PerfectParser
All insights
Guides

How to Automate ACORD Form Data Extraction with AI (Forms 25, 28, 125)

How to automate ACORD form data extraction with AI. Extract fields, checkboxes, and coverage tables from ACORD 25, 28, and 125 forms into JSON or Excel.

Mike SmithMike Smith
12 min read
How to Automate ACORD Form Data Extraction with AI (Forms 25, 28, 125)

ACORD form data extraction uses AI to read completed ACORD 25, 28, and 125 forms and return their contents as structured JSON, CSV, or Excel. It replaces manual re-keying of policy numbers, limits, dates, checkboxes, and certificate holder details. The right approach differs by form: a dense limits grid on the ACORD 25, lender and loss-payee context on the ACORD 28, and multi-row tables on the ACORD 125. Tools like PerfectParser handle the reading side so downstream systems receive verified data.

ACORD forms are standardized, but getting their data into your systems usually isn’t. A certificate of insurance arrives as a scanned PDF. Someone opens it beside the agency management system (AMS) and retypes the limits, policy numbers, and holder details one field at a time.

This guide explains how ACORD form data extraction works with AI. It covers what to expect from the three most common forms (ACORD 25, 28, and 125), how to validate the output, and how to test a tool on your own documents.

What Is ACORD Form Data Extraction?

ACORD form data extraction is the process of reading a completed ACORD form and returning its contents as structured fields instead of a flat image. A purpose-built ACORD parser handles four key operations:

  1. Identifies the form type: Distinguishes between ACORD 25 certificates of liability, ACORD 28 property evidence, ACORD 125 applications, and attached endorsements or schedules.
  2. Locates each field and section: Maps visual bounding boxes to semantic field definitions across producers, insurers, coverages, limits, and certificate holders.
  3. Reads values and checkbox states: Classifies binary tick marks (OCCUR vs. CLAIMS-MADE, ADDL INSD, SUBR WVD) while parsing handwritten or printed alphanumeric values.
  4. Outputs structured data: Emits clean JSON, CSV, or Excel ready for ingestion into AMS360, Applied Epic, Salesforce, or underwriting pipelines.

Why Manual ACORD Data Entry Is a Bottleneck

If you want to reduce manual ACORD form data entry, start by seeing where the time goes. For each form, an agency specialist typically:

  • Opens the PDF and identifies the form and edition.
  • Re-keys the insured, policy numbers, dates, and limits into the AMS.
  • Checks each box by eye (occurrence vs. claims-made, additional insured, subrogation waived).
  • Copies the certificate holder or mortgagee details.
  • Double-checks everything, because a transposed digit can flow into an active coverage record.

Three real costs add up quickly:

  • Labor costs: Minutes per form, multiplied across hundreds or thousands of certificates every month.
  • Transposition errors: A wrong digit in a policy number or limit can trigger compliance disputes or delayed loan closings.
  • Intake turnaround delay: Documents sit in a queue while tenants, vendors, lenders, and brokers wait for verification. For commercial agencies, certificate processing is part of a larger intake challenge explored in our guide to commercial broker insurance data capture.

How to Extract an ACORD 25 Certificate of Insurance

The ACORD 25 (Certificate of Liability Insurance) is one of the highest-volume documents in commercial insurance. Because it is largely a dense matrix, accurate extraction depends on keeping every limit tied to its exact coverage row.

SectionTypical fields to extract
Producer & InsuredProducer name, contact details, named insured, mailing address
Insurers Affording CoverageInsurer letter identifiers (A through F) and respective National Association of Insurance Commissioners (NAIC) codes
Commercial General LiabilityPolicy number, effective & expiration dates, occurrence vs. claims-made, each occurrence limit, damage to rented premises, medical expense, personal & adv injury, general aggregate, products/completed operations aggregate
Automobile LiabilityPolicy number, effective & expiration dates, auto-type checkboxes (any auto, owned, scheduled, hired, non-owned), combined single limit or split bodily injury/property limits
Umbrella / Excess LiabilityPolicy number, dates, occurrence vs. claims-made, each occurrence limit, aggregate, retention amount
Workers CompensationPolicy number, dates, statutory vs. other limits, E.L. each accident, E.L. disease each employee, E.L. disease policy limit, officer exclusion Y/N
Row-Level CheckboxesAdditional Insured (ADDL INSD), Subrogation Waived (SUBR WVD)
Description of OperationsFree-text descriptions, location schedule references, special endorsement wording
Certificate HolderHolder legal name, address, attention line
Cancellation & SignatureNotice period wording, authorized representative signature block, issue date

