The reserve roundtable is where every assumption in a claim file gets challenged on the record. The Atlanta slip-and-fall scenario - claimant alleges a $250,000 disc surgery, defense counsel reports the venue (Fulton County, GA) and the panel of plaintiff's experts suggest a likely jury verdict range of $125,000 to $325,000, paid-to-date is $42,000 ALAE on a $30,000 case reserve set 14 months ago, defense is requesting $50,000 mediation authority, and CCC IX's litigation-prediction model rates the file "moderate-high severity" with a 68% probability of plaintiff verdict if tried - is the canonical reserve-adequacy moment that forces every adjuster to articulate reserve reasoning on paper before the field manager, the chief claims officer, and the appointed actuary review it. AI does the structural work: pulling the loss-development data, benchmarking severity, threading the defense-counsel input through the case narrative, and producing a reserve recommendation memo with indemnity, ALAE, and case-reserve components that hold up under cross-examination. The adjuster owns the judgment call. The reserve memo is the artifact every other workflow - reinsurance reporting on Schedule P, rate adequacy in the next SERFF filing, the bad-faith defense file three years from now - depends on. This lesson is the AI-assisted reserve-recommendation workflow that survives the roundtable.

The Atlanta Slip-and-Fall on the Table

The file is 14 months old. The claimant slipped on a wet floor in a national retail chain's Atlanta store, alleges L4-L5 herniated disc with radiating left-leg pain, has had two epidural injections and one MRI showing the herniation, has been recommended for a $250,000 multi-level fusion by an orthopedic surgeon her plaintiff's-firm panel routinely refers to. The original FNOL set the case reserve at $30,000 indemnity plus $5,000 ALAE based on a soft-tissue assumption. The ALAE has now spent to $42,000 (defense counsel rates plus the MRI review by the carrier's medical consultant), the case has been in litigation for 9 months, and defense has filed a $50,000 mediation-authority request supported by a memo arguing the surgery recommendation is not yet medically necessary (claimant has not completed conservative care to exhaustion under Georgia case law) but plaintiff's counsel has demanded $325,000 in the latest demand letter. The venue, Fulton County, GA, sits in the upper third of Georgia jury venues on premises-liability verdicts.

The CCC IX litigation-prediction model (loaded with Fulton County premises-liability comparables, plaintiff's-firm verdict history, defense-firm verdict history, claimant age and occupation, and the surgery-recommendation flag) returns: 68% probability of plaintiff verdict if tried, expected verdict range $135,000-$340,000 with a median of $215,000, expected defense costs through trial $85,000-$110,000 additional ALAE. The reserve question on the table at the next roundtable: is the $30,000 case reserve adequate, redundant, or grossly deficient - and what is the recommended re-reserve including indemnity, ALAE-to-resolution, and any allocated case-reserve components for separate exposures (e.g., a Medicare set-aside if the claimant is over 62)?

The AI Pulls the Development Data

Before the adjuster touches the recommendation, the AI workflow pulls three categories of data the reserve discussion needs on the table. First, the carrier's internal loss-development pattern on premises-liability slip-and-fall in Fulton County: average paid-to-incurred at 14 months age (call it 22% based on the Schedule P pattern), average claim closure age (call it 31 months for litigated files), and average ALAE-to-indemnity ratio at resolution (typically 0.4-0.6 for litigated premises in this venue). Second, comparable case data: the carrier's prior cases with similar fact patterns (L4-L5 herniation, surgery recommendation pending, Fulton County venue, plaintiff's firm matched) and how those resolved - paid-to-resolution, indemnity-to-ALAE split, and case-reserve adequacy at the 14-month mark for the resolved cases. Third, defense counsel's input as text: the mediation memo, the prior status reports, the assessment of liability (contested? clear?), the assessment of damages (genuine? amplified?), and the assessment of trial risk (manageable? high?).

