Most compliance reviews stop at the certificate. Ours starts where the real risk lives, in the endorsement language itself. Here's how BCS reviews endorsements, and how that expertise is training the next generation of our AI.
A certificate of insurance tells you a policy exists. It doesn't tell you whether the coverage will actually respond when a claim hits. That answer lives in the endorsements, the dense, carrier-specific forms attached to the policy that grant, restrict, or quietly gut the protections your contracts require.
Anyone can check a box that says "Additional Insured endorsement received." The hard part, the part that determines whether your risk transfer actually holds up, is reading what that endorsement says. Does it require privity of contract? Does it exclude completed operations? Does it contain "sole negligence" language that a court could use to deny coverage entirely?
At BCS, we've spent years answering those questions at scale. The result is the BCS Endorsement Intelligence Matrix: a proprietary library of 3,700 reviewed endorsements, each analyzed against 18 distinct data points, spanning more than 5,826 requirement-type determinations across seven lines of coverage. It's the deepest endorsement knowledge base in the certificate of insurance tracking software industry, and it's the human-verified intelligence layer we're now using to train BCS's AI review capabilities.
What 3,700 Reviewed Endorsements Looks Like
Every endorsement in the matrix has been reviewed by BCS insurance professionals and classified by policy line and the requirement type it evidences: Additional Insured, Waiver of Subrogation, Primary & Noncontributory, or Notice of Cancellation. Because a single form can address multiple requirements, those 3,700 endorsements map to more than 5,826 individual determinations.
The distribution mirrors where risk transfer actually happens in construction and real estate:
Top requirement-type determinations in the BCS Endorsement Intelligence Matrix. CGL Additional Insured forms dominate, and they're also where the most dangerous limiting language hides.
Full coverage across seven policy lines
| Policy Line | Requirement Types Covered | Matrix Entries |
|---|---|---|
| Commercial General Liability | AI (Ongoing Ops), AI (Completed Ops), AI, P&NC, WoS, NoC | 5,122 |
| Auto Liability | AI, WoS, P&NC, NoC | 601 |
| Workers' Compensation | WoS, NoC, AI (Completed Ops) | 76 |
| Excess / Umbrella Liability | P&NC, WoS, NoC | 12 |
| Pollution Legal Liability | WoS, AI, P&NC | 7 |
| Professional Liability | WoS, AI, P&NC | 4 |
| Contractors Pollution Liability | WoS, AI, P&NC | 4 |
| Total | 3,700 unique endorsements | 5,826 |
AI = Additional Insured • P&NC = Primary & Noncontributory • WoS = Waiver of Subrogation • NoC = Notice of Cancellation
The 18 Data Points: How BCS Reads an Endorsement
Cataloging endorsements is table stakes. What sets the BCS review apart is the depth of analysis applied to each one. Every endorsement in the matrix is evaluated against 18 data points, a structured checklist built from years of seeing exactly where coverage fails in practice. They fall into three groups.
1. Coverage Scope: what the endorsement actually extends
2. Contractual Triggers: what has to be true for coverage to apply
3. Limiting & Red-Flag Language: where coverage quietly disappears
Why Title-Level Review Isn't Review
Consider a scenario our reviewers see constantly. A subcontractor submits an Additional Insured endorsement. The form title looks right. The checkbox on the COI is ticked. A title-level review marks it compliant and moves on.
A BCS review reads the form. Buried in the schedule language: Additional Insured status applies only "to the extent" of the named insured's negligence, coverage requires privity of contract with the named insured, and completed operations are silently absent. Three of our 18 data points just flipped that endorsement from compliant to a documented gap, before the project started, while there's still leverage to fix it. That's the difference between tracking documents and transferring risk.
A Worked Example: When a "Scheduled" Form Hides Blanket Language
Here is a review straight from our own system. A subcontractor submits a CG 20 10 (Ed. 12/19), the standard scheduled endorsement for Additional Insured, ongoing operations. This account's requirements are set to look for a schedule that names the specific parties who must be added as additional insureds, in this case Sales Demo Inc. and Business Credentialing Services Inc.
By form type, this is not a blanket endorsement, so a checkbox review, or a tool that only classifies the form, would pass it. The BCS matrix reads the actual language on the schedule instead, and here that language says: "Any person or organization where required by written contract to be added as an additional insured, provided the 'bodily injury', 'property damage', or 'personal and advertising injury' occurs subsequent to the execution of the contract or agreement."
That is blanket language. Even though the underlying form is a scheduled form, the text inside it extends coverage to anyone a written contract names and never actually lists the required parties. The matrix flags the result as an Unacceptable Form with missing entities, because Sales Demo Inc. and Business Credentialing Services Inc. are not specifically scheduled.
3,700 Endorsements Are Now Training Our AI
Here's where depth compounds. Every one of those 3,700 endorsements, with its 18 data-point analysis, is a human-verified training example. That corpus is the intelligence core of the AI review capability now in beta at BCS.
Generic AI can read an insurance document. It can't tell you that "caused by" wording is unacceptable where "arising out of" is required, or that a privity requirement leaves a project owner exposed, unless it has been taught by people who've made those calls thousands of times. Our AI isn't learning from the open internet. It's learning from the largest structured, expert-labeled endorsement dataset in the industry, built determination by determination by BCS reviewers.
The result is a system designed to deliver expert-grade endorsement analysis at machine speed, with the judgment encoded in, not bolted on.
What This Means for Your Compliance Program
If your current COI tracking process, or your current vendor, verifies endorsements by title and checkbox, your compliance rate is measuring paperwork, not protection. The questions worth asking of any review process: How many endorsement forms does it recognize? How many data points does it evaluate on each one? And can it show you, in the actual policy language, why a document passed or failed?
At BCS, the answers are 3,700, 18, and yes.
See the Matrix in Action
Get a walkthrough of how BCS reviews your certificates and endorsements against your actual contract requirements, and a preview of our AI review beta.
Talk to Our Team