Why it's hard: The same limit label (e.g., "$1,000,000") can appear under general liability, auto, and umbrella sections. Furthermore, the ADDL INSD and SUBR WVD checkboxes sit on a narrow row separated from the policy number they modify. PerfectParser uses vision-based spatial awareness to lock every limit and checkbox to its corresponding coverage line.

ACORD 28 Data Extraction: Evidence of Commercial Property Insurance

The ACORD 28 is issued when a lender, mortgagee, or loss payee requires evidence of commercial property coverage. Unlike the liability-focused ACORD 25, the ACORD 28 is organized around the physical property location and interested parties.

SectionTypical fields to extract
Agency, Insurer & InsuredProducer contact, carrier name, NAIC, named insured legal entity
Policy InformationPolicy number, effective date, expiration date, policy type
Location & PropertyProperty description, physical location address (street, city, state, ZIP)
Coverage & LimitsCoverage type (building, contents, business income), limit amount, deductible, causes of loss (special, broad, basic)
Coverage ConditionsCheckbox conditions: replacement cost, actual cash value, agreed value, coinsurance percentage
Additional InterestMortgagee, loss payee, or lender's loss payable name, address, loan/account number
Cancellation ProvisionsNumber of days for notice of cancellation to additional interests

Why it's hard: The same address structure appears multiple times (named insured address, physical property location, and mortgagee billing address). Contextual parsing is required to ensure loan numbers and loss-payable clauses map to the correct financing record. Because ACORD 28 forms summarize complex commercial assets, cross-referencing extracted limits with an AI insurance policy schedule parser ensures property sub-limits align with master policies.

ACORD 125 Commercial Application OCR

The ACORD 125 is the core applicant cover sheet for commercial submissions. As the front page of a multi-line insurance submission, it is the most information-dense of the three forms.

SectionTypical fields to extract
Agency & CarrierAgency code, applicant ID, proposed policy effective and expiration dates
Lines of BusinessCheckboxes for commercial property, general liability, commercial auto, umbrella, workers comp, inland marine, boiler & machinery
Applicant InformationLegal name, DBA, mailing address, premises address, FEIN, SOS entity type, website, inspection contact
Premises InformationMulti-row table of locations: street, city, state, ZIP, building number, leased vs. owned
General InformationMulti-question Yes/No table with explanation fields for past bankruptcies, uncorrected fire violations, or prior cancellations
Loss HistoryPrior carrier, line of business, date of occurrence, amount paid, amount reserved, claim status
Signature BlockProducer signature, applicant signature, date signed

General contractors and developers who manage high-volume subcontractor applications can coordinate certificate validation directly with their construction AP automation to unblock lien waivers and payment applications.

ACORD 25 vs. 28 vs. 125 at a Glance

AttributeACORD 25ACORD 28ACORD 125
Primary PurposeCertificate of Liability InsuranceEvidence of Property InsuranceCommercial Insurance Application
Primary ConsumerClients, vendors, general contractorsLenders, mortgagees, loss payeesUnderwriters, brokers, MGA raters
Key Failure ModeCheckbox misalignment, wrong limit rowAddress conflation, loan number dropsMulti-page tables, nested premises
Volume PatternHigh frequency, ongoing renewalsTriggered by loan closing or refinancingPeriodic at new submission or renewal
Critical FieldsDates, occurrence limits, holder addressLoan number, property address, perilsFEIN, premises schedule, loss runs

The Hard Parts: Checkboxes, Grids, and Scans

Checkbox State Detection

A checkbox is not text, so simple text-only OCR engines either omit it completely or misinterpret a cross as an X. An AI document parser classifies checkboxes into three distinct states: checked, unchecked, or uncertain. On an ACORD 25, correctly distinguishing OCCUR from CLAIMS-MADE or confirming whether SUBR WVD is selected prevents substantial coverage verification errors.

