← back to rankingAIG · American International Group, Inc.
Insurance - Diversified · mkt cap $39.1B · calls: Q1 FY2026 vs Q4 FY2025
60.0 conviction · conf-adj 60
conf 2/10 partial
enthusiasm:27.0 · trend:8 · quantifies:0 · impact:0 · under_radar:5 · credibility:12 · business_impact:4 · disruption:0 · commitment:0 · confirmation:4
Enthusiasm latest 9 / prev 8 (rising)
AIG’s AI thesis centers on underwriting and claims productivity via AIG Assist, an expanding Palantir ontology, and a shift from gen AI to multi-agent orchestration with Anthropic—management ties this to faster quoting, higher bind rates, and handling submission overflow (especially Lexington) without proportional headcount. Credibility is supported by repeated, line-specific productivity metrics in both calls, but P&L impact remains mostly implied: no AI-attributed revenue, margin, or expense dollars yet, with clearer ROI disclosure deferred to 2027–28. Enthusiasm and technical ambition rose quarter-over-quarter (agentic beta, global orchestration narrative) while human oversight and broker collaboration remain explicit guardrails.
GROUNDED NEXT-FY IMPACT vs CONSENSUS
Grounded on actual base — revenue $26.8B · net income $3.1B · net margin 11.6% · diluted EPS 5.43
These are next-fiscal-year annual uplift estimates, not next-quarter numbers.
Aggregate next-FY est. rev uplift: % · next-FY EPS uplift: % · vs analysts: unclear · priced in: medium (model's call-read: high; verdict above is the hard-data one used for ranking) · confidence: 2/10
| Claim | Figure | Arithmetic | Next-FY Rev % | Next-FY EPS % |
|---|
30% more submissions quoted (Lexington MM property / AIG Assist)
productivity · soft | 30% improvement | Throughput metric only. No disclosed Lexington MM property premium or % of $26,774M revenue → cannot compute rev_uplift_pct=100×Δ$/26,774M. EPS path needs undisclosed opex/FTE base. Illustrative bound: ~5% NWP lift on a hypothetical $1B line ≈$50M=0.19% of revenue — immaterial. | | |
55% reduction in time to quote (Lexington MM property underwriters)
productivity · soft | 55% reduction | Capacity/productivity claim, not a $ saving. Theoretical capacity lift 1/(1−0.55)−1=122.2%. No disclosed UW labor $ or FTE count → cannot compute after_tax_saving=saving$×(1−0.21); eps_uplift_pct=100×after_tax_saving/$4,014.9M adj NI. Feeds capacity, not standalone $. | | |
~40% increase in binding of submissions (Lexington MM property)
engagement · soft | approximately 40% increase | Segment bind-volume +40%. rev_uplift_pct=100×(0.40×P_MM)/26,774M needs P_MM (Lexington MM property premium) — undisclosed. Illustrative only: if P_MM=$1.0B → 1.49% rev, incr NI=400M×11.56%=$46.2M, eps_uplift_pct=1.15%. Not in est aggregate. | | |
26% YoY submission count increase (Lexington)
engagement · soft | 26% increase year-over-year | Top-of-funnel volume +26%. No disclosed Lexington premium $; submissions ≠ bound premium → cannot map to consolidated $26,774M revenue. | | |
35% submit-to-bind ratio increase (Lexington MM property)
engagement · soft | 35% increase | Most revenue-linked metric but bind-rate multiplier 1.35× with no P_MM/bind-count $ base → null. If paired with +26% submissions (same funnel): implied 1.26×1.35−1=+70.1% volume. Not summed with 40% binding claim (overlap). | | |
370,000+ submissions reached; 500,000 target by 2030
other · soft | 370,000+; 500,000 by 2030 | Volume count disclosed but no revenue-per-submission → unsizable to $. Linear glide (500k−370k)/5yr=26,000 subs/yr=7.027% on 370k (submission count, not revenue). Multi-year to 2030 so next-FY $ portion trivial/unpriceable. Upper-bound sensitivity only (not in est): 7.03% of revenue ≈+$1.88B. | | |
AIG Assist deployed across 8 lines of business
other · soft | 8 lines of business | Deployment breadth; no premium, revenue, or cost base disclosed. | | |
AIG Assist expanded to 7 additional lines of business (prior quarter)
other · soft | 7 additional lines of business | Rollout count only; no quantified P&L impact. | | |
Everest portfolio visibility within a week (Palantir ontology)
other · soft | within a week | Speed/insight metric; management says not solely an AI P&L item. No revenue, loss ratio, or expense $ disclosed. | | |
AI expense / benefit disclosure deferred ('27, '28)
other · soft | '27, '28 | Management states expense and revenue components are NOT yet quantified — self-confirms no sizable disclosed figure for next FY. | | |