The model produces a structured pre-roundtable brief in 8-12 minutes that the adjuster reviews before walking into the discussion. The brief has six sections: case facts (one-paragraph synthesis), liability posture (defense counsel's view plus jurisdiction-specific case-law citations), damages exposure (medical specials to date, lien posture, projected future medicals, projected wage-loss component), comparable-case data (three to five prior cases with disposition and ALAE split), CCC IX prediction summary (verdict probability, range, expected ALAE), and a draft recommended reserve range with indemnity, ALAE-to-resolution, and the qualitative defense of the range. The adjuster does not adopt the brief; the adjuster reads it, tests it against the file, and forms a judgment that the brief either supports or fails to support.

The Three-Component Reserve Recommendation

The reserve recommendation has three components and each component needs its own defense. The indemnity component is the projected payment to the claimant - settlement value (most cases) or expected verdict less defense-cost recovery (rare cases). The ALAE component is the projected defense cost from now to resolution - outside counsel rates × projected hours, expert-witness costs, mediation costs, deposition transcripts, jury consultants if applicable, court costs if tried. The case-reserve component captures any unusual exposures not in indemnity or ALAE - a Medicare set-aside for a Medicare-eligible claimant (CMS approval timeline 9-18 months adds carrying cost), a child-claimant minor's compromise (court approval and structured-settlement administration), a confidentiality-clause administration cost, or a structured-annuity purchase cost.

For the Atlanta case, the AI-drafted recommendation reads: indemnity reserve $185,000 (CCC IX median verdict $215,000 discounted for settlement compromise and held below the high end of the defense-counsel range $325,000 to reflect mediation likelihood of resolution); ALAE-to-resolution $75,000 (CCC IX-projected defense cost through likely mediation plus a 20% buffer for unexpected motion practice); case-reserve component $5,000 (Medicare set-aside processing because claimant turned 65 last quarter); total recommended re-reserve $265,000 from the current $42,000 ALAE-spent plus $30,000 indemnity case reserve. The adjuster's defensible recommendation, written in their own voice with the AI brief as input, takes the form of a 1.5-2 page memo titled "Reserve Re-Recommendation - File [X] - [Date]" that the field manager, chief claims officer, and appointed actuary will see.

Surviving the Field Manager Challenge

The field manager's first question is always the same: "Why are we adding $235,000 to this file 14 months in?" The defensible answer cannot be "the AI recommended it." The defensible answer is built on three legs: the facts have evolved (surgery recommendation now in the file changes the medical-specials base from $25K to a projected $275K), the venue and panel realities are unavoidable (Fulton County premises with a known plaintiff's firm and a real surgery recommendation does not settle at $30K), and the defense counsel's mediation request itself signals defense's read of exposure (defense counsel does not request $50K mediation authority on a $30K-reserved case unless the file is materially under-reserved). The adjuster's job is to walk the field manager through the three legs in plain language, citing the file documents - the orthopedic surgeon's recommendation letter dated [X], the defense counsel's mediation memo dated [X], the plaintiff's demand letter dated [X] - with the AI-generated comparable-case data and CCC IX output as supporting evidence rather than the source of the recommendation.

The field manager will push on each leg. On medical specials: "Has the claimant exhausted conservative care under Georgia law? If not, defense's argument that surgery is premature has weight, and the reserve should reflect the settlement range without surgery rather than with." On venue: "What's our most recent verdict in Fulton County on a comparable disc-injury premises case?" On defense counsel input: "Is the mediation authority request driven by exposure or by counsel's risk-aversion?" Each push has an answer in the file documents and in the comparable-case data; the adjuster's job at the roundtable is to surface the documents and answer the push with file evidence rather than model output.

Surviving the Chief Claims Officer and Appointed Actuary Challenge

The chief claims officer's lens is portfolio impact and reserve discipline. The reserve move from $30K to $265K is a $235K hit to current-period reserves; on a $400M annual loss reserve book, a single file's $235K move is not material - on a $40M reserve book at a regional carrier or MGA, it is. The CCO will ask: "Is this an isolated re-reserve or a pattern? Are we seeing premises-liability severity climb across the book in Fulton County and similar venues, and is the reserving on the rest of the book consistent with this one?" The AI workflow produces a portfolio-view companion exhibit on demand: all open premises-liability files in Fulton County with similar surgery recommendations or known plaintiff's firms, their current reserves, their ages, and a flag on which ones may warrant a similar review. The adjuster does not own the portfolio answer (that is the CCO's territory), but the adjuster's file is the test case the CCO will reason from.

