Product Manager

About the role

Harper is an AI-native commercial insurance company in San Francisco. We are not bolting AI onto insurance - we are rebuilding the entire business as software, on a simple bet: turning expert human judgment into compute is one of the largest transitions left to make, and a trillion-dollar industry still run 90 percent by hand is the place to prove it. The company has grown approximately 100 times in the last year and that pace continues; work happens on-site, in person, with long days and very high standards. Almost no one joins Harper for insurance; they join to build the company that replaces how it works. This role is for a Product Manager who will own a module of the business end-to-end, from the customer experience and operator workflows down to the AI agents that power them. You will translate nuanced commercial risks into encoded rules, prompts, and data structures, working alongside engineers and operators to make the system behave as well as a human in most places and better than a human in many. You will be responsible for moving key metrics, iterating on evals and backtests, and building the data flywheel that makes reliable autonomy possible. This is a deep, hands-on position where you run one surface fully, then move on to the next once you have mastered the current one.

Key facts

What you'll do

• Own the KPIs for your module, including conversion, handle time, accuracy, and autonomous-resolution rate, setting targets, instrumenting behavior, and driving movement on those metrics.
• Encode the specific nuances of your module's customers into rules, prompts, agents, and data structures so that the system reflects real commercial insurance complexity.
• Build and maintain a rigorous eval regime with regressions on every change, evals mapped to real outcomes, backtests against historical applications, and call-by-call review where it matters.
• Create a data flywheel by partnering closely with data labeling and validation teams to define and build golden datasets that train your module's models.
• Design a coherent cross-modal experience across web, voice, and human touchpoints, deciding where each modality excels and how handoffs and on-ramps should feel.
• Live with the operators in sales, service, underwriting, and ops, observing the work in real time to uncover problems before they are reported.
• Talk directly to customers every day, conducting real conversations rather than relying on aggregated reports or quarterly summaries.
• Prototype rapidly using tools such as Claude Code, Cursor, and Lovable, arriving in meetings with working demonstrations instead of static decks.
• Hyper-prioritize among competing requests, identifying the few initiatives that meaningfully move your KPIs and defending the decision to ignore the rest.
• Own the full lifecycle of experiments, from hypothesis through measurement, ensuring that learnings feed directly into product and model improvements.
• Partner with engineering to translate AI capabilities and constraints into product decisions that operators and customers can rely on.
• Surface insights from operator workflows and customer calls to refine routing, eligibility, and pricing logic within your module.
• Maintain a paranoid focus on silent regressions in probabilistic systems, using evals and real-world outcomes as leading indicators of trouble.
• Define what "right" looks like for data and behavior, then socialize and enforce those standards across labeling, evaluation, and production.
• Act as the primary decision-maker for your module, balancing speed, accuracy, and risk in alignment with company-wide priorities.

Requirements

• 1-3 years in product, or an early-career operator, engineer, or AI researcher who has been doing the work without the title.
• Demonstrated end-to-end ownership of a product or system, including KPIs, roadmap, and execution, with a track record of going deep on a domain and encoding what you learned into a system.
• You understand what an AI services company is: we do the work and sell the outcome, which means you ship behavior into a probabilistic system that real operators and customers must trust.
• You are obsessed with evals, preferring to ship a worse model with a great eval harness than the reverse, and you think in KPIs such as reducing handle time by a measurable percentage rather than merely shipping features.
• You can build using tools such as Cursor, Claude Code, and Lovable, and you can discuss AI tradeoffs including agents, LLMs, context engineering, data pipelines, and evals with the engineer who writes the code.
• You go deep before you go wide, choosing to own a specific thing rather than coordinating others' work, and you want a surface area you can truly master.
• If "PM" sounds like meetings, this is not the role for you; here the position involves direct building, testing, and system-level responsibility.
• You are comfortable working on-site in San Francisco during Monday-Friday hours that typically run from roughly 5 AM to 8 PM.

Nice to have

• Experience in AI or ML products, voice AI, agent frameworks, or workflow automation.
• Background in eval, prompt, or context engineering.
• Prior experience in insurance, fintech, or other regulated industries.
• History working at a startup where you wore multiple hats and moved quickly.

Practical notes

• On-site presence in San Francisco is required, Monday through Friday, with hours roughly from 5 AM to 8 PM.
• The learning curve is steep and the hours are long because the company moves at a fast pace and hires early-career PMs on purpose.
• Compensation details are listed as $125,000-$170,000 base plus performance bonus and equity.
• This is a full-time position based in San Francisco.