New opportunity

Oliver Wyman – Associate Director - Data Science (AI and Generative AI) - Gurugram

Oliver Wyman Group · Gurugram, India

About this role

Oliver Wyman’s India Data and Analytics practice seeks an Associate Director to lead client-facing AI and Generative AI solution design, evaluation, governance and implementation. Use cases include knowledge assistants, document intelligence, workflow automation, decision support and advanced analytics. The role combines hands-on technical leadership with client communication, commercial judgment and team development.

Responsibilities

Lead AI, machine learning and GenAI workstreams from use-case discovery and feasibility assessment through prototyping, evaluation, deployment readiness and adoption support. Define architecture options, workplans, quality standards and success metrics; translate business needs into data, model, platform and governance requirements.
Guide development of RAG assistants, enterprise copilots, workflow agents, semantic search, summarization, extraction and prompt-driven analytics. Advise on commercial and open-source language models, small language models, vector databases, cloud AI services, integration patterns and build-versus-buy decisions.
Oversee prompt engineering, retrieval design, model adaptation, benchmarking, performance monitoring and remediation. Establish responsible AI, privacy, security, human-in-the-loop, auditability, MLOps, LLMOps and model-governance practices. Review code, designs, documentation and client deliverables; manage and coach junior colleagues and contribute to proposals and thought leadership.

Required qualifications

9 to 12 years of experience in AI/ML, data science, GenAI engineering, NLP, software or data engineering, advanced analytics, AI strategy, or related consulting or technology roles, including leadership of AI/GenAI or analytics workstreams.
A bachelor’s or master’s degree in a quantitative or technical discipline; an advanced degree is preferred. Strong knowledge of ML, NLP, LLMs, embeddings, RAG, agents, prompt engineering, fine-tuning, evaluation, monitoring and AI application architecture; hands-on Python and SQL proficiency.
Ability to assess data quality, reliability, hallucination risk, fairness, explainability and controls. Strong stakeholder communication, technical documentation, project management and coaching skills. Cloud platforms, APIs, orchestration frameworks, MLOps/LLMOps tools and large-scale data experience are advantageous.

Preferred experience includes enterprise AI delivery in regulated or complex environments, especially financial services; AI governance and model-risk controls; productionizing prototypes; and client-facing consulting. Valued tools include Azure OpenAI, AWS Bedrock, Google Vertex AI, Hugging Face, LangChain, LlamaIndex, MLflow, Databricks, Snowflake, Spark and Kubernetes. The Gurugram-based role involves global collaboration and travel when required. Marsh’s hybrid policy expects at least three days per week in the local office or onsite with clients, including a team anchor day.

For full details about this role, click “Apply on company site” to view the employer’s original posting.

Skills for this role

Generative AIMachine LearningData ScienceNatural Language ProcessingLarge Language ModelsRetrieval-Augmented GenerationAI AgentsPrompt EngineeringEmbeddingsFine-tuningModel EvaluationModel MonitoringResponsible AIAI GovernanceMLOpsLLMOpsPythonSQLAPIsCloud PlatformsVector DatabasesAzure OpenAIAWS BedrockGoogle Vertex AIHugging FaceLangChainLlamaIndexMLflowDatabricksSnowflake

Checking your profile…

Preparation suggestions, not the employer’s actual interview questions. Always check the original posting for current requirements.

Understand the role

What matters most

Lead AI workstreams end to end
Source-backed: Lead discovery, feasibility, solution design, prototyping, evaluation, deployment readiness and adoption support; define workplans and success metrics. Preparation suggestion: Be ready to explain how a workstream moves from a business use case to a measurable delivery plan.
Make defensible architecture choices
Source-backed: Guide RAG assistants, document intelligence and other GenAI applications; advise on models, platforms, integration patterns and build-versus-buy decisions. Preparation suggestion: Organize examples around the requirements and trade-offs behind a design choice, not just the tools used.
Own evaluation and governance
Source-backed: Oversee benchmarking, monitoring and remediation, and establish responsible AI, privacy, security, auditability and model-governance practices. Preparation suggestion: Be ready to connect quality measures and controls to a solution’s intended business use.
Translate technical work into client decisions
Source-backed: Partner with senior client stakeholders, turn business needs into technical and governance requirements, and contribute to proposals and practical recommendations. Preparation suggestion: Practice explaining architecture, limitations and value in business, risk and technology terms.
Maintain delivery quality while developing the team
Source-backed: Review code, designs, documentation and client deliverables; manage multiple workstreams and coach junior colleagues. Preparation suggestion: Prepare a clear account of how you set review standards, delegate work and resolve delivery risks.

