Principal Product Manager, Person Data and AI Evaluations

Principal Product Manager, Person Data and AI Evaluations

ZoomInfo | Product | Core Data

About ZoomInfo

ZoomInfo is where careers accelerate. We move fast, think boldly, and empower you to do the best work of

your life. You'll be surrounded by teammates who care deeply, challenge each other, and celebrate wins.

With tools that amplify your impact and a culture that backs your ambition, you won't just contribute, you'll

make things happen, fast.

The Opportunity

ZoomInfo's Core Data team builds and maintains the person and company data that powers the Go-To-

Market Intelligence Platform: hundreds of millions of contact records, resolved to the right person at the right

company, kept accurate, and delivered to more than 35,000 customers.

That data has always been produced by deterministic pipelines: take many competing sources, weight them,

decay them, and select a winner per attribute. That model is being replaced. Core Data is moving to an

inference-default operating model, where an agent reads the full body of evidence for a record, applies our

business policy, proposes the answer, and a deterministic verification layer decides whether it lands.

Selection logic goes away; verification and evaluation are what remain. The product manager's job changes

with it. Instead of writing requirements for hand-built selection rules, you run the evaluation engine that

decides whether the agent's output is good enough to publish, and you make it better every week.

We are hiring a Principal Product Manager to own Person Data outcomes end to end and to build the AI

evaluation discipline that the whole Core Data team will run on. You inherit a live portfolio: person data

quality (removing bogus or outdated executive contacts, unlinking contacts from the wrong company, title

classification, email deliverability), person coverage and extraction, and the privacy roadmap for person data.

You will carry that portfolio through the pivot from rule-based selection to agent-adjudicated records, and

you will define how we know the new system is right.

This is a role for someone who has shipped data pipelines and entity resolution at scale and who is already

using AI every day to make data systems better. You should be as comfortable reading a golden set and an

eval report as you are writing a roadmap, and you should be excited that our product managers commit code

and work alongside agents as a normal part of the job.

What You'll Do

Own Person Data end to end. Set the strategy, roadmap, and monthly priorities for ZoomInfo's contact data:

accuracy, coverage, freshness, and compliance. Own the outcomes and the metrics that prove them, from

cleaning up bogus executive contacts and verifying employment through leadership extraction and person

location. Partner with the Person Data product manager already on the team and with Privacy, Trust and

Identity engineering to deliver the roadmap.

Build and run the AI evaluation engine. Define what "correct" means for each attribute the agent emits.

Curate golden sets and trap records with our research team, stand up LLM-as-judge gates, set the confidence

thresholds that route a record to auto-accept, auto-reject, or human adjudication, and track precision, recall,

and drift in production. Every model or prompt change to the person pipeline ships through the gate you own.

Drive the pivot from rules to agents. Move person attributes, one at a time, from deterministic selection to

inference over full evidence. Write the policy clauses the agent follows, the grading guides research uses to

score it, and the acceptance rules the pipeline enforces. Decide the order, prove each step in shadow mode

against the gold set, and retire the legacy logic when the numbers say it is safe.

Bring AI into the pipeline without tanking the data. Introduce models where they earn their place and keep

classical tooling where it wins: normalization, string similarity, registry lookups, deterministic rules. Reason

explicitly about cost per row, latency, determinism, and auditability. Know when a bigger model is the wrong answer.

Treat entity resolution as the hard core. Person-to-company matching, wrong-company links, duplicate

people, and identity across sources are the problems that make or break contact data. Bring a clear mental

model for canonical identity, match confidence, and the cost asymmetry between a false merge and a missed match.

Own the privacy roadmap for person data. Own the privacy roadmap for person data — suppression, opt-

out, and notification — with Legal and Privacy.

Build with the team, hands on. Prototype adjudication agents, evals, and analysis in code with the AI

engineer and data science. Commit to the shared repository. Use Claude and agentic tooling daily to inspect

results, tune prompts, and verify output so that no single absence blocks an iteration cycle.

Drive cross-functional execution and communicate strategically. Act as the hub between Person Data

engineering, Match, Research, Data Science, Web Data acquisition, Privacy, and the application teams that

consume contact data. Represent the roadmap and its rationale clearly to product and data leadership, and

write the release notes and customer-facing narrative with Product Marketing.

What You'll Bring

8+ years of product management experience, with meaningful time owning data products or data

platforms at scale

• Hands-on experience with data pipelines, ideally incremental or streaming rather than batch rebuilds,

and a working understanding of how records move from ingestion through processing to a served dataset

• Direct experience with entity resolution, record linkage, or matching systems: canonical IDs, match

confidence, survivorship rules, and the trade-offs between precision and recall

• Demonstrated, current use of AI to improve data systems, not just AI as a feature: you have brought

an LLM into a production data process and can describe how you kept accuracy from degrading

• Experience designing evaluations for model output: golden sets, precision and recall, LLM-as-judge,

human-in-the-loop routing, and production drift monitoring

Comfort working in code: you can read a pipeline, write a prompt and a validator, run an analysis in

SQL or Python, and commit alongside engineers

      Experience partnering closely with engineering and data science on complex, data-intensive systems,

and a track record of influencing technical direction without formal authority

• Strong bias for action and a demonstrated ability to drive focus and finish work in a fast-moving,

ambiguous, and frequently chaotic environment

• Excellent written and verbal communication with the ability to move between technical depth and executive narrative.

Preferred

Experience with contact, person, or identity data, or with B2B company data

• Familiarity with privacy regulation as it applies to published personal data (GDPR, CCPA, DNC, and

international notification regimes)

• Experience with Claude or comparable frontier models, agent frameworks, and eval harnesses
• Background at an AI research lab or an AI-native data company

Who You Are

AI-native, not AI-curious. You use AI tools every day and you have opinions about where inference beats

hand-built logic and where it does not. You would rather over-index on AI fluency and learn our data than the reverse.

An entity-resolution thinker. You see a data quality problem and immediately ask which entity it is about,

what the canonical form is, and what evidence would settle it.

An evaluator by instinct. You do not ship a model change without a gold set, a threshold, and a plan for what

happens at ten times the volume. "We'll A/B test it" is not an eval plan.

A builder. You prototype, you read code, you commit. You are comfortable being the person who inspects

agent output at nine in the morning and tunes the prompt by ten.

Tenacious. ZoomInfo is fast, distributed, and mes