Assumptions: Adopter-side only (AIG uses AI in underwriting/ontology, not selling AI compute). Incremental net margin for any revenue bridge: 11.56% (=$3,096M/$26,774M GAAP FY2025 net margin). EPS % denominator: adjusted/consensus basis EPS $7.04 × 570.3M shares = $4,014.9M adj NI (NOT GAAP $3,096M/EPS $5.43 — one-offs distort GAAP). Tax 21% reserved for explicit $ opex saves (none disclosed). Phasing irrelevant — no next-FY $ claim. Lexington MM property is a sub-line of AIG NA commercial E&S, small vs $26.774B; no segment premium invented; no double-count of overlapping bind metrics. Bookings = N/A.
Top line: Real but unsizable. Every figure (30% more quoting, 40% more binding, 35% higher submit-to-bind, 26% more submissions) is a percentage on the Lexington middle-market-property funnel with NO disclosed premium $ base. Lexington E&S is a fraction of the $26,774M top line; an illustrative ~5% NWP lift on a hypothetical $1B line ≈$50M=0.19% of revenue. Only anchored volume math is the 370k→500k glide: +26,000 subs/yr=7.027% on the 370k base — not mappable to revenue without $/submission. No defensible consolidated rev_uplift → null rather than invented.
Bottom line: 55% faster quoting is a productivity claim but no underwriter cost/headcount base is disclosed → after-tax saving uncomputable, $0 sized EPS. Net margin (11.56%) is healthy, not thin, so no thin-denominator artifact risk. With no $ saving and no $ revenue, est_eps_uplift_pct is null. Management defers quantified AI expense/benefit disclosure to '27–'28.
[impact n/m (all claims soft/unanchored)] Consensus 2025 revenue $27.21B sits only +1.6% above actual $26.774B, and consensus adjusted EPS climbed 5.05→7.04 (+39.5%) across estimate windows — driven by underwriting margin/buyback dynamics, not isolable AI. AI efficiency gains, bounded well under ~0.4% of consolidated revenue even on generous undisclosed sub-segment bases, are immaterial against that trajectory and plausibly absorbed in consensus noise. No quantifiable AI gap above consensus to call 'ahead' and none below to call 'behind' → unclear, with no isolable AI premium to price.
MODEL CONSENSUS (impact)
partial
Conflicts reconciled
- vs_analyst_expectations: X=inline vs Y=unclear -> used unclear because no isolable AI premium exists to call inline/ahead/behind (Y better justified)
- priced_in: X=high vs Y=medium -> used high (tied; more conservative/less optimistic)
- 370k claim soft: X=false vs Y=true -> used true because volume is disclosed but no $/submission base makes it unsizable, consistent with null pcts
- confidence: not given by X, Y=3 -> lowered to 2 due to category conflicts
| Field | Opus 4.8 | GPT-5.5 |
|---|
| Rev uplift % | – | – |
| EPS uplift % | – | – |
| Priced in | medium | – |
| vs analysts | unclear | – |
| Confidence | 3 | – |
| Top line | Real but unsizable. Every figure (30% more quoting, 40% more binding, 35% higher submit-to-bind, 26% more submissions) is a percentage on the Lexington middle-market-property funnel with NO disclosed premium base. Lexington E&S is a fraction of the $26.774B top line; an illustrative ~5% net-written-premium lift on even a (hypothetical, undisclosed) $1B line is ~$50M = 0.19% of revenue. No defensible consolidated rev_uplift can be computed — set to null rather than invented. | – |
| Bottom line | The -55% underwriter time-to-quote is a productivity claim, but no underwriter cost/headcount base is disclosed, so after-tax saving is uncomputable. Net margin (11.56%) is healthy, not thin, so no artifact risk — but with no $ saving and no $ revenue, est_eps_uplift_pct is null. Management itself defers expense/benefit quantification to FY27-28. | – |
| Reasoning | Consensus 2025 revenue $27.21B sits only +1.6% above actual $26.774B, and consensus adjusted EPS climbed 5.05→7.04 (+39.5%) across the estimate windows — a move driven by underwriting margin/buyback dynamics, not isolable AI. The AI efficiency gains, bounded at well under ~0.4% of consolidated revenue even on generous undisclosed sub-segment bases, are immaterial against that trajectory and plausibly absorbed within consensus noise. There is no quantifiable AI gap above consensus to call 'ahead', and none below to call 'behind' — hence unclear, with no isolable AI premium to price. | – |
Rows highlighted where the two models disagreed.
