As Generative AI (GenAI) transitions from experimental pilots to enterprise-scale adoption, banking and capital market (BCM) firms are exploring its potential to transform functions ranging from risk compliance and trading intelligence to customer service and operational efficiency.
However, amid the promise lies complexity. Procurement leaders face unprecedented challenges in evaluating solution maturity, defining ROI-centric use cases, and managing hidden cost structures tied to large language model (LLM) usage.
This article offers a procurement-focused roadmap for sourcing GenAI solutions effectively. Leveraging insights from the 2025 BCM market, we outline critical success factors across vendor evaluation, pricing transparency, integration readiness, and compliance governance.
Key themes addressed include:
- How to structure an RFP to reflect compute-heavy cost drivers
- Evaluating model lineage and explainability in vendor proposals
- Aligning AI deployment roadmaps with organizational risk frameworks
- Case examples from global financial firms driving GenAI at scale
Introduction
Banking and Capital Markets are undergoing a paradigm shift. The convergence of digital transformation, real-time data analytics, and regulatory evolution has created fertile ground for disruptive technologies, including Generative AI.
Yet, as GenAI moves out of sandbox environments and into enterprise-critical workflows, procurement teams face a growing mandate: to balance innovation with accountability. Traditional sourcing methods struggle to account for GenAI’s variable cost models, model bias risks, and evolving vendor ecosystem. For example, unlike off-the-shelf SaaS applications, GenAI deployments are highly context-sensitive, often requiring fine-tuning on proprietary datasets and integrating with complex IT infrastructures.
So, how can procurement lead the development of scalable, ethical, and cost-efficient GenAI strategies?
GenAI Sourcing Challenges in 2026
The adoption of GenAI across financial services is accelerating in 2026; however, sourcing strategies are being reshaped by macroeconomic instability, geopolitical tensions, and tightening regulatory scrutiny. Ongoing conflicts impacting energy markets, persistent inflationary pressures, and recession risks in key economies are driving budget constraints and cost volatility, directly influencing GenAI investment decisions.
At the same time, the GenAI ecosystem remains highly fragmented, with rapidly evolving vendors, pricing models, and compliance expectations. Traditional procurement approaches are no longer sufficient to manage token-based economics, model risk, and cross-border data regulations. As a result, sourcing leaders must balance innovation with cost control, resilience, and governance.
Table 1: Key Challenges and Sourcing Implications:
| Challenge | Sourcing Implication |
| Rapid vendor proliferation & consolidation cycles | Difficulty in benchmarking vendor maturity as new entrants emerge while hyperscalers consolidate capabilities (e.g., platform bundling across cloud + GenAI) |
| Volatile pricing models (token, compute, GPU costs) | Unpredictable TCO as inference costs fluctuate; GenAI pricing can vary 30–50% across vendors, requiring dynamic cost benchmarking |
| Macroeconomic pressure (recession risk, inflation) | Increased scrutiny on ROI; enterprises prioritizing high-impact use cases with 20–40% productivity gains over experimental deployments |
| Geopolitical & supply chain disruptions (AI chips, energy) | GPU shortages and export restrictions impacting model training/inference costs and vendor availability |
| Regulatory fragmentation (EU AI Act, U.S., APAC policies) | Need to embed explainability, auditability, and data residency clauses directly into contracts |
| Lack of standardized benchmarks & SLAs | Difficulty in defining measurable KPIs for accuracy, latency, and hallucination risk in RFPs |
| Data privacy & sovereignty risks | Increased due diligence on where models are trained, hosted, and how enterprise data is used or retained |
| Open-source vs proprietary model trade-offs | Cost vs control dilemma; open-source offers flexibility but requires higher internal capability and governance investment |
Unlike conventional enterprise software, GenAI solutions often bundle proprietary LLMs, vector databases, fine-tuning services, and integration APIs. The absence of consistent performance metrics and reference pricing makes apples-to-apples comparisons elusive. Furthermore, the rise of open-source vs. closed-source debates, and region-specific AI acts (like the EU AI Act and U.S. Executive Orders) require sourcing leaders to incorporate AI governance requirements directly into commercial agreements.
