Ankit Goyal

Two decades building enterprise data and AI infrastructure, the last twelve at Google. I build the layer where policy and meaning are authored once in business language and compiled down to whatever each system underneath enforces — first for access to sensitive data, then for data quality, and now for what AI agents are allowed to run on. I hold the manager’s instruments alongside the product: promotion calibration, discretionary budget and the compensation envelope.

↓ Download PDF

$731M+Audited impact

2.5B+Queries secured

35 → 88%AI response accuracy

~10 → 150+People, across six sites

Career Highlights

Both semantic layers, twenty years apart. Built the metrics semantic layer at Oracle — OBIEE and the RPD, business definitions mapped to physical tables and compiled to SQL at runtime — and shipped the business-facing dashboards and packaged consumer reports on top of it as a product. Built the ontology layer at Google: the Enterprise Knowledge Graph behind agent grounding.
Complex-query accuracy, 35% to 88%. Added governed knowledge-graph similarity as a third retrieval signal beside semantic and keyword rather than replacing either — the graph resolves which entity a question is about so the embeddings have something correct to be similar to. Golden prompt sets and an evaluation harness existed before the result and ran as a release gate; scoped to complex queries only.
Built the org, then the succession. On the leadership team through growth from ~10 to 150+ across six sites, restructuring roughly every two years. Handed the platform I founded to a PM I hired, staging the transfer so decisions visibly came from him — promoted to senior PM and given a team of his own.
Audited impact from the Data Security Framework. Founded as a 20% project, now a Google-wide standard. $731M+, signed off by Google Finance in 2024 — on a two-dimension cost-avoidance and productivity model I built myself and took to Finance for independent review.
Chose the layer, and refused the adjacent ones. Positioned DSF as the policy administration point rather than building another access-control system against the 80–90 already running, and kept it out of the live query path — policy propagated in batch and near-real-time instead, so the platform never sat between a user’s query and their data access check, and no separate SRE organisation was required to run it.

Experience — Google · Mountain View, CA · Aug 2014 – Present

Oct 2025 – now

Group Product Manager — Enterprise Data Platform & AI-Ready Data

Built the headless context engine layer. Exposed multi-hop entity resolution, reasoning, inference and contract verification as services consumed by an orchestrator rather than a UI, so any agent runtime could ground against the same governed layer. Domains attach their own entities to a thin core, so the tool surface grows without a central bottleneck.
Defined AI-ready data and made it enforceable. Asked in an executive review which of five launching agents would run on AI-ready data, answered none of them, and built the mechanism: business outcomes mapped to data products mapped to consuming agents, with contracts checked in the reasoning layer at query decomposition. Cut manual data preparation and time-to-deployment for new AI features by 80%.
A temporal layer over the graph. Approval and exception history from correspondence, ticketing and incident records, added as a time axis so agents could answer point-in-time questions directly rather than a person reconstructing them from source systems.
Designed the five-layer enterprise data platform — business entry and orchestration interface, agentic capabilities (entity resolution, multi-hop reasoning, the context graph), federated data domains, foundational services (catalog, quality, security, RAG), hybrid infrastructure — with contracts as the interface between layers rather than documentation beside them.
Composed enterprise context with temporal and personal context — episodic and short/long-term memory, personalised graphs holding individual, team and organisation preferences, and carried-forward session context, resolved across three teams as a federated substrate with personalisation opt-in and disableable.
Lead a top company-wide OKR. AI-ready data, with monthly status into the most senior leadership forum and biweekly updates to org leaders — investment framed against whether it beats the alternative use of the same people.
Platform GTM, internally and externally — ran adoption as a forward-deployed motion with domain partners across 10+ business units, hardened into reference architectures presented to Google Cloud enterprise customers and as keynotes at major data infrastructure summits.

Nov 2023 – Oct 2025

Group Product Manager — Enterprise Data Frameworks & Governance

Security at scale. Grew the Data Security Framework (DSF) to protect 20,000+ sensitive 1P and 3P data assets and secure 2.5B+ queries, evolving the measure as each stopped discriminating: assets protected, then queries under fine-grained policy, then audited financial impact.
Shipped agents that act, not just answer. Replaced static access policies with agent-managed access and policy packages over compensation and financial data, cutting stewardship toil by 80% on the most sensitive data in the company. Architected the first SAP-to-first-party policy integration, so policy travelled with the data across a hybrid-cloud finance transformation.
Enterprise Data Catalog. Incubated to product-market fit and adopted across 10+ business units with 2,000+ certified assets, agents, semantic models and business processes, each carrying a fit-for-purpose data contract tiered by use — daily refresh for analytical consumers, millisecond-SLA APIs for transactional ones.
One authoritative connector per third-party vendor. Replaced per-domain hooks into the same SaaS systems with a single owned integration surface per vendor, serving external agents reading in as well as internal consumers.
Killed my own product when it converged on another team’s. Ended our catalog’s discovery effort and pivoted to metadata enrichment writing back into the infrastructure team’s surface. Zero attrition; the competing team became its strongest advocate.

