Can AI Fill ACORD Forms? From Dec Pages to Completed Applications

Yes, AI can fill in ACORD forms from a client's existing policies, but only as a draft. Extraction from dec pages and loss runs is fast on clean files. Before submission the agent must check every limit, date, prior-carrier detail and loss figure against its source, and the client must review and sign what was filled.

The forms that agencies fill most often are the ACORD 125 (Commercial Insurance Application, Applicant Information Section), the ACORD 126 (Commercial General Liability Section), the ACORD 140 (Property Section) and the ACORD 25 (Certificate of Liability Insurance). The 126 and 140 are line sections designed to sit behind a 125, so an error on the 125 travels into both. The value of AI is in lifting data that already exists in a client's policy documents. The risk is the same as with any data entry: a wrong number on a signed application. What follows is a form-by-form guide to what extracts well, what the agent must check, and where the errors and omissions exposure sits.

Follow me on Instagram@sagnikteaches

Where the time goes, and what AI can take

A commercial application is mostly re-keying. The applicant's name and address, the prior carriers, the policy periods, limits, locations, class codes, payroll, sales, vehicle and building details already exist in expiring policies, schedules and loss runs. Staff copy them across by hand, form by form, then again into each carrier's portal. Typical steps where AI helps:

Connect on LinkedInSagnik Bhattacharya

• Reading the source documents (dec pages, schedules of locations, loss runs, prior applications) and pulling out named fields.
• Mapping those fields to the right boxes on the right form, using the form versions your agency system holds.
• Flagging gaps where a required field has no source.
• Preparing a review sheet showing each filled value next to the page it came from.

What it cannot do is decide what the client actually has, what they want covered, or whether an answer is true. That remains the agent's job, and the checks below are built around it. If your submission flow also involves insurer portals, the tutorial on AI submission intake and re-keying covers that side.

Subscribe on YouTube@sagnikteaches

What the tools are and what they claim

The market for this is mostly insurance-specific extraction platforms rather than general chatbots. Two whose own pages I read for this tutorial: Acord AI says it is built for agencies, MGAs, brokers and carriers, with separate modules for email intake, policy checking against source documents, extracting data into ACORD forms "for validation and review", quote comparison and certificates, and says it routes critical items to insurance professionals for review and approval. Cara AI describes enterprise AI for agencies, brokerages and wholesalers that automates daily manual tasks, client requests and new-sale and renewal workflows; its home page lists certificate generation but does not mention ACORD application forms, so ask what it fills before you book a demo for that job. Note that Acord AI is a vendor name, not ACORD, the standards body that publishes the forms, and its site states no link to ACORD. Check that distinction in any demo.

Do not overlook the system you already have. Agency management systems ship ACORD forms with their own field maps (Vertafore's AMS360 help, for example, publishes a field map for the ACORD 127), so the cheapest set-up may be an extraction step that feeds data into your existing forms rather than a tool that produces its own.

Prices are generally on request, so ask for them in writing, with what happens to your data, which form editions are supported, and whether the tool writes into your agency management system or produces a PDF you then upload. You can also use a general AI model for a smaller job, such as extracting fields from a dec page into a table, and then copy the table into your system. That approach costs less and lets you see every step, but it puts more of the checking and the data-handling decisions on you.

Because vendor claims on accuracy are marketing until you test them, use the test in the section below before you commit. The pre-submission checks apply whichever tool you pick.

ACORD 125: the commercial application

This is the form with the broadest data: applicant details, nature of business, premises, prior carriers and loss history, and the lines of coverage requested. It is also where the most expensive mistakes hide, because it carries the general answers that later forms rely on. One carrier's agent guide (Merchants Insurance Group's) calls it a required part of every commercial submission except workers compensation, and notes that it holds information not repeated on the other application forms, so a gap here is a gap everywhere.

What extracts well

Named insured, mailing address, policy numbers, effective and expiry dates, carriers, and premiums from dec pages. Prior-carrier history from the last three to five years of policies on file. Business description from a prior application.

What the agent must check

• Named insured exactly as it should appear. Models normalise "[Surname] & Sons Plumbing, LLC" into "[Surname] and Sons Plumbing LLC". A mismatch with the legal entity can create trouble at claim time.
• Nature of operations. The description must match what the client does now, not what a prior form said. If the plumbing firm began doing commercial work last year, an old residential-only description is a misrepresentation waiting to happen.
• Prior carrier and policy period. Dates in the wrong format, or the expiring policy listed as the one before it, are common extraction slips.
• Yes/no questions. Questions on bankruptcy, cancellations, prior declinations and uninsured losses need the client's answer. A tool will leave them blank or guess "No". Both are dangerous.