A strategic sourcing approach in GenAI should therefore not only evaluate technical merit or cost but also focus on deployment maturity, scalability for regulated environments, and contractual mechanisms that ensure continuous model governance. Without this, enterprises risk selecting tools that either fail to scale or breach compliance boundaries—leading to rework, inflated costs, or reputational harm.
Sources: KPMG Geni AI in Banking, *McKinsey & Company – The Economic Potential of Generative AI (2023–2024), Deloitte-AI in Banking & Capital Markets (2024)
GenAI Sourcing in Numbers: Use Cases, Costs & Governance
In the dynamic GenAI ecosystem, enterprises must rethink their sourcing lifecycle to align with AI-native operational risks, evolving pricing architectures, and domain-specific value generation. A phased sourcing model is essential – not only to align stakeholder expectations but to systematically de-risk vendor selection, technology integration, and long-term cost scalability.
The framework below outlines a three-phase sourcing approach tailored to enterprise-grade GenAI adoption, with key checkpoints and controls embedded at each stage.
Table 2: Sourcing Phases & Key Actions:
| Phase | Objective | Key Actions | Market Anchors (Real Data Points) |
| Phase 1: Demand Planning & Use-Case Prioritization | Identify high-impact GenAI use cases aligned to business value and data readiness | • Map use cases across fraud, AML/KYC, customer service, research automation • Assess availability of structured, semi-structured, and unstructured data (e.g., transactions, filings, CRM, research data) • Prioritize use cases with measurable ROI (>25–30%) and regulatory relevance | • 76% of financial institutions prioritize fraud detection as a top GenAI use case • 66% are investing in AI-led customer interaction automation • Early GenAI deployments in banking show 20–40% reduction in manual processing time (industry benchmarks) |
| Phase 2: Vendor Evaluation & Model Fit | Select vendors based on model performance, cost efficiency, and integration capability | • Evaluate model types (LLMs vs simulation models) based on use case • Assess performance across quality, latency, scalability, and explainability • Validate integration with enterprise data, APIs, and LLMOps stack • Conduct cost benchmarking across token usage and compute requirements | • GenAI pricing varies widely: ~$0.20–$0.75 per 1M input tokens and up to $4.5 per 1M output tokens for advanced models • Enterprises report 30–50% cost variance between open-source vs proprietary deployments • Model performance gaps (latency/accuracy) can drive 2–3x difference in operational efficiency |
| Phase 3: Commercial Structuring & Deployment Governance | Establish commercial guardrails and ensure scalable, compliant deployment | • Define token-based pricing, API usage limits, and compute caps • Negotiate model retraining terms, IP ownership, and auditability clauses • Implement monitoring, human-in-the-loop validation, and compliance controls • Align SLAs with latency, uptime, and regulatory requirements | • Shift to token-based billing models introduces variable cost structures per query/inference • Regulatory changes like T+1 settlement (2024) increase the need for real-time processing and control frameworks • Poor governance in AI deployments can increase operational risk exposure by 20–30% (industry risk estimates) |
RFPs should reflect GenAI’s distinct commercial structure, especially tokenized billing, where each query or inference has variable cost implications. Enterprises must negotiate compute ceilings, ensure transparency on retraining cycles, and retain rights over custom fine-tuned models. This enables both budget predictability and long-term control over AI-generated IP.
Prioritize vendors offering fine-tuned, domain-specific models (e.g., BloombergGPT-style) with proven performance on open LLM benchmarks like HELM or LMSYS Chatbot Arena.
Sources: Sources: KPMG Geni AI in Banking, *McKinsey & Company – The Economic Potential of Generative AI (2023–2024), Deloitte-AI in Banking & Capital Markets (2024), Gartner – Generative AI Cost & Value Trends (2024–2025)
Procurement Implication: Procurement is shifting from static licensing to usage-based GenAI sourcing, where costs scale per query and model performance. With 30–50% cost variance across models (Gartner) and up to 40% productivity gains (McKinsey), teams are prioritizing cost-per-use-case benchmarking, vendor flexibility, and governance-led contracts to control spend volatility and ensure compliance in real-time deployments.
