Building the AI Platform Team

The Talent Reality

Every banking institution wants to build an AI team. Very few have a realistic plan for doing so.

The market for AI engineers and data scientists is intensely competitive. Banks compete with technology companies that offer higher pay and the prestige of cutting-edge consumer products. Posting job listings and hoping is not a plan.

A successful AI team strategy for banking requires honesty about three things: what roles you actually need (not all of them), where you can realistically hire (not everywhere), and what you should build internally versus buy from vendors (not everything).

The Core Roles

ML Engineer / AI Engineer

What they do: Build, deploy, and maintain AI systems. They select and configure foundation modelsFoundation ModelA large AI model trained on broad data that can be adapted to many tasks. Examples include OpenAI's GPT, Anthropic's Claude, Google's Gemini and Meta's Llama families. Banks evaluate these for capabilities, safety, and regulatory fit.See glossary, build RAG pipelines, implement guardrails, and manage the technical infrastructure.

Why you need them: Without them, you depend on vendors and consultants for every AI decision. You need enough in-house expertise to evaluate vendor claims, make architecture decisions, and maintain production systems.

Hiring reality: This is the hardest role to fill. Experienced ML engineers command premium compensation. Consider hiring mid-level engineers with strong software engineering fundamentals and investing in AI-specific training.

Data Engineer

What they do: Build and maintain the data pipelines that feed AI systems. They handle data extraction, transformation, quality, and integration across the institution's data sources.

Why you need them: AI is only as good as its data, and many AI project failures trace back to data quality, not model quality.

Hiring reality: Easier to hire than ML engineers, and many banks already have data engineering talent. Upskilling existing data engineers with AI-specific skills (embedding pipelines, vector database management) is often the fastest path.

AI Product Manager

What they do: Translate business needs into AI product requirements, prioritize with business stakeholders, and measure outcomes.

Why you need them: Without one, AI teams build technically impressive solutions that nobody uses.

Hiring reality: Look for product managers with domain expertise in banking who are willing to learn AI concepts, rather than AI specialists who do not understand banking. Banking domain knowledge is harder to teach than AI concepts.

Prompt and Agent Designer (often part of the AI engineer or product role)

What they do: Design, test, and optimize the instructions that guide AI behavior -- promptsPrompt EngineeringThe practice of crafting effective instructions (prompts) to guide AI model behavior. Techniques include few-shot examples, chain-of-thought reasoning, and role-based system instructions.See glossary, system instructions, and increasingly the steps, tools, and hand-offs of AI agentsAgentsAI systems that can autonomously plan and execute multi-step tasks by calling tools, querying data sources, and making decisions without human intervention at each step -- typically within defined permissions and human approval checkpoints.See glossary. They build templates, implement orchestration patternsOrchestration FrameworkSoftware that coordinates LLMs, tools, and data sources into complex workflows. Frameworks like LangGraph, CrewAI, and vendor toolkits such as the OpenAI Agents SDK and Microsoft Agent Framework manage prompt chains, memory, and tool calling for multi-step AI tasks.See glossary, and tune AI interactions for quality and consistency.

Why you need them: The difference between a mediocre and an excellent AI application often comes down to how well it is instructed. Well-designed instructions improve output quality, reduce hallucinations, and keep outputs within institutional standards.

Hiring reality: Treat this as a skill, not necessarily a separate job title -- it often sits inside the AI engineer or product role. Look for strong writing, analytical thinking, and a systematic testing mindset; backgrounds in technical writing, QA, or compliance are surprisingly well-suited.

Citizen builders: Business staff now build their own assistants and agents too. BNY reported in April 2026 that more than half of its employees are building AI agents on its internal platform. The hub's job is to govern that activity -- approved tools, templates, and review -- not to do all the building itself.

Model Risk Analyst (AI-focused)

What they do: Validate AI models, test for bias and fairness, monitor ongoing performance, and ensure compliance with MRM frameworks.

Why you need them: In April 2026 the Federal Reserve, OCC and FDIC replaced SR 11-7 (and OCC Bulletin 2011-12) with updated model risk management guidance -- SR 26-2, OCC Bulletin 2026-13 and FDIC FIL-15-2026. The new guidance is aimed mainly at banks with over $30 billion in assets, and it explicitly places generative and agentic AI outside its scope while the agencies gather input on how banks use AI (the OCC said a request for information is coming). That is not a free pass: examiners can still act on unsafe or unsound practices or violations of law, and most banks continue to apply model-risk disciplines -- inventory, validation, monitoring, documentation -- to their AI systems. This role bridges your existing MRM function with AI-specific validation.