Dense Multi-Box Grids

Limits and deductible amounts are packed into tight cells. If vertical row boundaries shift, a $2,000,000 general aggregate can accidentally be read as the per-occurrence limit. Layout-aware vision models examine row geometry and label headers simultaneously to ensure values land in the right schema slot.

Fillable PDFs vs. Scans

Not every ACORD document requires the same processing path:

  • Fillable digital PDFs: Contain digital form fields (AcroForms) that allow deterministic field-level reading.
  • Scanned, faxed, or photographed PDFs: Require visual layout parsing, deskewing, and intelligent character recognition.

A production pipeline should automatically route clean digital documents through instant extraction and apply multi-modal vision models to degraded scans.

Form Editions

ACORD updates form revisions periodically (e.g., ACORD 25 edition 2016/03 vs. earlier revisions). Coordinate-based zonal OCR breaks whenever a carrier uses an alternate edition or custom margins. Vision models parse relative semantics, ensuring consistent output regardless of edition variations.

AI Solution for ACORD Form Data Entry: The Architecture

An enterprise ACORD data extraction pipeline operates across seven automated stages:

  1. Ingest: Files arrive via REST API, SFTP, shared inbox, or batch file upload.
  2. Classify: AI classifies whether the incoming PDF is an ACORD 25, ACORD 28, ACORD 125, or an attached policy schedule.
  3. Parse Layout: Spatial analysis detects grid headers, column separators, and checkbox clusters.
  4. Extract: Field values, checkboxes, and line items are mapped into a standardized JSON schema.
  5. Validate: Automated business rules run cross-field consistency checks (e.g., aggregate ≥ occurrence).
  6. Human Review: Low-confidence fields or validation discrepancies are flagged for side-by-side operator review.
  7. Deliver: Clean data is delivered via webhooks or API directly into AMS360, Applied Epic, or your custom database.

Sample JSON Output and API Workflow

Here is an illustrative output schema from an ACORD 25 extraction:

{
  "form_type": "ACORD_25",
  "edition_date": "2016/03",
  "producer": {
    "name": "Beacon Risk Partners",
    "contact": "Jane Doe",
    "phone": "555-0199"
  },
  "insured": {
    "name": "Apex Contracting LLC",
    "address": "450 Industrial Parkway, Austin, TX 78701"
  },
  "certificate_holder": {
    "name": "Midwest Development Corp",
    "address": "1200 Commerce St, Dallas, TX 75201"
  },
  "coverages": {
    "general_liability": {
      "insurer_letter": "A",
      "policy_number": "GL-889102-01",
      "effective_date": "2026-03-01",
      "expiration_date": "2027-03-01",
      "basis": "occurrence",
      "additional_insured": true,
      "subrogation_waived": true,
      "limits": {
        "each_occurrence": 1000000,
        "damage_to_rented_premises": 300000,
        "med_exp": 10000,
        "personal_adv_injury": 1000000,
        "general_aggregate": 2000000,
        "products_comp_op_aggregate": 2000000
      }
    },
    "workers_compensation": {
      "insurer_letter": "B",
      "policy_number": "WC-449102-09",
      "effective_date": "2026-03-01",
      "expiration_date": "2027-03-01",
      "statutory_limits": true,
      "limits": {
        "el_each_accident": 1000000,
        "el_disease_ea_employee": 1000000,
        "el_disease_policy_limit": 1000000
      }
    }
  }
}

Submitting an ACORD document through the PerfectParser REST API is straightforward:

import requests

url = "https://api.perfectparser.com/v1/extractions/start"
headers = {"X-API-Key": "YOUR_API_KEY"}

with open("acord25_certificate.pdf", "rb") as f:
    response = requests.post(
        url,
        headers=headers,
        files={"file": f},
        data={"parser_id": "acord-forms-v1"}
    )

job = response.json()
print("Job submitted:", job["job_id"])

Validation Rules That Catch Real Errors

Extraction without validation merely accelerates errors. A production-ready pipeline applies these validation rules:

ACORD 25

  • Chronology: Effective date must precede expiration date.
  • Limit Logic: General aggregate must be greater than or equal to the each-occurrence limit.
  • Completeness: Policy number must be present for every checked coverage row.
  • Holder Validation: Certificate holder name and address must match the contracting counterparty.
  • Endorsement Flags: Ensure ADDL INSD and SUBR WVD match contractually required endorsements. For vendor onboarding and subcontractor management, compliance teams frequently pair certificate intake with contract data extraction to verify that certificate limits satisfy underlying master services agreements (MSAs).

