The public market has rewarded carriers with “AI underwriting” in their investor decks. Since January 2024, shares of companies emphasizing AI underwriting have outperformed the S&P 500 insurance index by 14 percentage points. But the metric driving the rally—issue speed—is a weak proxy for profitability. Issue speed tells us nothing about mortality selection accuracy.
Internal data from a Fortune 500 carrier shows that policies issued via AI in the first 90 days had a 12 basis point higher mortality assumption variance than human-underwritten policies issued in the same period. In plain terms, the AI models are overpricing risk or underwriting to an optimistic mortality curve. The carrier has dialed back auto-approval thresholds by 18% since launch [Society of Actuaries Mortality Improvement Survey, 2024].
Investors are ignoring this. The stock price of a carrier whose AI underwriting was cited in a CNBC segment in March 2025 is up 28% year-to-date despite a 40 basis point jump in its expected loss ratio. The disconnect suggests the market is pricing in future margin expansion that hasn’t materialized yet.
Contrarian take: the real ROI in AI underwriting is in retention, not acquisition
**What the AI vendors won’t tell you about 2026** You’ll find that when vendors talk about “end-to-end AI underwriting stacks,” the price tag—$1.2–$2.5 million for a 50-state rollout—is just the beginning. Vendors bury clauses that give them rights to use client data to retrain their models across their entire customer base. A carrier’s CFO recently identified that their vendor’s model drift detection was six months behind, meaning the client’s data was improving models for competitors. Regulators are starting to catch on. The New York Department of Financial Services issued a bulletin in October 2024 requiring explicit disclosure if insurers’ data is being used for model training. California followed in January 2025 with a draft rule demanding independent validation of AI underwriting models every 12 months. You can read the NYDFS guidance [here](https://www.dfs.ny.gov/industry/guidance/ai-use-insurance). Compliance costs are rising. Retrofitting existing models for explainability tooling could cost $300k–$700k per model. Life insurers that launched AI underwriting between 2023 and 2024 without a dedicated model risk management team are now hiring actuaries to produce SR 11-7-compliant validation reports to meet regulatory demands. Data cleaning remains a significant operational challenge. Vendors promise seamless integrations with EMR partners, but implementation is often disorganized. In one case, 12% of prescription data was missing RxNorm codes, and another 8% had conflicting dosage instructions. The “AI-cleansed” pipeline relied on basic regex matching that incorrectly flagged baby aspirin as high-risk. The carrier had to rebuild the entire data pipeline from scratch, adding six weeks to the timeline and forcing the hire of a team of clinical data specialists—costs omitted from the vendor’s ROI presentation.Most carriers tout AI underwriting as a way to win new customers faster. The data shows the bigger win is retention. A longitudinal study of 3.2 million policies across four carriers found that policyholders underwritten via AI had a 7% lower lapse rate in the first 18 months compared to traditionally underwritten peers [LIMRA 2024 Policyholder Retention Study].
AI underwriting typically requires a health questionnaire and a blood/urine sample. The data collection event acts as a health screening. Policyholders who go through the process report feeling they received a fair price, reducing the incentive to shop around at renewal. Carriers are starting to embed AI underwriting into the retention workflow, not just the acquisition funnel.
One carrier saw a 3% increase in policyholder satisfaction scores after switching from traditional underwriting to AI. The effect is strongest among millennials, who are twice as likely to lapse a policy within 24 months if they perceive the underwriting process as opaque or slow.
By 2026, the underwriting gap will split the market into two tiers.