Example: a plumbing firm renewal

Say a four-person plumbing firm moves to a new carrier. The tool fills the ACORD 125 from the expiring policy and a prior application. The review sheet shows the business description as "residential plumbing repair", copied from three years ago. The owner has since taken on small commercial fit-outs. The agent updates the description and adds the subcontracted work, which changes the questions asked on the liability section. A pure extraction approach would have submitted a polished form describing the wrong risk.

ACORD 126: commercial general liability section

This section captures the liability exposure: classification, premises, sales or payroll as the exposure base, subcontracting, products and completed operations, and claims-made details where relevant. It is the form where numbers matter most to price. Two layout details trip extraction tools. The schedule of hazards holds only three locations or class codes, with an ACORD 211 Schedule of Hazards for more, and long remarks overflow to an ACORD 101 Additional Remarks Schedule. A tool that cannot spill onto those attachments silently drops the fourth location or truncates the remark. The same carrier guide says every question in the contractors section must be answered when the applicant is a contractor, and that both the applicant's and the producer's signatures are required.

What extracts well

Class descriptions and codes from the expiring policy, limits, deductibles, and the exposure basis if a prior application or audit is on file.

What the agent must check

• Exposure figures. Payroll or gross sales for the coming period, not the last one. The tool will happily copy last year's. Ask the client for a current estimate and write the date next to it.
• Subcontractor costs and insurance. Whether subs carry their own cover, and the cost of subcontracted work, which is often missing from a dec page.
• Claims-made versus occurrence. Retroactive dates and extended reporting details do not extract reliably and have serious consequences if wrong.
• Class code fit. A code carried over from a previous carrier may not match the carrier's own classification. Confirm it against the carrier's appetite and guide.

Example: a cleaning company's exposure base

Say a cleaning company's expiring policy shows annual receipts of $620,000. The tool carries that figure to the 126. The owner mentions on the phone that two large contracts started in January and she expects $780,000. The agent enters the new figure and notes the source in the file. Understating the base is the classic route to a premium audit and a dispute, so the extraction needs the human conversation to be right.

ACORD 140: property section

The property section describes buildings and contents: locations, construction type, year built, protection, occupancy, values, and the coverage requested. The data is detailed, and a lot of it is not on a dec page. Each 140 holds two locations, with coverage and rating recorded separately for each, so a five-location schedule of values needs three copies and a tool that maps rows to the right copy in the right order.

What extracts well

Addresses, building and contents limits, deductibles and mortgagee names from a schedule of locations. If the client has a statement of values, the tool can map it row by row.

What the agent must check

• Replacement cost versus the old limit. The limit on the expiring policy is not evidence of current value. Construction costs move, and a stale limit leaves the client underinsured.
• Construction, roof age and updates. These come from inspection reports, photos or the owner. A tool can read them from a document if one exists. If not, it should leave a flag, and you ask.
• Protective devices. Sprinklers, alarms and their certification or monitoring status should be confirmed, because they affect rating.
• Mortgagees and additional interests. Names and loan numbers must be exact, since a lender who is not properly named can dispute a claim payment.

Example: a roofing contractor's yard

A roofing company owns a small yard with a workshop and storage. The dec page shows a building limit set five years ago. The tool copies it and the roof year from the last application. The agent asks the client for the roof replacement date and a current estimate, learns that the building was extended last year, and updates the values and the construction description. The AI saved the typing of eight fields. The conversation fixed the three that mattered.

ACORD 25: certificate of liability insurance

The certificate is the form that goes out most often, and the one where AI automation is easiest to sell because requests are repetitive. It is also a document that third parties rely on, so errors reach beyond your client.

What extracts well

Policy numbers, carriers, effective and expiry dates, limits and coverage types from current policies, plus the certificate holder's name and address from the request.

What the agent must check

• Policies in force. A tool that reads an expired dec page issues a certificate for cover that has lapsed. Check the effective dates against today's date and the renewal status.
• Additional insured and endorsement wording. A request may ask for additional-insured status, waiver of subrogation or primary and non-contributory wording. The certificate must reflect endorsements that actually exist on the policy, and not describe cover the policy does not provide. ACORD's own certificates FAQ is blunt on the point: a certificate is not a policy and does not provide, amend, extend or alter its terms; only an endorsement, rider or amendment to the policy changes coverage. This is the point at which certificates go wrong in a way that matters.
• Certificate holder details. Name and address as on the contract or request.
• Description of operations. Keep it factual and tied to the contract reference.

