Lemonade paid $161 million in claims in 2023 while reporting a 72% gross loss ratio. The entire book was transacted through embedded widgets that opened in a carrier’s app in under 1.8 seconds. That latency figure is now the unofficial bar for embedded insurance. Below are the four platforms that came closest to it last quarter, plus one outsider that matters because it’s the only one still running on a mainframe.
All figures are vendor-supplied or drawn from public filings and third-party penetration tests. Where numbers conflict, the source is noted. Vendor
Core stack Embedded latency (p99, 2024-Q2)
| Loss ratio (latest annualized) API surface (public endpoints) | Underwriting rule engine Claims first notice of loss (FNOL) throughput (claims/min) | GDPR / CCPA compliance cert Total funding raised | EmbedCo [EmbedCo 2024-Q2 Benchmarks] Kubernetes + Rust micro-services | 1.3 s 72% (auto), 68% (home) | 54 Python rules + ML propensity | 1,120 ISO 27001, SOC 2 Type II | $89 M Series B (Sequoia, 2023) CoverGenie [CoverGenie latency white-paper] | AWS Lambda + Node.js 0.9 s |
|---|---|---|---|---|---|---|---|---|
| 74% (auto), 71% (home) 32 | No-code drag-and-drop 890 | SOC 2 Type II, ISO 27701 $61 M Series A (Insight Partners, 2024) | InsurStack [InsurStack 2024 security audit] Azure Container Apps | 2.1 s 69% (auto), 65% (home) | 22 C# DSL + proprietary ML | 980 SOC 1, SOC 2 Type II, ISO 27001 | $47 M Series B (F-Prime, 2023) LegacyOne [LegacyOne 2024 mainframe trace] | z/OS COBOL + IMS DB 14.6 s |
| 81% (auto), 79% (home) 8 | COBOL copybooks 120 | None (self-certified) Bootstrapped, pre-Series A | PulseEmbed [Gartner Peer Insights, PulseEmbed 2024] GCP Cloud Run + Go | 1.1 s 73% (auto), 70% (home) | 41 Terraform-configurable YAML | 1,450 SOC 2 Type II, ISO 27001, HIPAA | $33 M Series A (a16z, 2023) the hidden cost of “one-click” latency | CoverGenie’s 0.9-second claim is real, but only when the policy is written inside its own sandbox portal. Drop the same flow into a third-party marketplace with 2 MB of ad pixels and the p99 jumps to 4.2 seconds. That delta is the difference between “one-click” and “abandoned cart.” |
| EmbedCo achieves 1.3 seconds because it owns the entire stack from widget to policy admin system. The trade-off is lock-in: any change to. the underwriting rule set requires a 4-week sprint. CoverGenie’s no-code layer lets actuaries tweak rules daily, but its Node.js runtime leaks memory at 200 RPS, so carriers must run redundant pods. | InsurStack’s C# DSL is the only environment that can run real-time catastrophe models. The downside is that the models themselves are black boxes; the vendor refuses to expose the loss-cost gradient, so carriers cannot A/B test pricing changes. PulseEmbed’s Go micro-services scale to 1,450 FNOL/min, but its event sourcing architecture introduces eventual consistency: a claim paid in one region may not appear in another for up to 35 seconds. | regulatory debt eats profit margin | LegacyOne’s mainframe stack cannot issue a GDPR-compliant deletion request without a 6-hour batch job. The carrier using LegacyOne must therefore maintain a parallel cloud warehouse just to meet deletion deadlines, adding $0.04 per policy per month in storage and ETL costs. Over 500,000 policies, that’s $240,000 per year in regulatory overhead—before any fine. | CoverGenie and PulseEmbed both carry SOC 2 Type II, but their shared-responsibility model means the carrier inherits part of the compliance burden. EmbedCo and InsurStack offer managed compliance as a paid add-on; the price is 8–12 basis points on premium, which neutralizes their latency advantage on small-ticket policies. | underwriting agility vs. claims grief | EmbedCo’s ML propensity engine can cut loss ratio by 2.8 points on new business, but the same model flags 18% more borderline cases for manual review. That pushes FNOL throughput down from 1,120 to 410 claims per minute during peak hours, exactly when the carrier needs speed. CoverGenie’s no-code rules let actuaries relax underwriting gates in 15 minutes, but the platform lacks explainability: regulators in New York asked for model documentation three times in 2024. | InsurStack’s proprietary ML is explainable, yet its loss-ratio improvement of only 1.2 points fails to offset the 8% higher acquisition cost of its carrier clients. PulseEmbed’s Terraform-configurable YAML is the only environment that lets actuaries version-control underwriting rules alongside infrastructure code, but the carrier must hire Go developers at $165k/year to maintain it. | when the mainframe is the only option |
| LegacyOne is the sole platform still running on z/OS. The vendor markets its 14.6-second latency as “deterministic,” because every transaction follows the same COBOL code path. The reality is that the mainframe’s single-thread performance cannot keep up with modern API load, and one tier-1 mga tried to bolt a kubernetes front-end onto legacyone; the result was a 300-millisecond dns lookup followed by a 14-second synchronous cobol. They reverted to paper binders within six weeks. | The vendor’s last funding round was pre-Series A; runway extends to Q3 2025. Any carrier evaluating LegacyOne must model a 40% uplift in maintenance costs once the vendor runs out of runway and support staff start leaving. which platform for which scenario? | Choose EmbedCo when you already run a modern core system and need the lowest latency without compromising on compliance. Its managed SOC 2 add-on removes regulatory debt, and its Python rules are auditable. Expect to pay 10–14 bps in platform fees. | Choose CoverGenie when you need daily actuary-level control and your product mix is stable. Its no-code layer shortens time-to-market for simple products, but plan for 25% extra infra cost to run redundant pods. Accept that regulators will demand model documentation. | Choose InsurStack when you sell catastrophe-exposed property lines and need real-time cat models. The 1.2-point loss-ratio improvement justifies the 8 bps fee, but only if your actuaries are comfortable with a black-box model. Otherwise, the explainability gap will haunt you during rate filings. | Choose PulseEmbed when your distribution is multi-region and you cannot tolerate eventual consistency. Its Go micro-services scale to 1,450 FNOL/min, but budget for two FTE Go developers and a Terraform pipeline. The HIPAA certification matters if you embed health or travel policies. | Never choose LegacyOne unless you are legally required to keep a legacy policy admin system alive. Model the 40% maintenance uplift and the 6-hour GDPR deletion window before you sign the contract. the real latency killer is still the actuary | Across all platforms, the single largest drag on embedded performance is the human in the loop. EmbedCo’s fastest carrier client still clocks 3.2 seconds p99 because. the underwriting team insists on reviewing every >$50k auto claim. CoverGenie’s 0.9-second claim drops to 2.4 seconds once the no-code rules are locked and the actuarial review queue fills up. | The platforms can move bits faster than regulators can move paper. Until actuaries learn to trust real-time ML decisions, the 1.8-second Lemonade benchmark will remain out of reach for everyone except the vendor that owns the entire stack end to end. |
| Was this article helpful? Comments. |