Data Scientist

Most of what makes American healthcare expensive isn’t medical care. It’s the machinery wrapped around it: middlemen taking a cut, fraud nobody stops, and billing systems designed to fight over payment instead of deliver care. The result is higher premiums, denied claims, surprise bills, and a system patients increasingly experience as adversarial.

Arlo is rebuilding health insurance for small businesses from first principles: making sure as much of every premium dollar as possible goes to care instead of getting absorbed by the system around it. We do that by identifying fraud earlier, steering members toward higher-quality and lower-cost care, automating operational overhead, and eliminating vendors whose business exists mostly to take a cut.

AI is the foundation that makes this work. We use it across underwriting, operations, clinical programs, and member experience to build an insurer that becomes more efficient as the technology improves.

We’re already operating at meaningful scale: profitable, hundreds of millions in premiums, tens of thousands of members covered, and growing quickly through brokers, employers, and partners. Backed by Upfront Ventures, 8VC, and General Catalyst, with a team from Palantir, YC companies, and longtime healthcare operators.

Arlo quotes small businesses using AI-powered underwriting that predicts individual risk to determine group rates. But accurate pricing requires more than historical data alone — at the time of quoting, additional signals like prior rates, aggregate claims history, and self-reported health information can meaningfully sharpen our view of a group's risk profile.

This role sits at the center of four connected problems: learning from post-policy data to understand where our risk blindness lies, building automated quoting systems that act on softer signals, understanding how algorithmic changes ripple through to sales outcomes, and maintaining the integrity of the data pipelines that underpin all of it.

• Emerging risk detection. Analyze early-period claims data to surface patterns that indicate risk was underestimated at quoting — and quantify the gap.

• Blindness measurement. Build frameworks to systematically identify where the external claims database is incomplete or lagged, and estimate the magnitude of those gaps.

• Rate determination. Develop calibration factors that translate soft signals — based on competitive information, econometrics and emerging events — into rate adjustments layered on top of core model output.

• Feedback loop infrastructure. Create pipelines that carry post-quoting learnings back into upstream models so calibration improves continuously.

• Multi-signal fusion. Design models that ingest heterogeneous inputs and synthesize them into an enriched risk view that reduces reliance on manual review.

• Confidence scoring. Build the logic that decides when a quote can be issued automatically versus routed to a human, minimizing queue volume without increasing adverse selection.

• Threshold and rule design. Collaborate with underwriting to set and validate auto-issuance decision thresholds, and monitor their performance over time.

• Algorithm impact attribution. When we adjust rating algorithms, tighten auto-quoting rules, or change data quality requirements, measure the downstream effect on quote volume, win rates, and sales conversion — and distinguish model-driven changes from market-driven ones.

• Sales funnel diagnostics. Identify where quote kickouts, rate changes, or data submission friction are creating drop-off in the sales pipeline, and quantify the cost of each leakage point.

• Data quality incentive analysis. Understand whether brokers and groups that provide richer data at submission (more complete census, aggregate claims, prior rates) achieve better pricing outcomes — and help us make that case externally.

• Ingest delay profiling. Map the latency characteristics of our live policy data — claims, enrollment, eligibility — and identify which delays are structural versus operational, and where they introduce systematic bias into our models.

• Consistency monitoring. Build checks and alerting for data inconsistencies across our ingestion layer: duplicate records, mismatched member IDs, enrollment timing gaps, and reporting lags from carriers or TPAs.

• Upstream data partnerships. Work with engineering and data teams to document known data quality issues, prioritize fixes, and maintain a clear picture of what the data can and cannot reliably support at any given point.

• 3–5 years in a data science or quantitative analyst role

• Proficiency in Python (scikit-learn, pandas, statsmodels) and SQL

• Experience building and validating predictive models end-to-end

• Comfort working with messy, inconsistent real-world datasets

• Experience designing or auditing data pipelines for quality and consistency

• Ability to communicate model behavior and business impact clearly to non-technical stakeholders

• Background in insurance, actuarial data, or healthcare claims

• Experience with GLMs, survival analysis, or credibility theory for pricing

• Familiarity with group health underwriting or broker distribution models

• Experience building confidence-based routing or triage systems

• Exposure to sales funnel analytics or conversion attribution

• Familiarity with MLflow, dbt, or similar tooling

• You'll be an individual contributor embedded with underwriting, pricing, and sales teams — with direct access to the people who use your outputs daily.

• You'll own projects from data exploration through to production deployment and monitoring, without a separate ML engineering handoff layer.

• Work spans Python-based modeling and SQL-driven analysis equally; both matter.

• Your models directly influence pricing decisions and sales outcomes, so accuracy, explainability, and sound uncertainty quantification are all first-class concerns.

• You'll work closely with engineering on data ingestion questions — this role requires comfort sitting at the boundary between data science and data engineering.

• High ownership: You’ll get real responsibility from day one—our high-trust team empowers you to run with big problems and shape core parts of the company.

• Join an important mission: Your work directly influences how people access care and improves lives at scale.

• Growth & expansion: We’re moving fast, and as we grow, your scope will grow with us—new challenges, bigger opportunities, and rapid career velocity.

• Apply AI to a problem that matters: Instead of optimizing ads or cutting labor costs, you’ll use AI to fundamentally reimagine how people get healthcare.

• High pace, high collaboration: We operate with velocity, first-principles thinking, and a team that works closely, openly, and with ambition.

Exact compensation inclusive of salary and any bonuses is determined based on a number of factors including experience and skill level, location, and qualifications which are assessed during the interview process.
Arlo is an equal opportunity employer. We do not discriminate based on age, race, color, creed or religion, national origin, sexual orientation, gender identity or expression, military status, sex, disability, predisposing genetic characteristics, marital status, familial status, status as a victim of domestic violence, or arrest or conviction record, as defined under New York State law.
🔒 Your safety matters to us. If you're selected to move forward in our hiring process, you'll hear directly from a member of our Recruiting team via an @joinarlo.com email address. We will never ask for personal or financial information outside of our formal onboarding process. When in doubt, please reach out to us to verify at: [email protected].