Apr 2020 – Nov 2023

Senior Product Manager — Data Security Framework

Converted laterally from the analyst ladder into product management at the same level, on the strength of DSF.

Founder and primary driver. Conceived DSF as a 20% project in 2015 after hand-writing access filters on sensitive compensation data, built the business case, and drove it from prototype to Google-wide launch in 2017 — establishing define security policy once, apply it everywhere as a company standard. Migrated existing 20% users onto the platform at launch, so day one had active users rather than an empty funnel. Corporate Engineering & Product Excellence Award (2020).
Governance that produced evidence, not just enforcement. Built impact analysis and approval workflows into every policy change, so a single edit could not cause mass over-sharing or a data outage — and made the resulting decision and access record the audit trail for internal and external review. Interoperated with existing policy stores, including SAP GRC.
Architecture that held for a decade. The first design computed each user’s filter set, hashed it and wrote the hash as a column — fast reads, prohibitive write amplification across millions of rows. The second decoupled user from policy via roles held in a separate system; it is still in production and it is what made business-language policy expressible at all.
Won the positioning argument repeatedly. Defended DSF’s scope to an incoming director or VP roughly every two years on the same ground: one authored policy across many endpoints means incumbent access-management teams manage fewer surfaces, not more.

Aug 2014 – 2020

Business Systems Analyst → Staff Business Systems Analyst (L4–L6), Enterprise Finance Systems

Consolidated fragmented opex and capex systems into a single authoritative spend-analysis platform, still in production twelve years later. Owned the hand-written access filters on payroll and compensation data — the high-risk, low-reward work DSF was invented to eliminate.

Prior Experience

2009 – 2014

TEKsystems Global Services · Engineering Manager

Led delivery and practice teams for Fortune 500 clients under commercial contract; owned BI solution roadmaps across client engagements and partnerships with Oracle, Tableau and Informatica. Built business semantic layers where metrics are defined once in business language and compiled to physical SQL at runtime — the metrics-layer ancestor of the ontology work above.

2006 – 2009

Oracle · Senior Software Engineer

Led development for Oracle Manufacturing Operations Center (BI/ERP) — built the business semantic layer and delivered the business-facing dashboards and packaged consumer reports on it as product; ran customer POCs to drive North America adoption.

2005 – 2006

Infosys · Software Engineer

Product development for Finacle, Infosys’s flagship core banking platform.

Organization & People Leadership

Performance & promotion. Member of leadership promotion committees; run calibration sessions grading individuals and evaluate packets against the leveling bar across the organisation.
Compensation & budget. Own the performance-review compensation envelope — stock, bonus, equity and base increments by individual against full-year performance — for my own team and, jointly with a peer manager at my level, for the broader organisation we run together. Own budgets for travel, professional services and vendor management.
Org design. Method for a major restructure: draw the target platform architecture as if it were one team, place the best talent against each layer, structure teams around that, then minimise the disruption of getting there. Two remote-site models tried across six sites; neither clearly won.
Hiring & pipeline. 200+ interviews conducted; converted three rotators onto the PM ladder; worked with recruiting to hold a bench of pre-approved candidates so open roles could go straight to offer without a fresh loop.
Performance management. Moved a vendor lead off the account and reassigned a strong PM off a product that did not suit them, with an improvement plan and an escalation path ready in both cases.
Also practised. Succession planning, cross-org conflict resolution, executive influence without direct authority, portfolio consolidation without scope loss.

Core Competencies

Agentic data platform — enterprise knowledge graphs, metrics and semantic layers, entity resolution, temporal modelling, data products and contracts, federated / data-mesh architecture, catalog, lineage, data quality.

Agent tooling & orchestration — grounding and context engineering, NL2KG, hybrid and graph retrieval, RAG, service interfaces for agent runtimes, MCP, skill and workflow orchestration.

Evaluation & AI trust — eval design and golden sets, hallucination mitigation, release gating on accuracy and coverage, policy-as-code, audit trail and compliance, shift-left governance.

Hands-on — Python, SQL; builds and ships personal multi-agent systems with a declarative policy layer.

Education

2019	Product Management Certification, UC Berkeley
2001 – 2005	B.Tech, Information Science and Engineering, Visvesvaraya Technological University