Accelerated Underwriting rPPG: 5 Metrics Carriers Should Track

Accelerated underwriting rPPG metrics keep coming up in carrier strategy meetings, and the conversations tend to loop. Everyone agrees paramedical exams are a bottleneck. Everyone has seen a demo of camera-based vitals. But when it comes to measuring whether an rPPG integration is actually working after go-live, most carriers default to the same two words: eligibility rate and accuracy. Those are fine starting points. They are not enough. The metrics that separate a program generating real underwriting lift from an expensive pilot that quietly gets shelved are more specific, and the industry has been slow to define them.

"We observed a clear tradeoff between acceleration and placement rates; carriers with the lowest acceleration rates reported the highest placement rates, and vice versa." — Munich Re US Life, Accelerated Underwriting Trends Survey, 2024

Why Standard Underwriting KPIs Fall Short for rPPG Programs

Traditional accelerated underwriting programs run on MIB checks, prescription histories, motor vehicle records, and credit-based insurance scores. The metrics built around those data sources — straight-through processing rate, average cycle time — still matter. But they miss what makes rPPG different.

rPPG introduces a real-time biometric signal into the underwriting workflow. That signal behaves unlike anything else in the evidence stack: it degrades with poor lighting, varies with skin tone if the algorithm was not trained on diverse data, and depends on applicant cooperation during a 30-to-90-second phone scan. You cannot evaluate an rPPG integration using the same dashboard you built for electronic health records. The failure modes are different. The metrics need to catch up.

Gen Re's 2025 Individual Life Next Gen Underwriting Survey found that 86% of approved applications going through automated workflows were ultimately placed, compared to 78% for accelerated workflows and 63% for fully underwritten paths. Those numbers tell a story about the relationship between process speed and policyholder follow-through. But they tell you nothing about whether the biometric data feeding those decisions was any good.

Five metrics that actually tell you something useful:

1. Signal Quality Acceptance Rate

This is the percentage of rPPG scans that produce a usable signal on the first attempt. It is probably the most important operational metric for an rPPG underwriting integration, and it is the one carriers most often forget to track.

A scan can fail for several reasons: the applicant moved too much, the room was too dark, the phone camera was smudged, or the applicant's face was partially obscured. Some of these are recoverable with a second attempt. Some are not. The signal quality acceptance rate tells you how often the technology works without friction.

Signal Quality Tier	Definition	Target Range	What It Means
First-pass acceptance	Usable signal on first scan attempt	>85%	Technology works without applicant frustration
Second-pass acceptance	Usable after one retry	>95%	Acceptable with retry prompt in the flow
Total failure rate	No usable signal after all attempts	<2%	Applicants who must fall back to traditional evidence
Partial signal	Some vitals captured, others degraded	<10%	May indicate lighting or motion issues

A 2025 study published in PMC enrolled adult volunteers and captured rPPG data using front-facing mobile cameras for approximately 90 seconds per session. The researchers found that signal quality correlated strongly with ambient lighting conditions and subject motion, with head movement during recording being the primary source of signal degradation. Carriers running rPPG in an uncontrolled home environment should expect lower first-pass rates than what vendor demos show under studio conditions.

The MMPD benchmark dataset, developed at the University of Washington, specifically tests rPPG algorithms across diverse skin tones (Fitzpatrick types I through VI) and real-world smartphone conditions. Top-performing algorithms achieve mean absolute error below 2 bpm on controlled datasets like UBFC-rPPG, but that number climbs to the 4-7 bpm range on MMPD. Carriers should ask vendors what their signal quality acceptance rate looks like on diverse, uncontrolled data — not just lab results.

2. Biometric Concordance Rate

This metric measures how often the rPPG-derived vitals agree with traditional evidence when both are collected on the same applicant. During the pilot phase, many carriers run rPPG in parallel with existing evidence gathering. The concordance rate tells you whether the new data source is producing underwriting decisions that align with what traditional methods would have produced.

A concordance rate that is too low suggests the rPPG signal is unreliable or the risk classification thresholds need recalibration. A concordance rate that is 100% raises a different question: if rPPG never changes a decision, why are you collecting it?

The useful range is somewhere between 90% and 97%. Below 90%, you have a data quality problem. Above 97%, you might have a threshold problem — the rPPG data is not being weighted heavily enough to influence outcomes, which means it is adding cost without adding information.