QUANTIFICATIONS
Quoting more submissions (Lexington middle market property, AIG Assist): 30% improvement (Q1 FY2026 (current results cited on call), topline)
“AIG Assist has helped deliver a 30% improvement on quoting more submissions”
Time to quote for underwriters (Lexington middle market property): 55% reduction (Q1 FY2026, bottomline)
“reduced time to quote for the underwriters by 55%”
Binding of submissions (Lexington middle market property): approximately 40% increase (Q1 FY2026, topline)
“increased binding of submissions by approximately 40%”
Submission count (Lexington): 26% increase year-over-year (Q4 FY2025 full year / YoY, topline)
“Lexington's business has seen a 26% increase in submission count year-over-year.”
Submit-to-bind ratio (Lexington Middle Market property): 35% increase (Q4 FY2025 (since AIG Assist deployment), topline)
“for Lexington Middle Market property, our submit-to-bind ratio increased 35%”
Submissions processed (company ambition / progress): 370,000+ reached; 500,000 target by 2030 (end of 2025 actual; 2030 target, topline)
“our ambition of reaching 500,000 submissions by 2030. As of the end of last year, we've already reached over 370,000 submissions”
AIG Assist lines of business deployment: 8 lines of business (Q1 FY2026 current state, both)
“we began to deploy AIG Assist across 8 lines of business”
AIG Assist lines of business deployment (prior quarter): 7 additional lines of business (Q4 FY2025, both)
“we've expanded its use to 7 additional lines of business, including our Lexington business.”
Everest portfolio visibility via Palantir ontology (speed of insight, not solely AI P&L): within a week (Q1 FY2026 (Everest conversion context), both)
“we were able to get a look at the portfolio within a week about every upcoming month as to what the submission activity is going to be”
AI expense / ROI disclosure timing (guidance, not a realized figure): '27, '28 (forward, both)
“as we get into like '27, '28 ... you'll start to get a lot more clarity in terms of what the expense components are of how we deploy it and what the benefits are on the revenue side”
PAST (realized)
- Q1 FY2026 | Peter Zaffino: In 2025, we launched Underwriting by AIG Assist to help our underwriters review our submissions with more and higher-quality information in a fraction of the time.
- Q1 FY2026 | Peter Zaffino: After a successful launch, we began to deploy AIG Assist across 8 lines of business.
- Q1 FY2026 | Peter Zaffino: In Lexington middle market property ... AIG Assist has helped deliver a 30% improvement on quoting more submissions, reduced time to quote for the underwriters by 55% and increased binding of submissions by approximately 40%.
- Q4 FY2025 | Peter Zaffino: we leveraged our gen AI capabilities to evaluate the Everest portfolio and prioritize the accounts we want to renew in a fraction of the time.
- Q4 FY2025 | Peter Zaffino: we've expanded its use to 7 additional lines of business, including our Lexington business.
- Q4 FY2025 | Peter Zaffino: Lexington's business has seen a 26% increase in submission count year-over-year.
- Q4 FY2025 | Peter Zaffino: for Lexington Middle Market property, our submit-to-bind ratio increased 35%
- Q4 FY2025 | Peter Zaffino: As of the end of last year, we've already reached over 370,000 submissions
CURRENT (now)
- Q1 FY2026 | Peter Zaffino: This quarter, in close partnership with Palantir and Anthropic, we've begun the next phase of agentic AI at AIG that builds on early successes of AIG Assist.
