The Best's Credit Rating Methodology has four building blocks. Balance Sheet Strength, Operating Performance, Business Profile, and Enterprise Risk Management. The first three are where boards spend most of their attention, because they are the ones rating reports tend to lead with and the ones that read most cleanly off audited financials. ERM is the fourth block, and it is the one almost no board has stress-tested for AI exposure.
That asymmetry matters because ERM is also the block most exposed to the qualitative override AM Best calls a Comprehensive Adjustment. The Comprehensive Adjustment is the lever in the methodology that allows the rating committee to move a carrier's rating across notches based on factors the quantitative model does not fully capture. AI model risk management lives almost entirely inside that qualitative space. A carrier whose three quantitative blocks look strong can still move across the A- threshold if the rating committee concludes that the ERM posture does not support the indicated rating.
We think this is the cleanest unmapped pathway from AI governance weakness to existential balance sheet harm. The mechanism is published, and the 2023 downgrade data is already pointing in the same direction. Boards should be able to recite it before the next rating review.
The Four Building Blocks and What ERM Actually Covers
The Best's Credit Rating Methodology frames the rating as a build from four assessments. Balance Sheet Strength carries the heaviest weight and anchors the analysis. Operating Performance, Business Profile, and Enterprise Risk Management each adjust the indicated rating up or down within published ranges. ERM in particular is scored across a five-point scale from Very Strong to Very Weak.
What AM Best evaluates inside ERM is broader than most carriers assume. The published criteria look at risk identification, the ERM framework itself, risk culture, governance and control structures, and the carrier's documented capacity to escalate when controls are breached. The analyst is asking whether the carrier knows what risks are on the books, whether the people who operate those risks own them in writing, and whether there is a documented path from incident to executive attention.
Model risk management sits inside that frame. A machine learning rating model is a risk-bearing instrument that can introduce loss into the book without an underwriter ever touching a file. AM Best's ERM evaluation already has the vocabulary to read it as such. The question for a carrier deploying AI in pricing, underwriting, or claims is whether the analyst, sitting in the next interactive review, can find the documentation that would let them defend a "Strong" or "Adequate" ERM assessment to their committee.
For most regional and specialty carriers we observe, the answer in 2026 is that the documentation does not exist in a form a rating analyst can use. Commit histories, model retraining notebooks, and Slack channels do not satisfy the framework. The analyst is looking for a governance system, written down, with named owners and operating evidence that the system runs.
The Comprehensive Adjustment Is Where the Move Actually Happens
The Comprehensive Adjustment is the methodology's mechanism for letting the rating committee move the indicated rating based on holistic judgment that the four quantitative blocks do not fully capture. Material weaknesses in ERM are explicitly contemplated as a basis for downward adjustment.
For a carrier sitting at the A-/B++ boundary, the Comprehensive Adjustment is the fulcrum. A rating committee that reads the carrier's ERM as "Weak" or "Very Weak," and concludes that the weakness is not adequately reflected in the quantitative blocks, has the methodological authority to apply a downward adjustment that moves the published rating across the threshold. The qualitative read is sufficient, on its own, to produce the move.
There is a concept worth naming clearly here. In the FedNat sequence we walked through separately, the rating downgrade was the opening event of a chain that ran through treaty trigger to insolvency in 89 days. The Comprehensive Adjustment is the methodological hinge that lets a rating committee deliver that opening event without waiting for a quantitative metric to catch up. It is faster than financial deterioration, because it leads it.
What the 2023 Data Is Already Telling Us
The agency's appetite for downward action has shifted. AM Best recorded 55 P&C downgrades in 2023, against 30 in 2022. Personal lines saw the sharpest movement, with downgrades roughly doubling from 18 to 39, according to AM Best's annual rating action review covered in trade press at the time. Upgrades stayed roughly flat over the same window.
The composition of those actions is as interesting as the count. Reading through the 2023 rating commentary that AM Best published alongside specific carrier actions, ERM is referenced more frequently and more directly than it was in the 2018 to 2020 period. Governance language that previously sat in qualitative footnotes now appears in headline rationale. The agency is doing two things at once. It is moving more carriers downward in absolute terms, and it is leaning on ERM as a stated basis for those moves.