The appointed actuary's lens is reserve adequacy at the segment and total-line level under ASOP 36 (Statements of Actuarial Opinion on Property/Casualty Loss and Loss Adjustment Expense Reserves) and ASOP 43 (Property/Casualty Unpaid Claim Estimates). The actuary will ask: "Does this file's reserve change suggest the IBNR pick on the premises-liability segment in the South region is light? Does the loss-development factor (LDF) at 14 months on litigated premises in the segment need to be revisited? Is the reserve change in line with the trend the model is showing across the book, or is this file an outlier that requires segment-specific case-reserve discipline?" The AI workflow flags the file's place in the segment distribution (e.g., this file's incurred-at-14-months sits at the 78th percentile of comparable files in the segment); the actuary uses the flag as one data point in the broader segment review under ASOP 36/43.

The Memo That Survives the Deposition Three Years Later

The reserve memo is not just a roundtable artifact. It is the document plaintiff's counsel will subpoena three years later if a bad-faith claim follows an adverse verdict. The memo must establish that the carrier's reserve decision was made in good faith, with reasonable diligence, on the basis of the facts known and the analysis applied. The structure that withstands deposition has six elements: (1) facts as known at the date of the recommendation, with citations to specific file documents; (2) liability assessment with venue-specific case-law citations; (3) damages assessment with medical-specials base, lien posture, and future-medicals projection methodology; (4) comparable-case data with file references; (5) CCC IX or other model output documented as one input among many, not the recommendation; (6) the recommended reserve range with the adjuster's reasoning in the adjuster's voice, signed and dated.

The model-version stamp goes in the memo footer or in the file note that accompanies the memo: "Pre-roundtable brief generated by [model name and version], prompt template [X] at [archive location], input documents [list], reviewed and adopted with modifications by [Adjuster Name] on [Date]." The footnote is not legal cover; it is the chain-of-custody evidence that the adjuster used the model as a tool and not as a decision-maker. ASOP 41 and NAIC Model Bulletin §4 both rest on that distinction. A plaintiff's lawyer arguing bad-faith reserving on the adjuster's adoption of the AI recommendation without independent judgment loses if the file shows the adjuster reviewed the brief, tested it against the file, and produced an independent recommendation. The plaintiff's lawyer wins if the file shows the adjuster pasted the model output into the memo and signed.

When the Model Is Wrong and the Adjuster Must Override

The CCC IX prediction is one data point. The adjuster's experience on Fulton County premises files with this specific plaintiff's firm, this specific defense counsel, and the carrier's adjuster-network knowledge of the orthopedic surgeon's verdict history is the override. There are three common scenarios where the adjuster's judgment must override the AI output. First, when the model lacks recent comparable data (CCC IX's data is most powerful on high-frequency case types; on a thinly-traded plaintiff's firm or an unusual venue mix, the model's confidence is overstated and the adjuster's experience is more reliable). Second, when the defense counsel's read of exposure diverges materially from the model (defense counsel sees the deposition transcripts and the witness demeanor; the model does not). Third, when there is a settlement signal the model cannot price (claimant just lost a spouse, plaintiff's firm has a trial calendar conflict in 60 days, defense counsel has a prior relationship with the mediator that has produced favorable resolutions on similar files).

Each override goes in the memo with a documented rationale. "CCC IX model returns expected verdict range $135K-$340K; experienced override to recommend settlement reserve of $175K because [defense counsel reports witness credibility issues on the medical-necessity question that the model cannot factor], [plaintiff's counsel has a known pattern of accepting settlements at 65-75% of demand on cases with credibility issues], and [the mediator scheduled has produced 78% settlement closure on similar files in the prior 24 months]." The override is not a contradiction of the model; it is the adjuster's judgment doing the work the model cannot do, with the documentation that survives the deposition.