- Q1 FY2026 | Peter Zaffino: we are beta testing the use of multi-agentic solution to enhance our team's productivity, efficiency and learning and development.
- Q1 FY2026 | Peter Zaffino: when we worked with Palantir on the ontology, we were able to get a look at the portfolio within a week about every upcoming month as to what the submission activity is going to be
- Q1 FY2026 | Peter Zaffino: I think as we start to get AI more embedded into Lexington ... We see opportunities because we're not able to service the incredible submission flow
- Q4 FY2025 | Peter Zaffino: We're already seeing benefits from these efforts.
- Q4 FY2025 | Peter Zaffino: we remain on track to complete our accelerated rollout to the rest of North America, U.K. and EMEA in 2026.
- Q4 FY2025 | Peter Zaffino: we're exploring the next phase of our gen AI strategy, focused on the orchestration of AI agents
FORWARD (guidance)
- Q1 FY2026 | Peter Zaffino: we're creating a multi-agentic solution with a strong orchestration layer that coordinates specialized and trained AI agents
- Q1 FY2026 | Peter Zaffino: In this phase, we expect each AI agent to be purpose-built for a specific underwriting function.
- Q1 FY2026 | Peter Zaffino: This ontology, coupled with orchestration, will enable us to deploy multiple AI agent teams to integrate with our core systems, which will improve decision-making and reduce costs over time.
- Q1 FY2026 | Peter Zaffino: I absolutely think in a 5-year period that the global capabilities in terms of the AI orchestration across an organization, just not in underwriting but across from front to back office will be profound.
- Q1 FY2026 | Peter Zaffino: as we get into like '27, '28 ... you'll start to get a lot more clarity in terms of what the expense components are of how we deploy it and what the benefits are on the revenue side.
- Q1 FY2026 | Peter Zaffino: we're going to look at how do you use autonomous with a lot of guardrails and supervision to work through reengineering our workflow.
- Q4 FY2025 | Peter Zaffino: Our top gen AI priorities for 2026 include: deploying underwriting by AIG Assist and claims by AIG Assist across the majority of our commercial businesses; enhancing AIG's ontology ... developing an orchestration layer ... further utilizing gen AI for AIG's SPV strategy, portfolio analytics and compute.
- Q4 FY2025 | Peter Zaffino: our ambition of reaching 500,000 submissions by 2030
- Q4 FY2025 | Peter Zaffino: orchestrating that in an orderly way of being able to get that at scale is what we're going to focus on in 2026.
TRACK RECORD — PROMISE vs DELIVERY
74/100 track record too-early 6 calls reviewed
AIG makes few numeric AI commitments, mostly rollout deadlines; it shows real deployment and productivity metrics but the main 2026 majority-coverage and regional-rollout targets are still open, with only the end-2025 Lexington scope partially met.
Deploy underwriting by AIG Assist across the rest of the Lexington business by end of 2025 — promised Q3 FY2025
partial By Q4 FY2025 AIG reported expansion to seven additional lines including Lexington middle market, but pushed completion of the broader North America, U.K. and EMEA rollout to 2026; Q1 FY2026 cited eight lines live with strong Lexington middle-market metrics but not full Lexington or regional coverage.
Pull forward AIG Assist rollout to the rest of North America, U.K. and EMEA commercial lines by 6 months versus the prior plan — promised Q3 FY2025
too-early Q4 FY2025 said the accelerated program remains on track for 2026 completion; Q1 FY2026 described continued line-by-line deployment and agentic beta work without stating the pull-forward milestone was fully achieved.
Deploy underwriting and claims AIG Assist across the majority of commercial businesses in 2026 — promised Q4 FY2025
too-early Q1 FY2026 reported deployment across eight lines of business and multi-agent beta testing, with no claim that a majority of commercial businesses were yet covered.
