Business Analyst, AI & Enterprise Data Products

Business Analyst, AI & Enterprise Data Products

Duration: 12+ Months

Location: Remote

Work Authorization: USC

Position Summary

We are seeking a Business Analyst to define requirements for an AI-driven application serving internal business functions across Quality / Manufacturing / Clinical / Regulatory / Commercial.

The application sits on top of a substantial data layer, drawing from multiple enterprise source systems, and uses machine learning for surface insights / generate draft content / classify records / recommend actions for internal users.

This role is a hybrid of classic business analysis and data product ownership. You will spend as much time on data lineage, source-system semantics, and model evaluation criteria as you will on user workflows.

Success requires, someone comfortable writing requirements for systems that produce probabilistic rather than deterministic outputs, and who understands that in a regulated environment, an AI recommendation is only as defensible as the data provenance and audit trail behind it.

This is an internal-facing role. Your customers are our own colleagues.

Key Responsibilities

AI Solution Requirements

• Define and document the intended use of the AI capability — what decision it informs, who acts on the output, and what the consequence of an incorrect output is. This statement drives the risk classification and validation rigor for the entire system.
• Write acceptance criteria appropriate to probabilistic systems: performance thresholds (precision, recall, F1, or task-appropriate metrics), confidence thresholds, and defined behavior for low-confidence, out-of-scope, and no-answer conditions.
• Specify human-in-the-loop requirements — which outputs require human review before action, how reviewers are presented with model rationale and source evidence, and how reviewer overrides are captured and fed back into evaluation.
• Own ground truth curation: partner with SMEs to define labeling guidelines, coordinate annotation of evaluation datasets, adjudicate disagreements, and maintain the golden dataset used to benchmark releases.
• Define explainability and traceability requirements — for any given output, users and auditors must be able to identify the source records, model version, and data snapshot that produced it.
• Specify guardrails and failure modes: prohibited outputs, escalation paths, graceful degradation when upstream data is stale or unavailable, and user-facing communication of system limitations.
• Define monitoring requirements for model drift and performance degradation, including retraining triggers and the change control path a retrained model must follow.
• Partner with data scientists and ML engineers to translate business objectives into measurable model objectives, and to translate model performance back into business impact leadership can act on.

Data Layer & Integration Requirements

• Profile source data hands-on to establish what is actually available, at what quality, and at what refresh cadence — before committing to requirements that depend on it.
• Produce source-to-target mappings, data dictionaries, and canonical definitions for business entities that carry different meanings across source systems.
• Document data lineage and provenance end-to-end, from system of record through transformation to model input to user-facing output.
• Define data quality rules and thresholds across completeness, accuracy, consistency, timeliness, validity, and uniqueness; specify system behavior when thresholds are breached.
• Specify integration requirements: APIs, batch vs. event-driven patterns, latency and volume expectations, error handling, and reconciliation.
• Define master and reference data requirements, including entity resolution rules where records must be matched across systems.
• Specify handling of structured and unstructured content (batch records, deviation narratives, protocols, regulatory correspondence), including chunking, indexing, and retrieval requirements where applicable.
• Define access control, data classification, retention, and de-identification requirements in partnership with Privacy, Security, and Legal.

Compliance & Validation

• Work within the company''s CSV/CSA framework and GAMP 5 (Second Edition) principles, applying risk-based validation proportional to patient safety, product quality, and data integrity impact.
• Contribute to AI-specific risk assessment: model risk, data risk, and the incremental risk introduced by automation of a previously manual judgment.
• Ensure AI outputs and the human review of them meet 21 CFR Part 11 audit trail and electronic records expectations and uphold ALCOA+ data integrity principles.
• Author and contribute to validation deliverables: URS, validation plans, IQ/OQ/PQ protocols, traceability matrices, and summary reports.
• Support alignment with applicable AI governance expectations, including internal AI policy, the NIST AI Risk Management Framework, and evolving FDA guidance on the use of AI in regulatory decision-making.
• Prepare change controls and support internal audits and health authority inspections, including explaining system logic and evidence to inspectors.

Stakeholder Management & Adoption

• Serve as primary point of contact for assigned business functions; build relationships with process owners, SMEs, and super-users.
• Facilitate requirements sessions, design reviews, and demos across technical and non-technical audiences.
• Set realistic expectations about AI capability — actively manage both over-trust and under-trust in model outputs.
• Develop training materials, job aids, and SOPs covering not just system mechanics but appropriate reliance on AI outputs.
• Coordinate UAT, including evaluation of model quality by business SMEs, not just functional testing.
• Track adoption, override rates, and user feedback; translate patterns into a prioritized enhancement backlog.

Required Qualifications

• 7+ years of business analysis experience, including at least one delivery involving analytics, machine learning, or data-intensive applications.
• Working SQL proficiency — able to independently profile data, validate assumptions, and investigate discrepancies without waiting on an engineer.
• Demonstrated ability to write requirements and acceptance criteria for systems with non-deterministic outputs.
• Practical understanding of data modeling, source-to-target mapping, and integration patterns.
• Foundational understanding of machine learning concepts: training vs. inference, evaluation metrics and their tradeoffs, overfitting, drift, and the distinction between correlation and causation in model outputs.
• Experience in pharmaceutical, biotech, medical device, or another FDA-regulated industry.
• Experience validating or supporting a GxP-relevant AI/ML system, or familiarity with GAMP 5 Second Edition guidance on AI/ML.

Preferred Qualifications

• Familiarity with generative AI patterns where applicable: retrieval-augmented generation, prompt design, embedding and vector search, evaluation of open-ended outputs, and hallucination mitigation.
• Familiarity with predictive ML patterns where applicable: classification and regression, feature definition, class imbalance, and threshold tuning against business cost of false positives vs. false negatives.
• Hands-on exposure to modern data platforms: Snowflake, Databricks, Azure Synapse, dbt, or equivalent.
• Experience with enterprise source systems common in life sciences — Veeva Vault, SAP, LIMS, MES, TrackWise, Salesforce, ServiceNow.
• Familiarity with data governance tooling and concepts: catalogs, data contracts, stewardship models, lineage tracking.
• Working knowledge of HIPAA, GDPR, and PHI/PII de-identification practices.
• Certification such as IIBA CBAP/CCBA, PMI-PBA, CSPO, or a recognized data/AI credential.

Core Competencies

• Comfort with uncertainty — can specify, test, and communicate a system that is right most of the time rather than always, and can hold the line on what "good enough" means.
• Data skepticism — verifies what the data actually contains rather than accepting documentation or stakeholder description at face value.
• Analytical rigor — decomposes ambiguous problems into structured, measurable requirements.
• Compliance mindset — treats provenance, traceability, and documentation as part of the design, not overhead.
• Influence without authority — aligns competing stakeholder priorities toward a workable outcome.
• Ownership — follows issues to closure across organizational and vendor boundaries.