A completeness check is two jobs: is everything present, and does it agree with itself? Software is good at both and cannot be accountable for either.
Key takeaways
You check a submission is complete by turning the insurer's requirements into a structured list, extracting the corresponding fields from the documents you are about to send, and having software report three things: what is missing, what is blank, and what disagrees with something else in the pack. A broker then decides what to do about each.
The gain is not that a machine reads the application. It is that a machine reads it the same way every time, at eleven at night, on the twentieth renewal that week — which is precisely when a human check degrades.
A commercial submission is rarely one document. It is an application, prior loss experience, a statement of values or schedule, supplementary questionnaires for the classes involved, financial information for some risks, and whatever the insurer's appetite guide adds for the occupancy.
"Complete" therefore means complete for this insurer, this class, this renewal date — not complete in the abstract. That is why the useful build is not a generic checklist but a per-insurer requirement list your brokerage maintains, against which extraction is compared. The maintenance of that list is the real asset; the AI is the part that is easy to buy.
The submissions that come back with questions are usually not the ones missing a page. They are the ones where the named insured on the application does not match the name on the schedule, the values on the statement of values do not sum to the total declared, the construction year contradicts the appraisal, or the loss history covers four years where the insurer asked for five.
Those are exactly the comparisons software does well: normalise, cross-reference, and report the mismatch with both sources cited. A person is poor at this at volume and excellent at deciding which mismatches matter, which is the right division of labour.
A brokerage handles a multi-location retail renewal. The pack is an application, a 14-location statement of values, five years of loss runs, and a fire protection questionnaire.
The checker extracts the location count and total insured value from the statement of values and compares them to the figures on the application — they differ, because a location was added in March and the application was drafted from last year's numbers. It compares the policy period on the loss runs to the requested period and finds one year short. It reads the fire protection questionnaire and flags two unanswered items.
None of those is a judgment call. All three would have come back from the underwriter as a query, two days later, with the renewal date closer. The broker resolves them before the submission goes out and records what was changed.
Completeness is not merely a service standard; the consequences of an omission run to whether the policy responds. In British Columbia, section 17 of the Insurance Act provides that a contract is not rendered void or voidable by reason of a misrepresentation or a failure to disclose in the application unless the misrepresentation or failure to disclose is material to the contract, and that materiality is a question of fact.
Read that carefully, because it cuts both ways. It means an immaterial slip does not hand an insurer a defence. It also means the omissions that are material are precisely the ones a completeness check exists to catch. Provincial insurance legislation differs, so confirm the position in the province where the risk sits rather than transplanting one province's rule.
The downstream disputes are worth understanding before you design the check: our sister firm covers misrepresentation on an application and voiding a policy, whether a business's insurance covers what the owner thinks it covers, and an insurer's duty to defend.
The licensee is. In British Columbia, the Insurance Council of British Columbia states that its Council Rules establish licence conditions and requirements in addition to those in the provincial Financial Institutions Act, and that Council Rule 7(8) requires all licensees to comply with the Insurance Council's Code of Conduct, which defines standards of conduct for the practice of the business of insurance. Other provinces have their own regulator, and you should be reading your own.
No completeness tool changes that. If a submission goes out wrong, the claim that follows is against the brokerage — see suing an insurance broker in negligence. Which is the argument for the tool, not against it: a documented, repeatable pre-submission check is evidence of a system, and a system is what a standard-of-care argument turns on.
Three design rules keep this useful. Report, never send: the checker produces a list, and a licensed person clears it. Cite the source: every flag names the document and page it came from, or the broker spends longer verifying flags than checking manually. And log the clearance: who cleared which flag, when, and on what basis — that log is the artefact that is worth having two years later.
Submissions contain personal information about individuals as well as business data, so the Office of the Privacy Commissioner's principles for generative AI apply, and limiting collection matters: send the pages the check needs.
Can it submit to the insurer automatically?
It should not. Submission is an act by a licensed person that carries representations about the risk. Automate the assembly and the checking; keep the send button human.
Does this work for personal lines?
The mechanics do, but the payoff is smaller because the packs are shorter and more standard. The return is highest where a submission has many documents and internal totals to reconcile, which usually means commercial.
What about certificates and evidence of insurance?
Same mechanics, different list: the requirement usually comes from a contract rather than an underwriter. Whether a certificate actually reflects the coverage is a separate question — see how a certificate of insurance can shift out of date.
A 30-minute call is enough to tell you whether AI pays for itself here.
