Scaling AI Everest: Convincing Your CFO to Invest in Generative AI  

Bruce Feldman, Vice President of AI, Lohfeld Consulting Group 

While many organizations can effortlessly pitch the brilliance of artificial intelligence (AI) solutions to their Chief Information Security Officer (CISO), convincing the Chief Financial Officer (CFO) is a whole different ballgame. CFOs need to balance the costs and benefits of various acquisition strategies—make, buy, or lease—to find the sweet spot for their organizations. This article offers government contractors the insights they need to craft a compelling business case that will make their CFOs see the light (and the dollars) of implementing a Generative AI (GenAI) solution. 

Initial Investment in a GenAI Platform: Make/Buy/Lease 

The decision to acquire an AI platform profoundly impacts cash flow, primarily influenced by capital expenditure and amortization. The choice between hosting the AI platform on-premises or in a virtual private cloud (VPC) further complicates this decision. Let’s review the rationale for a make, buy, or lease decision.  

Make: Developing in-house AI components gives organizations control over the architecture but comes with varying degrees of responsibility—from building a user interface to developing a complete platform. McKinsey¹ (Table 1) identifies three archetypes for in-house models, with initial development costs ranging from $0.5M to $200M and annual sustainment costs between $0.5M and $5M. This approach requires significant resources and can divert focus from market-ready capabilities, making it a costly and time-consuming option. 

Table 1. McKinsey Data on Cost for the “Make” Decision on GenAI Archetypes.
The report does not quantify associated hidden costs, such as for organizational change management and data curation.

Buy: Purchasing a GenAI model can save time and reduce opportunity costs compared to developing one in-house. Vendors benefit from economies of scale and external investments, which lower the initial costs. However, customization to meet specific needs and environments is still necessary, adding to the overall expense. 

Lease: Leasing GenAI capabilities through SaaS (Software-as-a-Service) offers reduced initial investments and annual sustainment costs. This model outsources development and operation, providing flexibility and scalability. Leasing can be structured per seat, by usage tokens, or by other fixed models. This model is often more cost-effective and can deliver acceptable performance compared to making or buying. 

Annual Sustainment Costs and Model Hosting 

Annual sustainment costs for genAI platforms include operating expenses for model inference and maintenance. Hidden costs—such as keeping up with platform evolution, managing technical debt, ensuring compliance, training, and maintaining a key performance indicator (KPI) program—add to the overall expense. 

Model Hosting On-Premises vs. in the Cloud: Government contractors increasingly choose between hosting AI platforms on their IT infrastructure or in a VPC. A survey by Andreesen Horowitz (a16z)² found that 28% of enterprises host AI on internal networks, while 72% use VPCs. Both approaches leverage existing cybersecurity measures, but on-premises installations require more customization and integration, making them more expensive than VPC hosting options. 

Making Corporate Data Available to the GenAI Platform: Fine-Tuning vs. Retrieval Augmented Generation (RAG): Deciding between fine-tuning and RAG significantly impacts sustainment costs. Fine-tuning involves recurring training of the AI's Large Language Model (LLM) with corporate data, ensuring superior performance, but at a higher cost, due to the complexity and time required. In contrast, RAG is less expensive because it retrieves information from data sources without extensive fine-tuning. 

A survey cited in the Andreesen Horowitz (a16z) March 2024 article shows that 72% of enterprise decision-makers prefer fine-tuning, while 22% opt for RAG. Although fine-tuning delivers better performance, it is substantially more expensive, highlighting the need for organizations to carefully consider their budget and performance requirements. 

Conclusion 

Investing in genAI for proposal development presents government contractors with several strategic options, each with distinct financial implications. The make, buy, or lease decision, along with choices about hosting and data integration, must be carefully evaluated to balance costs and benefits. By understanding these factors, government contractors can make informed decisions that enhance their capabilities, improve efficiency, and maintain competitiveness in the federal marketplace. So, arm your CFO with these insights and deliver a strong business case for investing in your next AI project.

¹McKinsey Digital, “Technology’s Generational Moment With Generative AI: A CIO and CTO guide,” 2023. https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/technologys-generational-moment-with-generative-ai-a-cio-and-cto-guide#/

^2: Sarah Wang and Shangda Xu, Andreesen Horowitz, “16 Changes to the Way Enterprises Are Building and Buying Generative AI,” March 2024. https://a16z.com/generative-ai-enterprise-2024/.