Concordance Scenario	What It Tells You	Action
rPPG agrees with traditional, applicant placed	System working as expected	Monitor for drift
rPPG disagrees, traditional evidence overrides	Potential rPPG calibration issue	Review disagreement patterns by vital sign type
rPPG disagrees, rPPG turns out correct post-issue	rPPG adding genuine value	Document for actuarial review
rPPG agrees with traditional, applicant not placed	Both methods caught the risk	Confirms redundancy value

Research published in the Journal of Clinical Monitoring and Computing (2022) evaluated camera-based respiratory rate measurement across 963 hospital patients and found 96% agreement with manual clinical counting. That kind of concordance study, run against actual clinical reference rather than another estimation method, is what carriers should demand from vendors before moving past pilot stage.

3. Acceleration Eligibility Lift

Munich Re's accelerated underwriting survey data shows that the average eligibility rate across carriers sits around 59% of individual life applications. The acceleration rate — how many of those eligible applicants actually get accelerated — has plateaued in recent years despite rising eligibility. There is a gap between being eligible for acceleration and actually receiving it, and that gap often comes down to missing or insufficient health evidence.

rPPG can close that gap by providing instant biometric data for applicants who would otherwise need to schedule a paramedical exam or wait for electronic health records. The metric to track is the incremental lift in acceleration rate attributable to rPPG data availability.

Calculate it as: (acceleration rate with rPPG available) minus (acceleration rate without rPPG, same population segment). If rPPG is not meaningfully lifting the acceleration rate, the integration is not justifying its cost.

The Gen Re 2025 survey found that carriers with higher automation levels reported better throughput but not necessarily better placement. The implication is that speed alone does not convert applicants to policyholders. The biometric data needs to be accurate enough to support confident underwriting decisions without triggering excessive fallout to full underwriting.

4. Demographic Consistency Index

This is where the conversation gets uncomfortable, and where carriers need to be most rigorous. rPPG algorithms extract pulse signals from facial video by detecting color changes in skin caused by blood flow. The physics works differently across skin tones. Melanin absorbs more light in certain wavelength ranges, which can reduce signal-to-noise ratio for individuals with darker skin. An rPPG system that performs well on Fitzpatrick I-III skin types but degrades on IV-VI has a fairness problem that is also, eventually, a regulatory problem.

The demographic consistency index measures the variance in signal quality and measurement accuracy across demographic groups: age, sex, skin tone, and BMI range. A well-calibrated system should show minimal variance across these groups.

Demographic Factor	Why It Affects rPPG	What to Measure
Skin tone (Fitzpatrick scale)	Melanin absorption affects signal-to-noise ratio	MAE by Fitzpatrick group; signal failure rate by group
Age	Skin elasticity and peripheral perfusion change with age	Measurement bias by age band (under 40, 40-60, over 60)
BMI	Facial adiposity can alter light reflection patterns	Signal quality acceptance rate by BMI range
Sex	Hemoglobin concentration differences between males and females	Bland-Altman bias by sex

A 2025 PMC study examining rPPG-derived cardiovascular parameters reported mean absolute differences of 2.69 mmHg for systolic blood pressure and 0.16 mmHg for diastolic blood pressure against standard cuff measurements. Those are strong numbers — but the critical question is whether they hold across the full demographic spread of an insurance applicant pool, not just the study population.

The NAIC and state insurance departments have been increasing scrutiny of algorithmic bias in underwriting. Colorado's SB 21-169 requires insurers to test for unfair discrimination in algorithms used for underwriting decisions. An rPPG system with demographic performance gaps is not just an accuracy problem; it is a compliance exposure.

5. Fallback-to-Traditional Rate and Reason Distribution

When rPPG fails to produce usable data or produces results outside of confidence thresholds, the applicant falls back to traditional evidence gathering. The fallback rate itself matters, but the distribution of reasons for fallback matters more.

If 80% of fallbacks are due to poor lighting, that is a UX problem you can solve with better scan-flow instructions. If 80% of fallbacks are due to the algorithm failing on certain demographic groups, that is a model problem requiring retraining. If fallbacks cluster around specific vital signs — blood pressure measurements failing while heart rate succeeds — that tells you which measurement channels need improvement.

