_INSIGHTS. Which one you get depends on the provider that runs the check for that country and template, so it can differ between two templates in the same account, and it is not a setting you select directly.

If you handle only one of the two, a template routed to the other provider will look to your integration like an empty or unrecognised result.

For the proof-of-address upload itself, the accepted file types, sub-types and template configuration, see User – Proof of Address (POA).

Automated Decisioning

Document Insights returns an overall document result that reflects the most severe warning-level decision raised on the document:

Each individual warning also carries its own default decision (see the warning tables below). These defaults are a starting point — on any template you can configure the decision applied to each warning code to match your risk tolerance. Some warnings (such as certain field mismatches) default to APPROVED, so an approved result does not necessarily mean every extracted field matched the supplied data.

Warning Codes

The full list of warnings Document Insights can raise, along with each warning's default decision, is documented alongside the other verification warnings:

• User Verification → Warning Codes
• Business Verification → Warning Codes