PRICED-IN (REFINED)
MEDIUMEst. revisions rising · Fwd P/E 15.4 · EV/Sales 1.8x
AI claim maps to General Insurance Segment, International, North America
Estimate momentum is mixed but net positive: buy ratings edged up (5→6) and holds down (15→14) while forward EPS ramps sharply (4.79→5.05→7.04), baking meaningful earnings growth into consensus; price targets have drifted lower (last-year ~85.6 vs last-quarter ~81.5), so revisions are not uniformly chasing the stock. Valuation is not stretched for a mature insurer (fwd P/E ~15.4, EV/Sales ~1.8, P/B ~1.0), so the market is not paying a clear AI premium despite rising EPS. AI-driven underwriting/claims and expense efficiency would most plausibly flow through General Insurance and both geographic books. Net: some upside is in the numbers (rising EPS) but multiples leave room—priced_in_refined is medium.
COVERAGE — ENTHUSIASM TRAJECTORY + CATALYSTS
8Q4 FY20247Q1 FY20256Q2 FY20254Q3 FY20255Q4 FY20256Q1 FY2026
AI enthusiasm across 6 calls — trend ↘ falling
Peaked with Underwriter Assist and Investor Day GenAI validation, then thinned to brief mentions overshadowed by M&A before modest property tie-ins.
RECENT AI CATALYSTS & NEWS
BUSINESS IMPACT - QUALITATIVE MATERIALITY
6/10 qualitative impact moderate medium-term · mixed evidence
Where AI matters: underwriting & claims workflow
AIG Assist shows credible Lexington funnel gains (30% more quoting, 55% faster quotes, ~40% higher binding) and rollout across eight lines, but management cites no AI-attributed revenue, margin, or expense dollars and defers clearer ROI to 2027–28, so consolidated uplift likely stays well under ~0.5% of ~$27B revenue.
Caveats: Segment metrics (Lexington MM property) may not replicate or scale to consolidated P&L; No isolable AI revenue/margin/EPS disclosure until 2027–28; Industry-wide AI adoption may compress underwriting edge and broker economics; Variable AI/compute costs and execution risk on multi-agent rollout
AI DISRUPTION / CANNIBALIZATION RISK tailwind · 3/10
AI automates submission triage and workflow, not the core risk-transfer product; AIG still earns on capital, claims paying, and distribution, and scale plus ontology/agent orchestration plausibly favor large adopters more than they commoditize the model.
OPTIONS / MARKET STRUCTURE
option liquidity: good
proxy inputs — dollar-ADV $310M · beta 0.541 · px $73.80
source: proxy (no options chain on FMP)
FMP /stable/ exposes no options-chain endpoint on this key, so ATM IV, bid-ask spread and open interest are unavailable. Liquidity below is a PROXY from dollar-ADV, beta and price level (a stand-in for option depth), not measured option-market data.
CONFIRMATION — INSIDERS · 13F · LANGUAGE
Confirming — insiders buying, institutions flat, management language 6/10 measured.
INSIDERS buying 1 open-market buy(s) vs 0 sell(s) — net accumulation
INSTITUTIONS (13F) flat as of 2026-03-31: 133 new / 176 closed positions; 504 increased / 453 reduced; institutional ownership -2.80pp; -41 net 13F holders
MGMT LANGUAGE 6/10 measured Shipped AIG Assist with hard metrics and partners; future agentic rollout stays heavily qualified with may, could, should, and potential.
commit “AIG Assist has helped deliver a 30% improvement on quoting more submissions, reduced time to quote for the underwriters by 55% and increased binding of submissions by approximately 40%.”
commit “After a successful launch, we began to deploy AIG Assist across 8 lines of business.”
commit “This quarter, in close partnership with Palantir and Anthropic, we've begun the next phase of agentic AI at AIG”
VERBATIM AI QUOTES
“outline the progress we're making on our AI and digital strategies.”
— Peter Zaffino, Q1 FY2026
“That work reinforced our conviction that AI has the potential to materially improve performance and drive better solutions for our clients and for AIG.”
— Peter Zaffino, Q1 FY2026
“Our approach to using AI has been focused on 3 important components. First, you have to have an understanding of the technology and capabilities of large language models. Second, you have to have pattern recognition in order to know how to apply AI to your business. And third, you have to have a culture and a track record of execution in order to effectively deploy AI within an organization.”
