Pet Insurance Loss Ratios: Addressing the Behavioral Data Gap
Authors: Andrew Hellman, Jeff Pinard (Stride Systems)
Pet insurance loss ratios run roughly 14 points above personal lines. Approximately half reflects veterinary cost inflation; the remainder reflects an addressable behavioral data gap. This note quantifies the addressable component and its recoverable value.
Introduction
Pet lines is the fastest-growing segment in global P&C. Gross written premiums stand at roughly $20B today and are projected to reach $80B by 2033: a compound annual growth rate of ~17%. The canine policy sub-segment accounts for roughly 80% of the global book. North American policy penetration sits at just 4%. Demand trends are unambiguous.
The segment remains meaningfully under-capitalized relative to its growth trajectory. The principal reason for under-capitalization stems from pricing confidence. Pet lines loss ratios run 11 to 17 points (14 at the midpoint) above the personal lines average, a structural gap that makes aggressive participation economically irrational for disciplined capital allocators.
This note examines what is causing this gap, why one component of it is addressable while the other is structural, and what closing the addressable portion of the gap would mean for the segment's capital deployment trajectory and future growth.
1. Persistent Underperformance
Pet lines loss ratios sit in the 75 to 78% range against a personal lines benchmark of 61 to 64%. This persistent underperformance can be broken into two distinct components with very different addressability profiles. Roughly half of the gap is driven by veterinary cost inflation. Care costs have risen approximately 40% since 2020, propelled in part by private equity consolidation across the veterinary services market. This driver sits largely outside insurer control: vet data is fragmented across thousands of independent and roll-up practices, inconsistent in format, and structurally difficult to normalize at portfolio scale. Insurers can adjust pricing in response, but the input itself remains uniquely challenging and the current response largely passive.
The other half of the gap is a behavior observability problem which is uniquely addressable for the 80% of the global pet book related to canine policies. Decades of peer-reviewed clinical research into canine-human behavior have established clear, breed-type-specific behavioral patterning which becomes highly measurable within the right framework. Taken together with the following two structural features of current risk modeling, both the limitations of the current approach and the path to remediation become equally obvious:
Static inputs for a dynamic risk. Pet lines underwriters are currently asked to model a highly dynamic behavioral risk using static inception attributes and lagging claims events. Breed, age, ZIP code, and prior claims form the foundation. Everything that happens between inception and a loss event - the consistency of care, the stability of routine, the patterns that precede claimable conditions - goes unobserved.
The owner is an unpriced risk element. A dog is influenced by human behavior in a way no other insured asset is. The risk profile is fundamentally dyadic (pair-driven): the same animal carries materially different risk depending on the owner managing it. An experienced owner matched to a demanding breed and a first-time owner who chose the same breed on its appearance are two different risks wearing the same breed and age profile. Underwriting prices the animal and treats it as the whole, when the owner is the half that determines the outcome, and the half no inception attribute currently captures and no telemetry sensor can reach.
These two features compound: a dynamic risk modeled on static factors; policies rated on the animal without regard for the human half of the behavioral pair. This compound risk is the addressable portion of the pet lines loss ratio gap.
2. Why Current Inputs Are the Constraint
The pet lines underwriting stack is structurally identical to where personal auto sat before telematics, where homeowners sat before geospatial change detection, and where workers' comp sat before functional demand classification refined occupational pricing. In each of those lines, advanced analytics existed long before the inputs that allowed it to be predictive at portfolio scale. Compute investment ran ahead of data signal.
Pet lines is in the same position now. Reinsurers and primary carriers are investing heavily in AI and data science capabilities for the segment. Those investments are constrained by the inputs available to model against. When every competitor has access to the same static demographic factors, advanced models can only re-express the same limited information. Differentiation flattens. Marginal lift compresses. The compute investment becomes only incrementally accretive at best.
The mismatch is twofold. The risk is dynamic and the current inputs are static; the true risk is dyadic while current inputs capture only one side. Longitudinal behavioral inputs that capture how the dog-human pair actually behaves over the policy term would extend underwriting and portfolio monitoring beyond stated inception factors and claims events, and would sharpen insight into the dyadic risk itself. The observation gap along these two axes is precisely where the behavior that drives claims actually occurs. Instrumenting and measuring it would be, by definition, revelatory for 80% of the global pet lines market.