ACORD 28

  • Location Matching: Property location address must match portfolio assets on file.
  • Interest Verification: Additional interest name and loan/mortgage account number must reconcile with lending systems.
  • Term Coverage: Policy dates must cover the entire financing or escrow period.

ACORD 125

  • FEIN Validation: Federal Employer Identification Number must follow standard 9-digit format (XX-XXXXXXX).
  • Location Completeness: Every premises row must have a street, city, state, and ZIP.
  • Loss Reconciliation: Prior loss amounts must be numeric and dated.

What Extraction Does Not Do

To maintain compliance and sound risk governance, understand where automated extraction stops:

  • It does not issue or alter certificates: It reads incoming documents; it does not generate binder certificates or modify coverage terms.
  • It does not verify carrier solvency: It extracts the insurer name and NAIC code as printed, but does not assess A.M. Best ratings.
  • It does not replace coverage verification: An ACORD certificate is legally informational and does not alter the underlying policy. Critical risks should be cross-referenced with policy declarations.

How to Measure ROI

Calculate your team's baseline manual data entry expense:

Annual Cost = Forms per Year × (Minutes per Form ÷ 60) × Loaded Hourly Rate

For example, processing 1,000 certificates a month (12,000 annually) at 15 minutes per form represents 3,000 hours. At a $35/hour loaded labor cost, that equals $105,000 annually in repetitive keying. Automating 80% to 90% of this intake frees up operational capacity for client onboarding and underwriting analysis.

Automate ACORD Form Data Extraction with AI

Extract fields, limits, checkboxes, and certificate holder details from ACORD 25, 28, and 125 forms directly into JSON, Excel, or your AMS. 20 free pages included.

Start Extracting ACORD Forms Free →

Frequently Asked Questions

What is ACORD form data extraction?

It is the automated reading of completed ACORD forms into structured fields such as policy numbers, limits, dates, and checkbox answers, delivered as JSON, CSV, or Excel.

Can AI read checkboxes on ACORD forms?

Yes. A good parser classifies each box as checked, unchecked, or uncertain and links it to its label. Uncertain boxes are flagged for human review.

Can it extract the certificate holder from an ACORD 25?

Yes, including name and address, along with insured details, policy numbers, effective/expiration dates, and limits.

Can it extract the workers' compensation policy number?

Yes. Workers' comp policy numbers, dates, statutory coverage, and employer's liability limits are standard ACORD 25 fields.

What is the difference between ACORD 25 and ACORD 28?

The ACORD 25 is a certificate of liability insurance. The ACORD 28 is evidence of commercial property insurance, typically used where a lender or mortgagee needs proof of property coverage.

Can it read fillable ACORD PDFs?

Yes. Fillable PDFs carry form fields that can be read directly, while scanned or flattened faxes and photos are processed through AI vision OCR.

Does it work on scanned or faxed ACORD forms?

Yes. Vision-based models read rendered page images. Accuracy depends on resolution, and low-confidence fields are flagged for reviewer validation.

Do I need a template for each form edition?

No. Modern layout-aware, context-based AI extraction identifies fields by meaning and relative alignment rather than fixed pixel coordinates.

Can it create or edit ACORD certificates?

No. It extracts data from completed forms. It does not generate, issue, or alter certificates.

Can it tell me whether the carrier is financially sound?

No. It extracts the insurer name and NAIC number as printed on the form, but does not assess financial strength or solvency ratings.

ShareXLinkedIn
Mike Smith

Mike Smith

Product Growth Lead at PerfectParser

Mike Smith leads product growth at PerfectParser, where he builds AI-driven data extraction workflows for complex business documents. Drawing on years of experience developing advanced AI systems, he is dedicated to helping finance and operations teams replace manual data entry with high-accuracy, intelligent automation.

Related Resources & Solutions

Put the guide into practice

Ready to automate your documents?

Join teams saving hundreds of hours on data entry. Sign up now and start with 20 free pages — no credit card required.

Start Free Trial

Free trial · No credit card required