AIG's AI Strategy: How the $200B Commercial Insurance Giant Is Reinventing Underwriting With Conversation

TL;DR#

American International Group (AIG) — the $200B global commercial P&C carrier — is rebuilding its underwriting stack around AI-led conversation instead of broker-form data collection. Under CEO Peter Zaffino, AIG has consolidated 750+ legacy applications, deployed Anthropic's Claude internally across underwriters, and partnered with Palantir and Salesforce to build "AIG Underwriter Companion," a system designed to read submissions, surface coverage gaps, and ask the next question a senior underwriter would ask. The strategic bet: commercial underwriting is a narrative business — risk variance on a single $50M property schedule is wider than every line of personal auto combined — and you cannot capture narrative through a 40-field broker form. AIG reports its underwriters now spend an estimated 50%+ less time on data ingestion and submission triage, freeing them for the judgment work that actually prices risk. This is the commercial-side counterpart to Lemonade's conversational AI playbook on the consumer side — and a leading indicator for the $700B global commercial P&C market. Carriers still relying on Acord forms plus underwriter brain power are about to face a structural disadvantage.

What is AIG doing with AI in commercial insurance?#

AIG is replacing its broker-form-and-spreadsheet submission workflow with a conversational, AI-assisted underwriting layer that reads risk submissions, asks clarifying questions, and surfaces the narrative context senior underwriters need to price complex commercial risk. The program — internally branded "AIG Underwriter Companion" — combines Anthropic's Claude (for reasoning and dialogue), Palantir Foundry (for the underlying ontology of policies, exposures, and historical losses), and Salesforce (for the broker-facing workflow). According to AIG's 2025 investor materials and remarks from CEO Peter Zaffino, the system has materially compressed submission triage time and is being rolled out across Lexington (E&S), Glatfelter, AIG Re, and the core commercial property and casualty lines through 2026.

The bigger shift behind the tooling is philosophical. AIG is treating underwriting submission intake the same way Lemonade treated consumer claims FNOL — as a conversation that should adapt to the risk in front of it, not a fixed schema every broker has to translate into. That single design choice cascades through broker engagement, risk-narrative capture, and the economics of underwriting headcount.

Why commercial underwriting breaks under traditional broker-form workflows#

Commercial underwriting breaks under broker-form workflows because commercial risk has unbounded variance, depends on narrative context, and rewards iterative follow-up — none of which a static Acord form can deliver. A personal auto policy has maybe a dozen real risk dimensions; a $50M manufacturing property schedule has hundreds, and a directors and officers policy on a pre-IPO biotech has almost no fixed fields at all. Yet for thirty years the industry has tried to compress that variance into the Acord 125, 126, 140 family of forms.

Three structural problems make form-based commercial intake fundamentally lossy:

• Risk variance is unbounded. A 1990 paper mill and a 2024 lithium-ion battery plant share the Acord 140 schema but share almost nothing else. Senior underwriters know to ask 40 different follow-up questions for each — but the form is the same, so the broker gives the same data.
• Narrative context is where the money is. McKinsey's 2023 Insurance 2030 research identifies risk-narrative quality as the dominant driver of loss ratio in commercial lines — yet traditional submissions strip narrative in favor of structured fields. The conversational signals — "they just spun up a new product line in Mexico" — never enter the system.
• Follow-up depth is missing. A form has no idea you answered something interesting. If a broker checks a box that opens up $2M of additional exposure, the form doesn't probe — but a junior underwriter would, and the AI now can.

This is the same failure mode we cover in our broader analysis of why forms fail at customer research — but with carrier loss ratios on the other side of the lossy capture. The form problem isn't unique to insurance; it's just particularly expensive in commercial P&C because the dollars per submission are so large.

Inside AIG Underwriter Companion: what changed in submission intake#

AIG Underwriter Companion changed three things at once: submission intake became conversational, broker engagement became iterative, and risk-narrative capture became structured by the LLM rather than by a fixed form. According to AIG's January 2025 announcement with Anthropic and Palantir, and confirmed in subsequent investor calls, the system now performs four functions inside the underwriter's workflow:

1. Ingests heterogeneous submissions. A broker emails a 90-page PDF property schedule, a loss run, and three Excel exposure tabs. The system parses all of it, normalizes against AIG's policy ontology in Palantir, and presents the underwriter with a single underwriting view.
2. Asks the missing questions. Where prior tooling required underwriters to manually request supplementary info (often via a separate email thread that died), the AI generates a structured list of follow-up questions — and where AIG's broker portal supports it, can send those as a conversational follow-up to the broker.
3. Surfaces narrative risk signals. The system extracts the unstructured-text risk signals — "new sprinkler system installed Q3 2024," "loss in 2022 from a contractor on uninsured premises" — and tags them against AIG's loss-cause taxonomy. Underwriters see narrative risk, not just structured fields.
4. Drafts underwriting rationale. Senior underwriters still own the decision, but the system drafts the rationale memo, pre-fills the binder request, and flags the specific exposures the underwriter chose to accept, decline, or surcharge.

The internal metric AIG has emphasized publicly is submission-to-quote time, which the company says has dropped by double-digit percentages in pilot lines. The harder-to-measure but more strategic metric is what underwriters do with the time they get back — which AIG describes as moving from "data ingestion" work to "judgment" work.

For carriers building similar capability, the pattern is recognizable in adjacent verticals — see the same playbook in our analysis of Travelers' conversational underwriting shift, Pie Insurance's AI-first workers' comp underwriting, and Root's behavior-based conversational risk interview. The common thread: the conversation is the underwriting artifact, not the form.

The broker channel: winning brokers as users, not gatekeepers#

AIG's bet is that the broker channel becomes a competitive moat when brokers prefer to submit business to AIG because AIG's intake experience is the fastest and smartest in the market. For decades, the commercial broker relationship has been a friction tax — submission packages get sent to three or four markets, the carrier with the most patient underwriter and most flexible appetite wins, and the broker eats the operational overhead in the middle. AIG's strategic question is: what if our intake didn't require the broker to do that operational work at all?

Three changes flow from that question:

• Submissions accept whatever format the broker has. A broker who runs on Applied Epic, one who runs on EZLynx, one who emails PDFs — all converge in AIG's intake without translation work. The conversational layer handles the variance.
• The system asks follow-ups in the broker's voice. Rather than the broker chasing the insured for missing info on AIG's behalf, AIG's AI can run a structured follow-up conversation directly with the broker (and, in some configurations, with the insured) — pulling the narrative that used to live in email threads.
• The broker sees AIG's underwriting reasoning, not just the quote. When AIG declines or surcharges, the underwriter rationale is surfaced to the broker through the same conversational layer — which means the broker can negotiate with their insured against actual risk findings rather than a black-box "no."

This pattern of treating intake as a conversation rather than a form is the same logic Perspective AI customers apply when they deploy AI concierge agents for high-stakes inbound flows — the intelligent intake product for legal, insurance, and financial-services workflows is built around exactly this principle. According to the 2024 Council of Insurance Agents & Brokers Commercial P/C Market Index, the commercial market has been hardening for 16 consecutive quarters — making broker experience and submission speed a real determinant of carrier win-rate.

What this signals for the $700B commercial P&C market#

AIG's conversational underwriting move is the leading indicator that the commercial P&C market is about to bifurcate between carriers with AI-led intake and carriers still running broker-form workflows. The global commercial property and casualty market is roughly $700–800B in direct written premium, with the U.S. alone at approximately $470B according to Insurance Information Institute data. Three structural forces compound to make conversational intake a competitive necessity, not a nice-to-have:

The carriers building this capability now — AIG, Travelers, Chubb (see our Chubb specialty insurance AI analysis), Pie Insurance, and the European leaders like Allianz and Zurich (see Allianz's AI customer research strategy and Zurich's commercial lines customer engagement) — are quietly resetting the cost structure of commercial underwriting. The carriers that wait until 2027 to start will face a 2–3 year capability gap on a moat (broker preference) that compounds annually.

The downstream implications run further than underwriting:

• Claims will follow. What works for submission intake works for first-notice-of-loss; see our analysis of AI claims processing and the conversational FNOL shift.
• Personal lines is already there. Lemonade has proven the model in pet, renter, and homeowner lines; Hippo did it for home with IoT data; commercial is the larger, slower-moving prize.
• Group benefits and life are next. See our Prudential conversational policyholder research analysis and MetLife's group benefits AI strategy for how the pattern lands in life and benefits.

For carriers, MGAs, and brokers building their own capability, our practical commercial insurance AI guide for brokers, MGAs, and carriers maps the build-vs-buy decision tree.