A carrier whose ERM block has not been seriously evaluated for AI exposure should read that pattern as a forward signal. The framework the analyst will apply at the next interactive review is already written, the willingness to apply it adversely is already demonstrated, and the documentation that would defend an AI-driven business model under that framework is, for most carriers, not yet built.
The Downstream Chain Boards Should Map
Step one is the AM Best ERM assessment moving from "Adequate" to "Weak" or "Very Weak" because the analyst cannot find a defensible AI governance system. The Comprehensive Adjustment then pulls the published rating across the A- floor, even if Balance Sheet Strength and Operating Performance look fine on a static read.
Step two is the treaty rating-trigger clause activating. Standard catastrophe excess-of-loss wordings include rating triggers that grant the reinsurer the right to non-renew, cancel, or reprice if the cedent's interactive rating drops below a contractual threshold, commonly AM Best A- or S&P BBB+. We covered the wording exposure separately in our analysis of why the reinsurance treaty is where AI risk becomes existential. The cedent does not control the timing once the trigger fires.
The remaining steps run quickly. The cat layer becomes unavailable, partial, or repriced beyond what the carrier's surplus can absorb. The next cat loss attaches against a retention sized for a treaty layer that no longer exists. Receivership follows. Each link in this sequence is mechanically supported by existing law, treaty practice, or regulatory authority. The novel element is that the chain now opens with a qualitative read of governance rather than a quantitative deterioration of surplus.
This is the same chain FedNat ran in 2022. The trigger has shifted from a Demotech reserves read to an AM Best governance read. Everything downstream of the trigger continues to work the way it did in the prior decade.
What the Analyst Will Actually Ask
A carrier preparing for the next interactive AM Best review should expect the ERM portion of the meeting to include questions a 2018-vintage governance package was not built to answer. Several patterns are already visible in published rating reports for carriers that have been downgraded in the past eighteen months.
Ownership comes up first. The analyst will ask who owns the model, meaning the named risk owner with documented authority to pull it from production, not the data science team that built it. A name in response keeps the carrier in the conversation, while a department name leaves the question open.
Performance monitoring is the second area. The analyst is asking what specific drift thresholds trigger formal escalation and to whom, rather than whether monitoring exists in the abstract. A dashboard nobody is mandated to act on reads as a metric in the framework, not as a control.
Disclosure to the reinsurer is where the AI question intersects most directly with treaty exposure. The analyst will ask what the carrier communicated to its reinsurer at the last renewal about model changes during the treaty period. A retraining cadence that was never formally communicated leaves the carrier exposed to the disclosure warranty, and the ERM analyst is reading this with one eye on whether the reinsurer would consider themselves adequately advised. That read feeds directly into the qualitative judgment behind the Comprehensive Adjustment.
Incident response is the fourth area, and this is where governance documentation either holds together or falls apart. A carrier that cannot produce a documented escalation path, with named owners and timing expectations, has a gap the framework was specifically built to find.
None of these are exotic asks: they are standard ERM questions applied to a class of risk-bearing instrument that did not exist in the framework's original drafting. The question for the board is whether the documentation that would satisfy them currently exists in the building.
The CEO Question for the Next Board Meeting
The question worth putting on the next agenda is short. Could our AM Best analyst defend our AI model governance to their committee?
The honest answer for most regional and specialty carriers we observe is that the analyst could not, because the supporting documentation does not yet exist. Model change logs written for treaty underwriters, named risk owners with pull-from-production authority, drift escalation policies with response time commitments, and a treaty disclosure register tracking what was communicated at each renewal: these are the artifacts the framework is asking for. No state insurance department requires them today. The forcing function is the rating review, and the consequence of failing it runs through the Comprehensive Adjustment to the treaty trigger to surplus.
The asymmetry should make boards uncomfortable. Building governance documentation a rating analyst can defend takes months of work and a modest standing operating commitment, while arriving at the next interactive review without it costs the carrier the delta between the rating it expected and the rating the committee actually publishes, multiplied by what that delta does to the cat program at the following renewal.
We have written before that AI fines are survivable and reinsurance loss is not. The Comprehensive Adjustment is the methodological bridge between the two: the mechanism by which a governance weakness that would otherwise sit as a qualitative footnote becomes the opening event of an existential chain. Mapping the pathway now buys the carrier the runway to build documentation that would carry it through the next interactive review, and the carriers that wait are operating on a clock the rating committee controls without their input.