Cost Considerations for Gen AI Adoption
As GenAI adoption scales, Total Cost of Ownership (TCO) is shifting from fixed budgeting to real-time, usage-driven cost management. Costs are influenced by token consumption, compute intensity, data pipelines, and continuous model optimization.
On-premise deployments offer control but require high upfront investment in GPUs, integration, and talent. Cloud-native models enable faster scalability but introduce variable costs tied to API usage, compute, and model complexity, making spend less predictable.
Going forward, sourcing teams must evaluate TCO beyond pricing, factoring in data readiness, governance, retraining cycles, and regulatory compliance, as these will drive long-term cost sustainability.
Table 3: GenAI TCO Comparison:
| Cost Component | On-Premise Deployment | Cloud-Native Deployment |
| Model Access / Licensing | High upfront | Medium; usage-based |
| Compute & Infrastructure | High (GPU-heavy capex) | High, variable (GPU/TPU usage) |
| Integration & Data Engineering | High | Medium |
| Security & Governance | Medium–High | Medium–High |
| Monitoring & Retraining | High | Medium |
| Support & Maintenance | High | Medium |
Sources: OpenAI – API Pricing (2025), Amazon Web Services – Bedrock Pricing & Model Access
Case Studies: Real-World GenAI Adoption
Case 1: Morgan Stanley – AI Copilot for Wealth Management (2024–2026)
Morgan Stanley scaled its GenAI assistant (built with OpenAI) across ~16,000 financial advisors in 2024. The tool enables natural language search across internal research, filings, and knowledge bases, improving advisory efficiency and reducing research time by ~20–30%.
The deployment leveraged internal document embeddings and a modular API-led architecture, ensuring secure integration with proprietary data while maintaining flexibility in vendor selection.
Source: Morgan Stanley press release (2024); OpenAI partnership updates (2024–2025)
Case 2: JPMorgan Chase – GenAI for Risk, Research & Internal LLM Platform (2024–2026)
JPMorgan expanded its GenAI initiatives in 2024 with internal LLM platforms and tools such as “IndexGPT,” supporting research automation, compliance documentation, and internal knowledge management. These deployments are delivering 30–40% productivity improvements in document-heavy workflows.
The bank adopted a hybrid cloud strategy with strict governance controls, focusing on explainability, auditability, and regulatory alignment.
Source: JPMorgan investor day materials (2024); public filings; CEO commentary (2024–2025)
Case 3: Goldman Sachs – GenAI for Developer Productivity (2024–2026)
Goldman Sachs rolled out GenAI tools to thousands of developers in 2024 to support code generation, testing, and documentation. Early results indicate 20–30% productivity gains, accelerating software development cycles across trading and risk systems.
The firm emphasized secure, controlled environments with human oversight, ensuring compliance with internal and regulatory standards.
Source: Goldman Sachs CIO statements; industry interviews (2024–2025)
Case 4: HSBC – AI-Driven AML & Fraud Detection (2024–2026)
HSBC continues to expand AI and GenAI capabilities in AML and fraud detection, leveraging large-scale transaction data and behavioral analytics. The initiative has improved efficiency and reduced false positives, aligning with industry trends where fraud remains the top GenAI use case (~76%).
The bank’s approach combines advanced analytics with strong data governance and regulatory compliance frameworks.
Source: HSBC innovation updates (2024–2025); KPMG GenAI in Financial Services (2025)
Risk Management & Governance in GenAI Sourcing
The deployment of GenAI in banking and capital markets must be coupled with robust governance mechanisms to manage risk, ensure regulatory compliance, and protect customer trust.