Example: a request from a general contractor

An electrical contractor's client receives a request from a general contractor asking for additional-insured status on a primary and non-contributory basis, with a waiver of subrogation. The tool prepares the certificate and ticks the boxes because the request asked for them. The agent opens the policy and finds a blanket additional insured endorsement but no waiver of subrogation endorsement. The agent contacts the carrier to add it, and issues the certificate only when it can be truthfully stated. Acord AI's certificate module, for instance, says it validates policy details and prepares certificates for review and delivery. That is the right design, and the final sign-off belongs with the agent.

Loss runs: the document that fills three forms

Loss runs feed the prior-loss sections across the application set. They also vary more than dec pages in layout between carriers, which makes extraction harder.

• Valuation date. A loss run is a snapshot. Check its date and whether the carrier's cut-off matches your submission window.
• Open versus closed. The tool may count an open claim's reserve as paid. Reserves and payments are different columns.
• Claim count versus claim amounts. Many carriers want the count of claims and incurred totals separately. Add them yourself and compare against the carrier's total line.
• Zero-dollar claims. Incident-only records appear in some loss runs. Whether they belong in the application's loss history depends on the question asked.

Say a client has loss runs from two prior carriers over five years. The tool lists seven claims. The agent adds the incurred columns by hand, finds the extraction counted a $1,850 reserve as paid, and corrects it. A small slip like that changes the answer to a yes/no question about losses above a threshold.

Other forms in the same set: 127 and 130

Most commercial packages also include the ACORD 127 (business auto section) and the ACORD 130 (workers compensation application). Both follow the same pattern, so the checks carry over, with their own traps.

• Business auto. Vehicle schedules extract well from a dec page: year, make, VIN, garaging address. The agent must check that vehicles bought or sold since the last renewal are reflected, and that drivers listed are current employees with licence details confirmed by the client. A sold van still on the schedule is paid-for exposure. A new one missing from it is an uninsured vehicle.
• Workers compensation. Payroll by class code is the heart of it. A tool can lift last year's payroll from an audit, but the figure the carrier needs is a projection for the policy year, split by job role. For a cleaning company with office staff and field crews, putting everyone in one class is a mistake that shows at audit.

If your agency uses additional forms for a particular line, apply the same discipline: source for each field, client answer for each question, current figures for exposure.

A prompt for a general model, with a sample output

If you start with a general AI model on redacted files rather than a platform, ask for a table, not a finished form. The table is easier to check.

From the attached dec page, extract these fields into a table with
three columns: Field, Value, Source (page and line). Fields: named
insured, mailing address, policy number, carrier, effective date,
expiry date, each coverage with its limit and deductible, and every
endorsement listed. Do not infer anything. If a field is not on the
document, write NOT FOUND. Do not answer any application question.
Quote dates exactly as printed.

An illustrative result:

Field                 Value                         Source
Named insured         [Insured name]                p1, header
Policy number         [number]                      p1, top right
Effective date        01/15/2026                    p1, policy period
Expiry date           01/15/2027                    p1, policy period
Each occurrence       $1,000,000                    p2, limits table
General aggregate     $2,000,000                    p2, limits table
Waiver of subrogation NOT FOUND                     n/a

What you fix: check the date format against the page (a model can flip day and month), confirm that "NOT FOUND" really means the endorsement is absent and not just on a page the model did not get, and move the figures into your form yourself or via your agency system. The source column is the feature that makes checking quick.

Data handling in the agency

Dec pages, loss runs and applications hold personal and business information, and sometimes health or claims detail. Set three rules before anyone uploads a file. First, redact or use placeholders for anything the model does not need, such as tax identifiers. Second, use only tools your agency has approved, on a plan whose terms say what is retained and whether inputs train the vendor's models. Third, decide where drafts live: not in a personal chat history. Say a four-person agency gives each staff member the same short rule card and a named person to ask. That is enough to start, and it is far easier than undoing a habit later.

A pre-submission checklist for the agent

Use one list for every file, whichever form and tool. Tick every line, and keep the review sheet in the client file.

1. Source match. Every filled value has a source page, and the agent has seen it. Spot-check is not enough for money fields.
2. Dates. Policy periods, valuation dates, request dates and signature dates are consistent and current.
3. Money and limits. Totals add up; exposure figures are for the coming period and dated.
4. Names and addresses. Legal entity name, premises addresses, mortgagees and certificate holders match the source documents exactly.
5. Yes/no and narrative answers. Each answered by the client, not defaulted. The agent has asked, and the answer is in a note.
6. Form edition. The carrier or market accepts the version used, and it is a current ACORD edition. ACORD's FAQ warns that withdrawn editions are no longer kept up to date for regulatory requirements, and tells producers to ask their agency management system vendor for updates.
7. Blanks. Every empty required field is either filled or deliberately marked not applicable.
8. Client review and signature. The client has read the completed application, not a summary, before signing.
9. Retention. Source documents, review sheet and the signed application are stored together.