Hiring reality: Extend your existing model risk team. Current model validators with quantitative backgrounds can learn AI validation methodologies more readily than AI engineers can learn banking regulation.

AI Operations and Agent Supervision

Once agents run in production, someone has to watch them: monitoring their work, handling the exceptions they escalate, and switching them off when something goes wrong. At BNY, according to press reports, each "digital employee" has a human manager. Plan for this as an operations role -- often filled from the business line the agent serves -- rather than leaving it to the engineers who built the agent.

Team Models

Centralized: AI Center of Excellence

All AI talent sits in a single team that serves the entire institution.

Works when: The institution is in early stages (1-3 production use cases), AI talent is scarce and must be shared, and standardization is more important than speed.

Risks: Becomes a bottleneck as demand grows. Business units feel underserved. Projects are prioritized by the CoE rather than by business value.

Embedded: AI in Business Lines

AI talent is distributed across business lines, with each major division having its own AI engineers and data scientists.

Works when: The institution has mature AI capabilities, sufficient talent to distribute, and well-established governance standards that embedded teams follow.

Risks: Inconsistent standards, duplicated effort, difficulty attracting and retaining talent in smaller teams, governance gaps.

Hybrid: Hub-and-Spoke

A central AI platform team provides shared infrastructure, standards, and governance. Business line AI teams (or liaisons) drive use case identification and domain-specific implementation.

Works when: The institution has growing AI capabilities and wants to balance governance with business responsiveness. This is the most common model for banks at Stage 2-3 of the maturity model.

Where the hub reports matters. Put AI leadership close enough to the top that it can set standards across business lines. JPMorganChase, for example, said in February 2026 that its CIO and its Chief Data and Analytics Officer both sit on the firm's Operating Committee, with a dedicated AI specialist, to organize and measure AI across the firm.

The hub provides: Shared AI infrastructure, model hosting, guardrails framework, governance standards, training and capability development, vendor management.

The spokes provide: Business domain expertise, use case identification, stakeholder management, domain-specific prompt engineering, outcome measurement.

The Build-vs-Buy Decision

Not every AI capability needs to be built in-house:

Build internally:

• Proprietary RAG systems over your institution's data (competitive advantage from your data, not from the technology)
• Prompt engineering and system instructions (encode your institutional knowledge and compliance requirements)
• Integration with internal systems (nobody knows your systems like your team)
• Governance and monitoring (cannot outsource regulatory accountability)

Buy from vendors:

• Foundation models (do not train your own LLM -- use commercial models from providers such as OpenAI, Anthropic, and Google, accessed directly or through cloud platforms such as AWS Bedrock or Microsoft Azure. Large banks typically use several models at once)
• AI infrastructure (use managed services for model hosting, vector databases, and orchestration)
• Specialized tools (document parsing, PII detection, content safety -- buy proven solutions rather than building from scratch)
• Initial consulting (bring in expertise for architecture design and first deployment, then build internal capability to maintain and extend)

Be skeptical of vendor labels. Gartner (June 2025) warned of "agent washing" -- vendors relabeling existing chatbots and automation as agents -- and estimated that only about 130 of the thousands of vendors claiming agentic AI offer the real thing.

Realistic Staffing Timeline

Months 1-3 (Exploring):

• 1 senior AI engineer (technical lead)
• 1 data engineer (may be repurposed from existing team)
• 1 AI product manager (may be part-time, repurposed from existing PM role)

Months 3-9 (Experimenting):

• Add 1-2 junior AI engineers
• Add prompt and agent design skills (a dedicated designer or a trained AI engineer)
• Assign a model risk analyst to AI validation (part-time from existing MRM team)

Months 9-18 (Scaling):

• Grow to 5-8 AI engineers
• Add dedicated prompt and agent design capability, plus a governed program for citizen builders
• Assign operational supervision for any agents in production
• Full-time model risk analyst for AI
• Consider hub-and-spoke model with business line liaisons