Key Focus Areas
- Bias & Fairness Audits
Vendors should demonstrate ability to undergo regular third-party audits to detect and mitigate bias in model predictions. - Explainability & Traceability
In high-regulation environments such as under Basel III, Dodd-Frank, or MiFID II, institutions must adopt GenAI solutions that provide explainable outputs and traceable model lineage. - Model Drift Monitoring
Banks must incorporate SLAs that require continuous monitoring and retraining schedules to ensure GenAI models remain accurate as data evolves.
Table 4: Recommended Tools & Practices
| Objective | Recommended Tools |
| Model Lineage & Audit Trails | MLFlow, Azure OpenAI integration |
| Model Interpretability | SHAP, LIME, Captum |
| Drift Detection | Arize AI, WhyLabs, Seldon Core |
Robust model governance policies, especially those supporting auditability, will be essential to secure regulator confidence in GenAI deployments in financial institutions.
Sources: U.S. Securities and Exchange Commission, World Economic Forum – AI Governance in Financial Services (2024)
Strategic Recommendations for GenAI Sourcing
To successfully integrate GenAI while mitigating procurement risks, firms must shift toward a more proactive and technically grounded sourcing approach.
Develop a GenAI Sourcing Playbook
- Standardize Use-Case Evaluation:
Create templates that include cost-benefit analyses, ROI estimates, and technical feasibility scoring for GenAI projects. - Embed Legal Clauses Specific to AI:
Contracts must include clauses addressing model ownership, data usage rights, retraining protocols, and AI hallucination liabilities. - Prevent Shadow IT Risks:
Establish a collaborative framework between procurement, CIO/CTO office, and risk/compliance teams to govern AI tool adoption.
Future-Proof the Vendor Portfolio
- Encourage Pilot Programs:
Use sandbox testing environments to evaluate performance before scaling. - Benchmark Vendors Against Open Metrics:
Require vendors to publish performance metrics on industry-standard benchmarks like MMLU, HELM, and ARC to ensure transparency.
Conclusion
Generative AI is no longer an experimental capability in banking and capital markets; it is rapidly becoming a strategic enabler of efficiency, risk intelligence, and customer engagement. Successful adoption hinges on a data-first foundation, where institutions align use cases with high-quality structured and unstructured data, and select models based on performance, scalability, and cost efficiency.
However, scaling GenAI requires more than technology deployment. Firms must adopt a phased, governance-led sourcing approach, ensuring that vendor selection, pricing models, and deployment architectures are tightly aligned with regulatory expectations and operational realities. The increasing shift toward usage-based economics, hybrid deployments, and continuous model monitoring further reinforces the need for procurement and technology teams to work in tandem. Looking ahead, organizations that can balance innovation with control, leveraging GenAI for productivity gains while embedding robust governance frameworks, will be best positioned to unlock sustainable value and maintain competitive advantage in an increasingly complex financial ecosystem.
References
[1] ” Acquihire 2.0: A Practical Guide to Develop Generative AI Solutions for CIB & Capital Market Institutions. [Online]. Available: A Practical Guide to Develop Generative AI Solutions for CIB & Capital Market Institutions
[2] ” The Rise of The Mega Acqui-Hire (Medium). [Online]. Available: The state of AI in 2025: Agents, innovation, and transformation-Mckinsey & Co
[3] ” Fighting fraud in payments with AI. [Online]. Available: Fighting fraud in payments with AI
[4] ” The EU Artificial Intelligence Act: Up-to-date developments and analyses of the EU AI Act,” EU AI Act. [Online]. Available: The EU Artificial Intelligence Act-p-to-date developments and analyses of the EU AI Act
[5] ” GenAI may generate 20-40% productivity gains, no risk to tech spends: Axis Capital report” Economics Times. [Online]. Available: GenAI may generate 20-40% productivity gains, no risk to tech spends: Axis Capital report
[6] ” Banking & Capital Markets,” Deloitte. [Online]. Available: Deloitte-Banking & Capital Markets
Author
Saroj Das
Satyam Dash
Related Reading
16 Apr, 2026
Navigating 2026 Payroll Compliance: Why Outsourcing Gives Global Teams an Edge
16 Apr, 2026
Assessing the Impact of National Cybersecurity Strategies in African Emerging Markets