3. What a Behavioral Data Layer Looks Like
The data layer required to close this gap has specific structural properties derived from the observations outlined in Section 1. It is longitudinal rather than point-in-time since all behavioral drivers change over time by their nature. It is captured at a materially higher cadence than vet visits or loss events since dyadic optionality changes daily and not by imposed schedule. It is event-neutral, meaning capture occurs independently of injury, illness, or claims activity, which removes the negative-event bias that distorts claims-derived data. It is owner-mediated, providing the contextual layer that passive telemetry lacks and acknowledging that the risk is natively dyadic (an animal-human behavioral pair). It is structured for cohort-level aggregation, which is what makes it useful at portfolio scale and defensible under heightened governance expectations.
Practically, this means a set of behavioral signal classes that have direct analogs in adjacent insurance lines:
Engagement persistence and routine stability. Whether household care patterns hold steady or destabilize, the pet-lines equivalent of the engagement signals that drive UBI segmentation in personal auto.
Coverage and completeness. Explicit treatment of missingness as a signal in its own right.
Directional momentum and regime stability. Whether risk is improving, weakening, or shifting into a different pattern, mechanically analogous to credit-score trajectory in personal auto.
Context-attributed change. Separating explained shifts (household moves, family composition changes) from unexplained deterioration.
Stimulus, rest, and recovery patterns. The behavioral precursors that, in health insurance populations, have demonstrably preceded claimable events.
Read alongside the static segment keys insurers already use, these inputs make the dynamic and dyadic dimensions of pet lines risk observable for the first time. They complement existing underwriting inputs and are the missing layer that allows advanced models to do what they are built to do.
4. Sizing the Recovery Opportunity
Two paths lead to a defensible estimate of the loss cost recovery opportunity.
The first is top-down. The observed loss ratio gap of approximately 14 points (midpoint of range) may be split into roughly equal halves between vet cost inflation, which is structural, and the behavioral observability gap, which is addressable. A 7-point addressable gap against the global dog book through 2033 represents approximately $22B in recoverable loss cost system wide, before any premium growth or capital efficiency effects are layered on.
The second path is bottom-up, built from the research base in adjacent personal lines. Pet lines does not yet have its own validated longitudinal evidence, but the analog research is mature. The ranges below are drawn directly from that research, organized by the four underwriting workflow stages where longitudinal behavioral signals operationalize.
New Business Qualification. The question at policy bind is whether a given risk maps to a cohort whose observable patterns are consistent with responsible stewardship, or to one carrying elevated structural risk before the policy is written. Engagement persistence, routine stability, and breed-environment fit are the operative inputs, functioning in many ways like the engagement signals that segment usage-based cohorts in personal auto and the location-risk classification work that has refined homeowners pricing.
The analog research at this stage indicates uplift in the range of approximately 4 to 8 points. (Base case = 6).
Sources: Insurance Thought Leadership / Value Momentum (2023); Swiss Re & IoT Insurance Observatory consumer survey, 10,000 respondents (2023); North American Actuarial Journal, Vol. 26 No. 3 (2021); Verisk/ISO Geographic Information Solutions; OECD, Leveraging Technology in Insurance (2023), citing Arturo (2021).
Pricing and Segmentation. Once the risk is bound, the question shifts to whether the dog's activity and exposure profile challenges standard pricing in the relevant geography, particularly where local care costs are elevated. The mechanics here function in many ways like the functional demand classification logic that refined occupational pricing in workers' comp.
Analog research at this stage indicates uplift in the range of approximately 1 to 2 points. (Base case = 1.5).
Sources: NCCI, Class Ratemaking for Workers Compensation; Casualty Actuarial Society.
Mid-Term Monitoring. The question through the policy term is whether the cohort pattern is still consistent with how the risk was priced at inception, and whether behavioral or stress-accumulation patterns are signaling deterioration ahead of claimable events. The directional logic functions in many ways like credit-trajectory monitoring in personal auto and pre-event utilization detection in health insurance, both well-established as predictive of forward loss propensity.
Analog research at this stage indicates uplift in the range of approximately 4 to 7 points. (Base case = 5.5).