— Peter Zaffino, Q1 FY2026
“We started our AI journey at the core of our business in underwriting, where we felt the impact will be most profound.”
— Peter Zaffino, Q1 FY2026
“In 2025, we launched Underwriting by AIG Assist to help our underwriters review our submissions with more and higher-quality information in a fraction of the time.”
— Peter Zaffino, Q1 FY2026
“After a successful launch, we began to deploy AIG Assist across 8 lines of business.”
— Peter Zaffino, Q1 FY2026
“In Lexington middle market property, which is an area we have targeted for growth, AIG Assist has helped deliver a 30% improvement on quoting more submissions, reduced time to quote for the underwriters by 55% and increased binding of submissions by approximately 40%.”
— Peter Zaffino, Q1 FY2026
“This quarter, in close partnership with Palantir and Anthropic, we've begun the next phase of agentic AI at AIG that builds on early successes of AIG Assist.”
— Peter Zaffino, Q1 FY2026
“Using Palantir's Foundry platform, we expanded our ontology, a digital map of our business that included our underwriting processes, workflows and data relationships.”
— Peter Zaffino, Q1 FY2026
“we're creating a multi-agentic solution with a strong orchestration layer that coordinates specialized and trained AI agents to seamlessly supplement our underwriters' analysis”
— Peter Zaffino, Q1 FY2026
“Human oversight is and will continue to be essential to our underwriting processes.”
— Peter Zaffino, Q1 FY2026
“we are beta testing the use of multi-agentic solution to enhance our team's productivity, efficiency and learning and development.”
— Peter Zaffino, Q1 FY2026
“We've been deliberate in our growth and believe our AI implementation, which I will discuss later in more detail, will further enable this.”
— Peter Zaffino, Q1 FY2026
“As Peter has shared in depth, we are implementing a leading AI strategy designed to rapidly evolve alongside other advances in technology to deliver growth, data insights and quality decision-making.”
— Eric Andersen, Q1 FY2026
“Gen AI capabilities”
— Eric Andersen, Q1 FY2026
“when we worked with Palantir on the ontology, we were able to get a look at the portfolio within a week about every upcoming month as to what the submission activity is going to be and what we like for pricing and what we thought we needed to restructure.”
— Peter Zaffino, Q1 FY2026
“I think as we start to get AI more embedded into Lexington, it's not because we see massive growth opportunities because the market is there. We see opportunities because we're not able to service the incredible submission flow, which has still been very strong.”
— Peter Zaffino, Q1 FY2026
“highlight our progress on our gen AI and data and digital strategies.”
— Peter Zaffino, Q4 FY2025
“To support the conversion, we leverage our gen AI capabilities to evaluate the Everest portfolio and prioritize the accounts we want to renew in a fraction of the time.”
— Peter Zaffino, Q4 FY2025
“we leveraged our gen AI solution underwriting by AIG Assist to accelerate the conversion process in key lines of business, increasing renewal speed significantly.”
— Peter Zaffino, Q4 FY2025
“This is also the first time we've deployed our gen AI capabilities in an SPV transaction. Partnering with Palantir, we use large language models to match data and define risk characteristics within Amwins program business that were aligned with the Syndicate's risk appetite.”
— Peter Zaffino, Q4 FY2025
“Our top gen AI priorities for 2026 include: deploying underwriting by AIG Assist and claims by AIG Assist across the majority of our commercial businesses; enhancing AIG's ontology by developing a comprehensive digital twin of AIG's processes, workflows and data elements to drive enhanced speed and efficiency; developing an orchestration layer to coordinate AI agents to drive better decision-making and reduce costs across the organization; and further utilizing gen AI for AIG's SPV strategy, portfolio analytics and compute.”
— Peter Zaffino, Q4 FY2025
“we've expanded its use to 7 additional lines of business, including our Lexington business.”
— Peter Zaffino, Q4 FY2025
“Lexington's business has seen a 26% increase in submission count year-over-year.”
— Peter Zaffino, Q4 FY2025
“for Lexington Middle Market property, our submit-to-bind ratio increased 35%”
— Peter Zaffino, Q4 FY2025
“We believe our use of gen AI gives us a strong advantage going forward in this dynamic market.”