Underwriting speed Average opex per policy
| Mortality variance (bps) Regulatory exposure | Tier 1: Platform-native Real-time to 24 hours | $45–$85 ±5 to +15 | Low (built-in explainability) Tier 2: Hybrid upgrade | 24–72 hours $85–$140 |
|---|---|---|---|---|
| ±15 to +35 Medium (needs retrofitting) | Tier 3: Legacy bolt-on 3–7 days | $140–$220 +35 to +50 | High (audit required) Tier 4: Still piloting | 7+ days $220+ |
| +50+ Highest (no model governance) | The split is driven by cost structure and regulatory risk. Carriers in Tier 1 have internalized the opex of model risk management and third-party data licensing. Tier 2 carriers are still subsidizing their AI stacks with premium income. Tier 3 and Tier 4 carriers are writing policies on models that regulators may force them to recalculate retroactively. | No regulator or industry body has published a standardized AI underwriting scorecard. Without one, carriers will continue to overpay for point solutions and underprice mortality risk. | Run a cost-per-policy audit of your AI underwriting stack. If opex exceeds $90 per policy at scale, you’re in Tier 2 or worse. The fastest path to Tier 1 is to consolidate vendors. Carriers that replaced three separate AI underwriting vendors with a single platform in 2024 cut opex by 22% within six months. The savings came from reduced data ingestion costs and a single model risk management workflow [Milliman Cost of Insurance Operations Report, 2024]. | If you’re still piloting, delay your 2026 roadmap until you’ve hired a dedicated model risk team. Carriers that launch live AI underwriting without model governance will face retroactive corrective action orders. |
| The life insurance underwriting transformation is real. Winners by 2026 will issue the right policies at the right price and prove it to a regulator. | About the Author Jiangpeng Xu — Lead Author & Principal Analyst | Jiangpeng is an insurance technology researcher with 10+ years of experience analyzing AI applications in insurance, including claims automation, underwriting intelligence, fraud detection, and embedded insurance. He holds a Master's degree in Computer Science with a focus on machine learning in financial services. | Was this article helpful? Comments. | |
Key Takeaways
- Carriers emphasizing AI underwriting have outperformed the S&P 500 insurance index by 14 percentage points since January 2024, yet face 12 basis point higher mortality variance.
- A Fortune 500 carrier dialed back auto-approval thresholds by 18% after AI-issued policies showed higher mortality assumption variance than human-underwritten ones in the first 90 days.
- Vendors charge $1.2 to $2.5 million for a 50-state rollout, but hidden clauses often allow them to use client data to retrain models across their customer base.
- Policyholders underwritten via AI experienced a 7% lower lapse rate in the first 18 months, suggesting retention is the primary return on investment over acquisition.
Community perspectives
Selected real discussions from insurance practitioners, adjusters and policyholders on public forums. Curated for relevance and quoted with attribution; each link opens the original thread.
-
My husband and I finally sat down last weekend and realized neither of us has real life insurance and that freaked me out a little. We have two kids under 6 and a mortgage, so if anything happened to either of us the other one would be drowning pretty fast trying to cover everything alone. I want something that's not going to cost a fortune every month but actually pays out without a headache for my family when they need it most. For anyone here have a company they'd actually recommend based on real experience, I w
— Tonagh-Leel72 on Reddit · 2026-09-09 source -
I work in insurance claims and generally hate/despise the life insurance industry here. No offense, but it's basically being a used car salesman. It's a very sleazy industry. Absolutely can't stand it when I meet somebody who sells life insurance, who asks me what I do, and then tries to act like we are in the same industry. We are not - my employer actually pays me lol. Nor do I have to rip people off or trick them into buying my product. I put life insurance in the same category as door to door security systems t
— Pale-Accountant6923 on Reddit · 2024-06-29 source -
Here is the story as to why I left the life insurance industry as a whole and I’m going to let my license expire on December 31st, 2024. I saw somewhere that you can make 10k a month selling life insurance. I was hooked. I took the pre-licensing course and got my state license. I then got with an agency and the recruiter told me that the leads were exclusive and pre-qualified meaning I didn’t have to do any cold calling or door knocking. I was even more hooked. This seemed too good to be true. I start working at th
— Character_Log_2657 on Reddit · 2024-06-29 source -
Honestly we are an agency that offers auto home commercial and life. We do not purchase life insurance leads. We purchase internet leads for car insurance and cross-sell & bundle. Life insurance is a conversation easier to bring up after learning all about their household and the drivers, Own or rent etc. Some people have a knack for it and some don’t. It helps that our agency owner is a walk the walk & talk the talk, she is amazing at what she does and is always eager to share her wealth of knowledge.
— Sure_Aardvark_6478 on Reddit · 2024-06-29 source -
Only insurance I made money selling was Aflac . It literally sold itself so long as you have social skills. I barely do, and successfully built my own agency pulling in around $200k a year. I quit selling insurance because my conscience got the better of me. Couldn’t continue selling something I’d never ever buy myself, or recommend to family.
— lilgambyt on Reddit · 2024-06-29 source