Choosing the Right Implementation Strategy

When J.P. Morgan and Goldman Sachs announced major GenAI initiatives last year, they took notably different approaches—one prioritized vendor partnerships, the other emphasized building proprietary models in-house. For investment banks evaluating their own GenAI strategies, this divergence highlights a critical question: what deployment approach best fits your organization's capabilities, risk tolerance, and strategic priorities?

An Enterprise GenAI Blueprint isn't just about what you build—it's fundamentally about how you build it. The right approach depends on your firm's technical resources, regulatory constraints, and the specific workflows you're targeting. Let's examine four common deployment strategies with their respective trade-offs.

Approach 1: Custom In-House Development

Some firms, particularly those with established quantitative trading operations and deep ML expertise, opt to build proprietary GenAI capabilities from the ground up. This means training custom models on internal data, developing bespoke infrastructure, and creating specialized applications for functions like derivatives pricing or portfolio optimization.

Advantages:

• Complete control over training data, model architecture, and deployment environment
• Ability to incorporate highly proprietary trading strategies and analytical frameworks
• No vendor dependencies or concerns about competitive intelligence leaking through shared platforms
• Custom integration with legacy systems and existing workflows

Disadvantages:

• Requires significant investment in ML talent, GPU infrastructure, and ongoing maintenance
• Slower time-to-value as teams build everything from scratch
• Higher technical risk if you lack deep expertise in production ML systems
• Difficult to keep pace with rapid advancement in foundation models

Best for: Bulge bracket banks with strong technology divisions, use cases involving highly sensitive proprietary methodologies, firms with patient capital willing to invest 12-18 months in foundational capabilities.

Approach 2: Commercial Platform Integration

Many investment banks are partnering with enterprise AI platforms that provide pre-trained foundation models, governance frameworks, and deployment infrastructure. These platforms handle the heavy lifting of model training and infrastructure management while allowing customization for specific use cases like M&A due diligence or regulatory reporting automation.

Advantages:

• Faster deployment (weeks to months vs. quarters to years)
• Access to continuously updated foundation models without in-house ML research
• Built-in governance, security, and compliance features designed for regulated industries
• Lower upfront capital investment and more predictable operating costs

Disadvantages:

• Less control over model behavior and update cycles
• Potential vendor lock-in if switching costs are high
• Must ensure vendor's data handling practices meet your compliance requirements
• May not support highly specialized use cases without significant customization

Best for: Mid-sized investment banks, firms prioritizing speed-to-market, organizations with limited ML expertise, use cases that don't require proprietary model architectures.

For this approach, selecting platforms with proven financial services implementations significantly reduces integration risk and accelerates time-to-value.

Approach 3: Hybrid Architecture

The hybrid approach combines commercial foundation models for general capabilities (document analysis, summarization) with custom-trained models for specialized functions (credit risk scoring, trade surveillance). This strategy is increasingly popular among firms that want both speed and differentiation.

Advantages:

• Balance between speed (commercial platforms) and specialization (custom models)
• Ability to use vendor platforms for lower-risk use cases while maintaining control over proprietary analytics
• More flexibility to shift strategies as technology and business needs evolve
• Reduced overall risk through diversification

Disadvantages:

• Higher complexity managing multiple systems and vendors
• Requires governance frameworks that span different deployment models
• Integration challenges between commercial and custom components
• Need expertise in both vendor platform management and custom ML development

Best for: Large regional banks, firms with some ML capability but not top-tier talent depth, organizations that need both generic productivity tools and specialized analytical models.

Approach 4: Consortium or Industry Utility Models

Some investment banks are exploring industry-consortium approaches where multiple firms jointly develop or fund GenAI capabilities for common functions like KYC automation, regulatory reporting, or market data analysis. These shared utilities distribute costs and regulatory burden while maintaining competitive separation for client-facing activities.

Advantages:

• Dramatically lower per-firm costs for expensive infrastructure and model development
• Shared regulatory compliance work and audit frameworks
• Industry-wide standardization can improve interoperability
• Less competitive pressure on non-differentiating workflows

Disadvantages:

• Slower decision-making and product development due to multi-stakeholder governance
• Limited ability to customize for firm-specific workflows
• Potential competitive concerns even for "non-core" functions
• Dependency on consortium continuation and financial health

Best for: Non-differentiating back-office operations, regulatory compliance functions, smaller firms that can't justify individual investments, use cases where industry standardization adds value.

Making Your Decision

Your Enterprise GenAI Blueprint should clearly articulate which approach(es) you'll use for different categories of use cases. A practical framework:

Use custom development for:

• Proprietary trading strategies and alpha-generating analytics
• Core competitive differentiators in M&A advisory or structured finance
• Use cases requiring extremely sensitive client data that can't leave your environment

Use commercial platforms for:

• Productivity tools (document drafting, research summarization)
• Well-understood problems with available industry solutions
• Pilot projects where you want to learn quickly

Use hybrid approaches for:

• Complex workflows that span generic and specialized capabilities
• Situations where you want optionality to shift strategies

Use consortium models for:

• Regulatory compliance and reporting
• Standardized post-trade processes
• Industry data analysis where collaboration adds value

Conclusion

There's no single "correct" deployment approach for enterprise GenAI in investment banking. The most sophisticated Enterprise GenAI Blueprint recognizes that different functions, risk profiles, and strategic priorities demand different solutions. What matters is making deliberate choices aligned with your firm's capabilities and competitive strategy rather than defaulting to whatever approach vendors happen to be pitching this quarter.

As you finalize your blueprint, explore specialized solutions like AI Agents for Finance that offer flexibility across deployment models while maintaining the governance rigor investment banking demands. The winners in the GenAI transformation won't be the firms that move fastest, but those that move most strategically.