— Peter Zaffino, Q4 FY2025
“we think of these AI agents as companions that operate alongside our teams with specific roles such as knowledge assistants that can provide relevant information in real time, advisers that can provide additional insight based on historical use cases and critic agents that challenge the knowledge and adviser agents as well as the underwriters' decisions.”
— Peter Zaffino, Q4 FY2025
“we've been a leader in gen AI.”
— Peter Zaffino, Q4 FY2025
ANALYST QUESTIONS ON AI
Q (Q1 FY2026, Meyer Shields (KBW)): as leading carriers and brokers, both successfully adopt AI, how does that impact what the carriers pay to the brokers? ... how you expect that to play out.
A: Peter Zaffino: how we interact with the brokers ... is going to be more of how we all get so much more efficient in exchanging data and information on submissions. ... as we look at the way in which data is being ingested through the mechanisms of a variety of large language models, I think we will be able to augment information that we get in submissions to be able to make better underwriting decisions. ... as enterprise becomes a much bigger part of large insurance companies and large insurance brokers, the ability to collaborate will get even stronger.
Q (Q1 FY2026, Jian Huang / Bob Huang (Morgan Stanley)): multi-agent collaboration ... orchestration layer ... 5 years down the road, 10 years down the road, there should be global-wide capability ... underwriting and your understanding of risk would be much more uniform globally ... differentiation between you and other more regional underwriters?
A: Peter Zaffino: 5 to 10 years, we couldn't predict a year out. ... I absolutely think in a 5-year period that the global capabilities in terms of the AI orchestration across an organization, just not in underwriting but across from front to back office will be profound. ... you need to have size, scale and ability to beta test ... Europe ... very hard to beta test or roll something out in Europe without it being tested somewhere else ... We're doing that in our Japan business.
Q (Q1 FY2026, Jian Huang / Bob Huang (Morgan Stanley)): AI expense costs ... Claude 2.0 ... variable cost ... how that factors into your ROE considerations
A: Peter Zaffino: as we get into like '27, '28 ... you'll start to get a lot more clarity in terms of what the expense components are of how we deploy it and what the benefits are on the revenue side. ... first case was to go to the heart of the company, which is underwriting and then to go to claims. ... moved more from Gen AI to agentic and now, we are going to look at how do you use autonomous with a lot of guardrails and supervision to work through reengineering our workflow. ... efficiencies that will create the bandwidth to reinvest in the business. ... more capabilities, more insight, more benefit for brokers and clients, create our own bandwidth for investment by reengineering process and having the ability in certain markets to be able to grow exponentially when there's opportunities.
Q (Q4 FY2025, Taylor Scott / Alex Scott (Barclays)): expense ratio ... moving pieces with ... some of the AI initiatives. ... what we can expect from the expense ratio over the next few years
A: Peter Zaffino: (No AI-specific answer.) Focused on parent expense apportionment, PCS cleanup, full-year expense discipline, and path to sub-30% expense ratio by 2027 without quantifying AI savings.
Q (Q4 FY2025, Jian Huang / Bob Huang (Morgan Stanley)): orchestration layer ... sits on top of all the technology for AIG ... or ... just for localized AI systems
A: Peter Zaffino: orchestrating that in an orderly way of being able to get that at scale is what we're going to focus on in 2026. ... we've been experimenting with multiple agents on the underwriting side than the functional side. ... back office, we outsourced to Accenture. ... We share in the savings. We share in the design of the orchestration and how it actually comes into our workflow. ... it will be on the technology stack, but I'm talking more about orchestrating a significant amount of agents within the organization that are more organized.
Q (Q4 FY2025, Jian Huang / Bob Huang (Morgan Stanley)): low-hanging fruits ... show up in the numbers ... complicated projects ... further out 5 years down the road?
A: Peter Zaffino: the first one is absolutely to reduce cycle time with the higher quality data to the underwriter. ... massive shift in our ability to process a significant submission flow way beyond our expectations without additional human capital resources. ... the real long-term opportunity is going to be getting the orchestration of agents in an organization to be able to scale ... shrink all of that with the implementation of gen AI and multiple agents with a proper orchestration. ... the acceleration and the opportunity is greater than I thought at Investor Day.