Sources: FTC, Credit-Based Insurance Scores: Report to Congress; EPIC Actuaries LLC, Relationship of Credit-Based Insurance Scores to Auto Insurance Loss Propensity; NEJM Catalyst, Corewell Health predictive model study (December 2023); Parkland Health predictive analytics program.
Renewal Reassessment. At renewal, the question is whether the risk has diverged sufficiently from its inception baseline to warrant rerating. The change-detection logic functions in many ways like the renewal-pricing work that has matured in property lines.
Analog research at this stage indicates uplift in the range of approximately 1 to 2 points. (Base case = 1.5).
Sources: OECD, Leveraging Technology in Insurance (2023), citing Arturo (2021); Verisk/ISO Change Detection product documentation.
Each stage is driven by composite modules built from underlying longitudinal behavioral inputs, packaged for direct model ingestion. The full architecture is detailed in the technical data memo - Longitudinal Behavioral Data for Pet Lines – available upon request.
In Aggregate. Summed across the four stages, the analog research points to a combined loss ratio uplift range of approximately 10 to 19 points, with a base case near 14.5. Two adjustments translate the raw analog evidence into a defensible aggregate. First, pet lines is a high-frequency, low-severity book where loss ratios are driven more by claim frequency than claim size, while the source analogs come from lines with different frequency and severity profiles. A 75% factor is applied across the stages as a deliberate hedge against transferability risk. Second, the resulting figures represent a directional ceiling on what longitudinal behavioral signals could contribute to loss ratio recovery, not a projection of what any individual insurer will achieve in deployment. After both adjustments, the cumulative loss ratio recovery range sits at approximately 7.5 points (Conservative case), 10.9 points (Base case), and 14.3 points (Aggressive case) over the 2027 to 2033 window, equivalent to roughly $19.3B, $28.0B, and $36.7B in cumulative recovered loss cost against the canine portion of the global pet lines book.
The bottom-up range exceeds the top-down estimate at every confidence tier. The analog evidence indicates that longitudinal behavioral signals carry more loss ratio recovery potential than the addressable portion of the gap requires, which is to say the methodology has headroom against the problem it is designed to address. The interactive Loss Ratio Recovery Simulation (simulation.stridesystems.io) allows reinsurers and carriers to model this analog methodology against their own participating pet lines book. The tool takes a participating dog GWP figure and surfaces projected loss ratio recovery across the relevant workflow stages, using the analog-derived ranges discussed above. It is designed to make the math tangible at the level of an individual treaty or portfolio.
5. What Operationalization Looks Like
A behavioral data layer is only useful if it integrates into the way underwriting actually works. That means signals delivered as features ready for model ingestion, de-identified, cohort-thresholded, versioned, and documented with the lineage required for governance review.
It also means organization around the underwriting workflow itself. Behavioral signals become operationally useful when they are packaged to answer specific questions at specific stages: new business qualification (does this risk map to a cohort with characteristics consistent with responsible stewardship?), segmentation and pricing (does the activity and exposure profile challenge standard pricing in this geography?), mid-term monitoring (is the cohort pattern still consistent with how the risk was priced?), and renewal (has the risk diverged sufficiently from its inception baseline to warrant rerating?).
6. Next
Three implications follow.
For reinsurers, the strategic question is whether to source behavioral inputs exclusively, embed them in treaty economics, and use them as the input layer that makes existing analytics investment differentiated rather than commoditized. It follows that capital will flow into the systemic solution that can provide returns against unique risk insight. The window in which this is available on an exclusive basis is, by definition, finite.
For primary carriers, the relevant question is which reinsurers are positioning to make behavioral signals available through treaty terms, and what that implies for competitive positioning over the next 24 to 36 months. The structural advantage accrues to carriers participating in programs that close the data gap ahead of the broader market.
For the pet lines segment as a whole, the capital deployment trajectory depends on whether pet lines economics can be normalized toward broader personal lines benchmarks. Even a partial recovery within the addressable range would shift the calculus for capacity providers currently constrained by the gap. A more structural recovery would change the segment's relationship to capital entirely.
The pet lines market will reach roughly $80B by 2033 regardless of whether the data gap closes. What the data gap determines is who participates in that growth confidently, at what margin, and with what durable advantage.
- -
Stride Systems builds longitudinal behavioral data infrastructure for pet lines underwriting.
For technical data memo materials, contact andrew@stridesystems.io.