Likely challenges to think through

Choosing capability without unnecessary complexity
Preparation hypothesis, not a confirmed employer problem: A client use case may benefit from a sophisticated agent or large model, while a simpler approach could be easier to integrate, evaluate and govern. Prepare to compare options against data quality, success metrics, integration needs and build-versus-buy considerations.
Moving from a promising prototype to a usable solution
Preparation hypothesis, not a confirmed employer problem: Prototype performance may not hold up under reliability, security, monitoring and adoption requirements. Prepare an example of how you would test limitations, define remediation and assess deployment readiness before recommending wider use.
Balancing hands-on depth with senior-level oversight
Preparation hypothesis, not a confirmed employer problem: Technical reviews and coaching can compete for time with client decisions, proposals and concurrent deadlines. Prepare to describe how you would prioritize risks, set quality checkpoints and give junior colleagues ownership without losing accountability.

Practice your answers

Technical practice

Suggested practice: How would you prioritize client AI use cases?

What to coverThe posting calls for identifying high-value opportunities and assessing feasibility. Explain how you would compare business value, data readiness, technical risk and adoption needs, then turn a selected use case into scope and success metrics.

Suggested practice: How would you design a RAG knowledge assistant?

What to coverThe role includes RAG assistants, embeddings, semantic search and retrieval design. Walk through document preparation, retrieval, answer generation and access controls, including how you would diagnose irrelevant or unsupported answers.

Suggested practice: How would you evaluate a document intelligence solution?

What to coverThe posting names summarization, extraction and document intelligence as client use cases. Describe how you would assess data quality and measure output accuracy against human-reviewed examples, including when human review remains necessary.

Suggested practice: When would you use a workflow agent rather than a simpler application?

What to coverWorkflow agents and enterprise copilots are among the stated solution types. Compare an agent with a more constrained workflow on task complexity, reliability, permissions and auditability; explain how you would limit unintended actions.

Suggested practice: How would you advise on model and platform choices?

What to coverThe role calls for comparing commercial, open-source and small language models, cloud services, vector databases and build-versus-buy options. Discuss quality, cost, latency, privacy, integration and operating burden, making clear how client requirements drive the choice.

Suggested practice: When would you change prompts, retrieval or the model?

What to coverThe posting covers prompt engineering, retrieval design and fine-tuning or adaptation. Explain how you would isolate the cause of poor results before selecting an intervention, and how you would test whether it improves performance.

Suggested practice: How would you benchmark a GenAI application?

What to coverModel evaluation, benchmarking, hallucination risk and business-use alignment are explicit responsibilities. Propose representative test cases and measures for answer quality, reliability and limitations, then explain how results would inform a deployment decision.

Suggested practice: What governance controls would you put around a client AI solution?

What to coverThe posting calls for responsible AI, privacy, security, fairness, explainability, human oversight and auditability. Describe how you would identify material risks, assign appropriate controls and document residual limitations for stakeholders.

Suggested practice: How would you move a prototype toward production readiness?

What to coverThe role spans prototyping, deployment readiness and adoption support, with production scaling listed as valued experience. Cover integration, testing, security review, monitoring, issue remediation and measures of adoption or value.

Suggested practice: How would you use Python and SQL to assess data readiness?

What to coverHands-on Python and SQL proficiency and data assessment are stated requirements. Explain how you would investigate completeness, consistency and suitability of source data, and how findings would affect solution design or feasibility.

Behavioural practice

Suggested practice: Describe a time you explained an AI trade-off to senior stakeholders.

What to coverThe role requires turning complex technical topics into practical business, risk and technology implications. Give a real situation, the recommendation and communication choices you made, and the resulting decision.

Suggested practice: Tell me about leading competing technical workstreams.

What to coverThe posting calls for managing deadlines, risks, stakeholders and delivery quality across workstreams. Use a real example to show the constraints, how you reprioritized or escalated, and what resulted.

Suggested practice: Describe a time you coached a junior colleague through a quality issue.

What to coverManaging and coaching colleagues while reviewing code, designs and deliverables are stated duties. Explain the situation, the feedback and support you provided, and the effect on the work or colleague’s development.

Suggested practice: Tell me about challenging an attractive but risky AI proposal.

What to coverThe role calls for commercial judgment alongside feasibility assessment and governance. Recount a real situation, how you presented evidence and alternatives, and the outcome for the client or project.

Suggested practice: Describe collaborating across teams to shape a client solution.