Frequently Asked Questions#

What is AIG Underwriter Companion?#

AIG Underwriter Companion is the internal name for AIG's AI-assisted commercial underwriting system, built in partnership with Anthropic, Palantir, and Salesforce. It ingests broker submissions in any format, normalizes them against AIG's underlying policy and exposure data model in Palantir Foundry, uses Anthropic's Claude to read the submission and surface the narrative risk signals a senior underwriter would catch, and presents underwriters with a structured set of follow-up questions and a draft rationale. AIG began rolling it out across commercial lines starting in early 2025.

How is conversational underwriting different from form-based underwriting?#

Conversational underwriting captures risk through adaptive dialogue rather than fixed-field forms. A traditional Acord-form submission collects the same 40–200 fields regardless of risk type, which compresses unbounded commercial risk variance into a lossy schema. A conversational system reads what the broker submitted, identifies what's missing or ambiguous, and asks the next question — the same way a senior underwriter would. The output is a richer risk picture, faster quotes, and better loss-ratio signal at intake time.

Will AI replace commercial underwriters?#

No — AI is augmenting commercial underwriters, not replacing them, because commercial underwriting is fundamentally a judgment business. The submission-triage, data-ingestion, and rationale-drafting work that consumed 50%+ of an underwriter's day is being automated, but the judgment decisions on appetite, pricing, terms, and binder authority remain with human underwriters. AIG, Travelers, and Chubb have all publicly emphasized this framing. The likely impact is that each underwriter's portfolio capacity grows substantially, which is more about volume per underwriter than headcount reduction.

What's the difference between AIG's approach and Lemonade's?#

AIG and Lemonade both replaced forms with conversation, but at opposite ends of the market. Lemonade built a consumer-first, AI-native carrier from scratch for pet, renter, and homeowner lines — where the policy is simple and the bet was on customer-experience speed. AIG is retrofitting conversational AI into a $200B legacy commercial carrier where the policies are complex, the channel is broker-mediated, and the bet is on underwriter productivity and risk-narrative capture. Different markets, same conviction: the form is the wrong primitive.

Which commercial insurance lines are most affected first?#

E&S (excess and surplus), specialty lines, and middle-market property are the lines moving first because their underwriting depends most heavily on narrative risk context that forms can't capture. AIG has led with Lexington (E&S) and is extending across property, financial lines, and casualty. Workers' comp is moving in parallel — see Pie Insurance's AI-first workers' comp underwriting case study. Personal auto and small-business package policies will follow, but the highest-leverage applications are in complex risks where senior-underwriter time is the binding constraint.

How should mid-size carriers and MGAs respond?#

Mid-size carriers and MGAs should pilot conversational intake on one specialty line in 2026 rather than wait for an enterprise-wide AI program. The strategic risk is broker preference compounding against you — every quarter that AIG, Chubb, and Travelers offer faster, smarter intake is a quarter brokers route more business their way. Start with a single line where submission triage is the bottleneck, deploy a conversational layer (build or buy), and measure broker submission velocity. Our practical commercial insurance AI guide maps a 90-day pilot path.

Conclusion: The Form Is the Wrong Primitive for Commercial Insurance AI#

AIG's conversational underwriting program is the most important commercial insurance AI signal of 2025–2026 — not because the technology is novel (it isn't), but because the world's largest commercial P&C carrier has publicly committed to rebuilding submission intake around conversation rather than forms. The form is the wrong primitive for commercial insurance AI. Risk has unbounded variance, decisions hinge on narrative context, and the underwriters who price risk well are the ones who ask the next question. AI-led conversation finally lets that pattern scale.

The same logic plays out everywhere intake matters. Carriers, brokers, MGAs, and frankly anyone running a high-stakes inbound flow — legal intake, patient screening, B2B sales discovery, customer research — face the same form-versus-conversation choice AIG just made. Perspective AI is built on the same conviction: when the stakes are real, you replace the form with a conversation that adapts. Our intelligent intake product is designed for exactly the high-variance, narrative-heavy capture problems that have historically been jammed into web forms — from insurance submissions to legal intake to enterprise research.

If you're building commercial insurance AI capability, or running any intake flow where the form is the bottleneck, book a 20-minute walkthrough — we'll show you what conversational intake looks like in your domain and how to pilot it against one workflow in 90 days. For more on the AI-first carrier playbook, see our roundups of the best AI tools for insurance brokers and our analysis of AI underwriting software compared across personal, commercial, and life lines.