June 10, 2026 · Updated September 7, 2026
An application can arrive as a fillable PDF, a scan, a partly completed form, or a packet of documents that disagree. The task is to turn those sources into information the team can use without making uncertainty disappear.
Extraction and generation are different steps. Extraction reads information from a document. Generation uses reviewed information to prepare a new form. Test both steps if your goal is a usable application.
Identify the document before reading the fields
Record the form number, edition, account, and relevant period. Common commercial documents include the ACORD 125 application and line-specific sections such as the 126, 130, and 140. A prospect may instead start with dec pages or a schedule. The file received should guide the next question.
Confirm licensed access and current editions through your provider or the ACORD Forms Portal. A document found online is not automatically the correct form for the intended market.
Preserve three kinds of uncertainty
Unreadable: the source may contain an answer, but the text or mark cannot be read confidently. Ask for a clearer document or review it with the client.
Missing: the source does not provide an answer. A blank field is not zero, no, or not applicable. Keep it unanswered until the appropriate person confirms it.
Conflicting: two sources provide different values. Preserve both sources and resolve the difference. Do not let the first extracted value silently win.
For example, a property application lists one roof year and a newer client email lists another. Accurate extraction of the older number does not make it current. The reviewer needs the relevant source and date to decide what belongs in the new application.
Use a field review table
| Field | Source evidence | Review question |
|---|---|---|
| Named insured | Current declarations and client information | Is the correct legal entity applying? |
| Location | Schedule, declarations, or client answer | Has the operation moved or expanded? |
| Exposure figure | Stated payroll or sales and its period | Is this historical or projected for the requested term? |
| Coverage detail | Declarations and applicable forms | Does the summary omit a relevant condition or endorsement? |
| Unanswered question | No supported source | Who must provide the answer? |
The table is a review method, not a promise that every extraction product displays these exact fields or states. Ask the vendor to show how its interface makes the same questions answerable.
Evaluate tools with the documents you receive
Include a clear digital file, a scan, a partially completed form, and a packet with one known conflict. Record correct values, incorrect values, missing values, and how easily a colleague can verify them.
Generic OCR, document services, and insurance-focused tools have different capabilities. Avoid assuming that a category name establishes accuracy. A useful output needs the correct meaning and period, not just correctly recognized characters.
Keep the full source available. Reviewing only fields a tool flags assumes the tool detected every important uncertainty. The agency should define the additional checks required for its application process.
Carry the review into the generated form
In Relay, evaluate a supported document-to-application workflow with the files your team normally receives. Inspect the extracted facts and the actual draft. Check layout, form edition, missing answers, and the requirements of the intended market before use.
The document parsing page covers the input work. The ACORD generation page covers the draft deliverable. Success is a form the reviewer can use with a clear understanding of what still needs attention.
See a worked example of moving dec-page facts into an ACORD draft.
Start with the client documents on your desk and review the application Relay prepares.