What to coverThe posting describes work with partners, consultants and senior clients, including proposals and global collaboration. Give a real example of differing priorities, your actions to align people, and the resulting solution or decision.

Sharpen your knowledge

Topics to prepare

ML, NLP and LLM approach selection
Suggested exercise: compare a conventional ML or NLP approach, a small language model and a larger LLM for a document-extraction use case; explain when adaptation or fine-tuning would be justified. The posting explicitly requires knowledge of these approaches and judgment about model choices.
RAG and agent architecture
Suggested exercise: sketch a knowledge-assistant architecture and explain how embeddings, retrieval, prompts and an optional workflow agent interact, including where errors can arise. The posting explicitly names RAG assistants, agents, retrieval design and AI application architecture.
Evaluation and reliability
Suggested exercise: define separate measures for retrieval quality, answer accuracy, hallucinations and business usefulness, then specify how findings would drive remediation. The posting explicitly calls for benchmarking, model evaluation, monitoring and assessment of solution reliability.
Responsible AI and model governance
Suggested exercise: map privacy, security, fairness, explainability, human review and auditability controls to a hypothetical client-facing assistant. These controls are explicit responsibilities; regulated financial-services experience is listed as preferred, not required.
Use-case prioritization and production readiness
Suggested exercise: compare two hypothetical client use cases on value, data feasibility, risk and adoption needs; outline build-versus-buy options and criteria for moving beyond a prototype. The posting explicitly covers discovery, commercial judgment, deployment readiness and adoption support.

Skills to revise

Python and SQL data assessment
Suggested exercise: use Python and SQL on a sample document-metadata table to identify missing fields, duplicates and coverage gaps, then summarize implications for an AI use case. Hands-on proficiency in both languages and data-quality assessment are explicit requirements.
RAG prototype review
Suggested exercise: build or inspect a small knowledge-assistant prototype; trace a query through retrieval and answer generation, then document two retrieval or prompt improvements. The posting explicitly calls for leading and reviewing RAG applications and experimentation.
Evaluation and monitoring plan
Suggested exercise: create a test set for a summarization or extraction workflow, record failure categories and propose monitoring thresholds and remediation steps. The posting explicitly includes benchmarking, performance monitoring and issue remediation.
Senior-stakeholder solution brief
Suggested exercise: write a one-page client brief with architecture options, success metrics, limitations, controls and a recommendation stated in business terms. The posting explicitly requires client-facing communication and senior-stakeholder-ready documentation.
Workstream planning and technical review
Suggested exercise: draft a phased workplan from discovery through adoption, with quality gates, risks and review criteria; then annotate a sample design for feedback to a junior colleague. The posting explicitly calls for managing delivery, reviewing technical work and coaching team members.

Your preparation checklist

Map the stated requirements to evidence
Prepare a one-page evidence sheet for the posting’s 9–12 years of relevant experience, technical degree, AI workstream leadership, and hands-on Python and SQL proficiency. List specific examples to discuss without treating the sheet as an eligibility assessment.
Build an end-to-end project walkthrough
Prepare one client-facing AI or GenAI example covering use-case discovery, feasibility, data assessment, architecture choices, delivery plan, evaluation, stakeholder decisions, and measured outcomes. Note your own responsibilities and what you would change.
Practice a solution-design exercise
Sketch a proposed RAG knowledge assistant or document-intelligence solution, both use cases named in the posting. Explain retrieval, model and platform options, integrations, and build-versus-buy trade-offs; label the design as an interview exercise, not a company project.
Prepare an evaluation and governance plan
Create a sample scorecard for answer quality, retrieval performance, reliability, and business value. Add tests and controls for hallucinations, data quality, privacy, security, fairness, human review, auditability, and remediation.
Outline the path from prototype to adoption
Draft a brief delivery plan with workstreams, quality gates, testing, monitoring, deployment-readiness criteria, stakeholder ownership, and adoption measures. The posting specifically calls for leading work through deployment readiness and adoption support.
Rehearse senior-level communication and coaching
Prepare a short executive explanation of one technical trade-off and an example of reviewing deliverables or coaching junior colleagues. Be ready to explain how you managed deadlines, risks, and client expectations.
Write targeted questions for the employer
Ask how this team prioritizes AI use cases, evaluates production readiness, divides architecture and governance responsibilities, and collaborates across global teams. Clarify travel and the Gurugram office or client-site expectations stated in the posting.

Availability last checked: 5 Nov 2026. Status can change after a check; confirm on the employer’s website. Report an issue · Our checking policy

KEEP EXPLORING

Similar AI jobs

Related skills and specializations in India. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…