What the Adjuster and the Actuary Each Own

The adjuster owns the case-reserve recommendation. The appointed actuary owns the segment and total-line reserve opinion. The two roles do not overlap; they reinforce. The adjuster's memo is one data point in the segment review; the actuary's segment review informs the case-reserve culture (LDFs, average claim closure age, ALAE-to-indemnity ratios) the adjuster operates inside. AI assists both - the adjuster gets the pre-roundtable brief and the comparable-case data; the actuary gets the portfolio-view companion exhibits and the segment distribution flags. Neither role is replaced; both roles gain analytical depth and lose data-gathering hours.

The chief claims officer reads both. The CCO does not write reserve memos and does not certify segment reserves; the CCO sets the discipline and the escalation triggers. AI-assisted reserve work raises the CCO's signal quality - every adjuster's memo carries comparable-case data and model output documented as inputs, every segment review carries portfolio flags from individual file reviews, and the consistency of discipline across adjusters becomes measurable. The 88% of mid-market carriers that do not yet run this workflow leave portfolio-level reserve discipline to spreadsheet rollups and quarterly reviews; the 12% that do are seeing case-reserve accuracy at the 12-month mark improve 15-25% against the prior-period baseline.

Key Takeaways

• The reserve recommendation has three components - indemnity (projected payment), ALAE (projected defense cost to resolution), case-reserve (Medicare set-aside, minor's compromise, structured-settlement admin) - and each component needs its own defense at the roundtable. For the Atlanta slip-and-fall: $185K indemnity, $75K ALAE, $5K case-reserve, total $265K from the current $30K.
• The AI pulls three categories of data before the adjuster touches the recommendation: carrier's internal loss-development pattern (paid-to-incurred at age, closure age, ALAE/indemnity ratio), comparable cases (three to five prior files with disposition), and defense-counsel input as text (mediation memo, status reports, liability/damages/trial-risk assessments). The structured 8-12 minute pre-roundtable brief replaces 2-4 hours of adjuster prep.
• The field manager's "why $235K added 14 months in?" challenge has a three-leg answer: facts have evolved (surgery recommendation now in file), venue and panel realities (Fulton County with known plaintiff's firm), defense counsel's mediation request itself signals exposure. Each leg is defended from file documents, not model output.
• The chief claims officer's lens is portfolio impact and reserve discipline; the AI produces a portfolio-view companion exhibit on demand (all open premises files in Fulton County with similar surgery recommendations, their reserves, their ages). The adjuster's file is the test case the CCO reasons from on the rest of the book.
• The appointed actuary's lens is segment and total-line reserve adequacy under ASOP 36 and ASOP 43; the AI flags the file's place in the segment distribution (this file's incurred-at-14-months sits at the 78th percentile of comparable files). The flag is one data point in the actuary's broader segment review.
• The memo that survives a bad-faith deposition three years later has six elements: facts at recommendation date with file citations, liability with venue case-law citations, damages with methodology, comparable cases with file references, CCC IX (or other model) output documented as one input, and the adjuster's reasoning in the adjuster's voice signed and dated. The model-version stamp goes in the footer.
• Three common adjuster-override scenarios: thin comparable-data in the model, defense counsel sees what the model cannot (witness demeanor, deposition transcripts), settlement signals the model cannot price (claimant spouse death, plaintiff's counsel calendar conflict, mediator history). Each override goes in the memo with documented rationale.
• NAIC Model Bulletin §4 and ASOP 41 both rest on the distinction between adjuster as decision-maker and AI as tool. The adjuster who reviews the brief, tests it against the file, and produces an independent recommendation defeats a bad-faith claim on AI use. The adjuster who pastes model output and signs loses.
• The 12% of mid-market carriers running this workflow see case-reserve accuracy at the 12-month mark improve 15-25% against prior-period baseline. The 88% who do not run it leave portfolio-level reserve discipline to spreadsheet rollups and quarterly reviews - and pay the IBNR development surprise quarter after quarter.