The errors and omissions question

An application is a statement the client signs, usually with an acknowledgement that the answers are true and complete. If a material fact is wrong, the carrier may rescind or deny a claim, and the client may look to the agency. AI changes how errors are made, not who is responsible for them. Four steps keep the exposure sensible.

• Keep the audit trail. Retain the source, the draft, the corrections and the signed copy. If a field is challenged later, you can show what was extracted, what was checked, and what the client confirmed.
• Do not let tools answer for the client. Any tool that auto-answers yes/no questions without a client input is a risk to switch off.
• Read your tool's terms. Look for liability limits, data retention and use of your data for training. Compare against your agency's own policies.
• Tell your E&O carrier or broker. Ask whether they have a view on AI in application preparation, and whether any condition or questionnaire applies. Do it before you roll the tool out agency-wide, and record the answer.

The tutorial on comparing policy wordings with AI raises the same theme of human sign-off on things clients rely on.

Test a tool on ten files before you trust it

A demo uses a clean file. Your own book includes scans, fax-quality PDFs and policies that were renewed three times with endorsements. A short test gives numbers.

1. Pick ten real files with a completed application already in the file. Include two difficult scans and one multi-location property schedule.
2. Run the tool, then compare every field to the completed version. Tally errors by type: wrong value, missing value, wrong form field, hallucinated value.
3. Time two things: tool processing and human review. The saving is the old time minus the sum of those.
4. Check what the tool does when it is unsure. Good tools flag; poor tools fill.
5. Decide a threshold in advance. For example: no hallucinated values in ten files, fewer than two wrong values per file on money fields, review time under 15 minutes per application.

Illustrative result: on ten files, the tool fills 85% of fields correctly, leaves 10% blank with flags, and gets 5% wrong, mostly dates and class codes. Review takes twelve minutes per file instead of forty-five of typing. That is a clear saving, with a review step doing real work. If the 5% included invented values with no flag, the answer would be different.

Where this fits in a small agency's week

Start with the narrowest job where the source is clean and the stakes are visible: certificates for repeat clients, where policies are current and the requests are routine, with the agent signing off each one. Move to renewal applications for stable accounts next, because prior data is rich and little has changed. Keep new-business applications for complex risks manual or double-checked until your test numbers justify more. Agencies that want a sense of cost can read what AI costs a small brokerage in year one, and renewal workflows are covered in automating insurance renewals with AI.

If you would rather map this against your own forms and systems before spending on a tool, that is what my AI implementation consultation covers. Whatever you choose, the rule stays the same: AI fills the boxes, and the agent owns what the box says.

Questions agencies ask before automating ACORD forms

Who is responsible if AI fills a field wrongly?

The agency that submits the application. The client signs it and the agent presents it, so a wrong prior-carrier detail or an understated figure can lead to a coverage dispute or an errors and omissions claim. Tool vendors typically limit their own liability in their terms, so read them before relying on output.

Can I upload a client's dec page to a general chatbot?

Only if your agency's data policy and the tool's terms allow it. Dec pages and loss runs hold personal and commercial data. A business plan with no training on your inputs, or an insurance-specific tool with its own security controls, is the safer route. Ask what is stored and for how long.

Will AI know which edition of an ACORD form to use?

No. ACORD revises its forms and says withdrawn editions are not kept up to date for regulatory requirements. Use the current versions in your agency management system or the carrier's portal, ask your system vendor about updates, and let the tool map data to those fields rather than generating its own layout.

How do I test an extraction tool before buying?

Run ten real files you have already completed by hand, including one with a messy scan. Compare every field, count the errors by type, and note how long review took. Ask each vendor to show where the tool flags low-confidence fields.

Further reads

• Where AI Saves Time in a Small Insurance Brokerage — Where else AI saves time in a small brokerage.
• Can AI Answer Calls and Qualify Leads for an Insurance Agency? — Answer calls and qualify leads for an agency.
• How Insurance Brokers Use AI to Handle Claims Enquiries — Handle claims enquiries with AI, with limits.
• Compliance Calendar: Track Licences, Insurance and Certificates — Track licences, insurance and certificates on one calendar.
• Contractor Insurance Renewal: Gather the Figures Before Your Broker Calls — The figures contractors should gather before renewal.
• Best AI Tools for Insurance Agents: Seven for Quotes, Renewals and Calls — Seven AI tools for independent insurance agencies, with costs, test scripts, a six-person agency sum and the E&O checks before you buy.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Acord AI and Cara AI home pages; ACORD Certificates of Insurance FAQ; Merchants Insurance Group agent guides to the ACORD 125, 126, 130 and 140; Vertafore AMS360 help, ACORD 127 field map (all read 4 November 2026). Check the current edition of each form in your agency system.