Fallback Reason	Category	Typical Fix
Low ambient lighting	Environmental	In-app lighting guidance; flashlight prompt
Excessive motion	Behavioral	Clearer instructions; progress indicator showing stillness quality
Signal quality below threshold	Technical	Algorithm improvement; camera hardware requirements
Demographic performance gap	Algorithmic	Model retraining on underrepresented groups
Applicant refused to scan	UX/Trust	Better consent flow; explanation of how data is used
Device incompatibility	Technical	Expand supported device list; minimum camera spec requirements

Track these weekly during pilot, monthly in production. The distribution should shift over time as you address the most common causes. If the distribution is not shifting, you are not learning from the data.

Current Research and Evidence

The research base for rPPG has grown a lot since 2023. A few findings that matter for carriers:

Bland-Altman analysis, the gold standard for clinical measurement agreement first published by J. Martin Bland and Douglas Altman in The Lancet (1986), remains the right framework for evaluating rPPG accuracy. Carriers should be skeptical of vendors who report only correlation coefficients or mean absolute error without Bland-Altman limits of agreement. High correlation does not prove measurement agreement.

The Veeva clinical trial registry lists an active study specifically designed to validate rPPG-derived cardiovascular parameters — including blood pressure, heart rate, and cardiac workload — against standard clinical measurements, with a secondary aim of comparing rPPG-generated cardiovascular risk estimates (ASCVD risk, Framingham heart age) against clinician-calculated scores. Results from that trial, when published, will be directly relevant to insurance risk classification applications.

On the benchmark side, leading rPPG algorithms now achieve sub-2 bpm heart rate MAE on controlled datasets. The gap between controlled and uncontrolled performance remains the central research challenge, with most of the improvement coming from algorithmic post-processing rather than hardware changes. A 2026 study in Nature Digital Medicine showed that adaptive correction algorithms reduce Bland-Altman limits of agreement by 30-40% compared to raw signal extraction.

The Future of rPPG Metrics in Underwriting

These five metrics are a starting framework, not the final word. As rPPG matures in underwriting, longitudinal tracking will matter most: measuring the same applicant's rPPG-derived vitals over time and correlating them with actual mortality and morbidity outcomes. That is where the real actuarial value sits. But it requires years of data before it produces anything actionable.

For now, the carriers who get this right will be the ones who treat rPPG as a measurement system that needs ongoing calibration — not a feature that gets switched on and left alone. Dedicated analytics. Regular bias audits. A willingness to pull the technology back to parallel-run mode if the numbers start slipping.

Munich Re's survey data shows industry acceleration rates have plateaued. Getting past that plateau will take new data sources that are fast, scalable, and accurate enough to support confident decisions. rPPG can be that data source. But only if carriers track the right things.

Frequently Asked Questions

What is the most important rPPG metric for insurance carriers?

Signal quality acceptance rate. If the technology cannot reliably produce a usable signal from applicants scanning in their own homes, none of the downstream metrics matter. Carriers should target above 85% first-pass acceptance in real-world conditions before scaling beyond pilot.

How accurate does rPPG need to be for underwriting decisions?

For heart rate, a mean absolute error under 5 bpm against ECG reference is generally sufficient for risk classification, where heart rate is one input among many. For blood pressure, the ISO 81060-2 standard (mean error ≤5 mmHg, standard deviation ≤8 mmHg) provides a reasonable benchmark. Recent studies have reported rPPG blood pressure MAE of 2.69 mmHg systolic and 0.16 mmHg diastolic, which falls within those thresholds.

Does rPPG work equally well across all skin tones?

Not inherently. Melanin absorption reduces signal-to-noise ratio in darker skin tones, which can degrade accuracy if the algorithm was not trained on diverse data. Carriers must track their demographic consistency index and require vendors to demonstrate performance across Fitzpatrick skin types I through VI. The MMPD dataset from the University of Washington is a good benchmark to reference.

How should carriers handle rPPG scan failures?

Build a fallback workflow that routes applicants to traditional evidence gathering when the rPPG signal quality is below threshold. Track the fallback rate and, more importantly, the reason distribution. Most scan failures are environmental (lighting, motion) and can be reduced with better UX guidance. A total failure rate above 2% warrants investigation.

Solutions like Circadify are building rPPG measurement systems designed specifically for the uncontrolled conditions of home-based insurance applications, addressing the signal quality and demographic consistency challenges that generic rPPG solutions struggle with.