Artificial intelligence is reshaping work by delivering speed, efficiency and instant insights across industries. However, the risks of AI misuse are uniquely high for infrastructure investment teams, where a single hallucinated contract term or miscalculated financial ratio can put millions of dollars at risk. This danger has led many such firms to ban AI entirely, while others experiment in an ad hoc manner that could create substantial professional liability.
Neither of these paths is a viable long-term solution. For project financiers, thoughtful, practical AI application must include rigorous integrity, privacy and accuracy guardrails, with purpose-built systems that are validated for the demands of the industry.
In my experience building project finance software with Banyan Infrastructure, which uses technology to improve the efficiency and profitability of infrastructure portfolios, I have seen firsthand the risks that unchecked AI adoption can pose in bias, rigor and accountability. To counter these risks, we must start building a foundation for responsible AI use in project finance today.
Understanding artificial intelligence
Project financiers often conflate automation and AI, but these technologies serve different roles. Automation is deterministic and repeatable, like using Control + F to search a document; it delivers dependable results even when AI features are turned off. AI systems, by contrast, introduce probabilistic behavior and require stronger oversight and safeguards, especially in high-stakes workflows like project finance.
While some tasks benefit from AI’s probabilistic reasoning, others are better suited to predictable, AI-free automation. To adopt AI responsibly, project finance professionals must first understand the kinds of intelligence available and the real outcomes each can enable.
- Retrieval-augmented generation: Retrieval-augmented generation connects a large language model to one or more internal or external knowledge sources, such as an organization’s document repositories and systems of record. In project finance, these systems can enable teams to retrieve and extract critical information from term sheets, financial models and due diligence reports and save that information in a standardized format.
- Agentic AI: AWS defines AI agents as software that can “perform self-directed tasks that meet predetermined goals.” In project finance, an AI agent could detect overdue covenants, draft reminder emails or compile monthly reports for approval.
- Model Context Protocol: Model Context Protocol is an open, emerging standard that lets AI applications and agents connect to external tools and data sources. Such integrations are still early, but they point toward a future where a single AI interface can reach across many systems and reduce much of today’s context switching between platforms.
Where AI breaks down
AI risk in project finance rarely appears as obvious failure. More often, it shows up as small shortcuts that quietly weaken rigor and accountability. For example, models trained on historical deal data can be a helpful tool. But without human oversight, they can reinforce existing patterns — like overweighting certain asset classes, sponsors or structures — and inadvertently create bias in due diligence.
Similarly, incorporating AI-generated outputs directly into credit memos or reports can lead to decisions that lack a clear audit trail. If a ratio is miscalculated or a covenant misread, it becomes unclear how a decision was made if there is no traceability to source data and human approval.
Project financiers can head off these risks by adopting checks and balances in their AI workflows. No matter the task, AI outputs should always be reviewed by a project finance professional. Careful review also supports higher quality work and encourages teams to engage deeply with the underlying documents.
Finally, proper governance is a key step in responsibly adopting AI. With technology evolving rapidly, AI adoption can move faster than controls, increasing risk.
Principles of responsible use
Responsible AI adoption begins with responsible system design. AI tools must be built so users can understand how they work, know when to rely on them and maintain clear oversight of their outputs.
This “human-in-the-loop” approach ensures project finance professionals remain involved at defined points in the AI workflow to ensure accuracy, safety, accountability and ethical decision-making. Simply put, responsible AI is designed in, not bolted on. Effective use depends on concrete guardrails, privacy controls and integrity checks that are embedded at every stage of development, not added as an afterthought.
Trustworthy software providers should have internal AI policies that govern review, approval and escalation procedures. Software should always include features that make AI usage explicit and governable, such as creating clear opt-in settings so that users always know when AI is in use. Separating automation from predictive AI can also ensure usability and value even when companies have strict AI usage policies.
Finally, data privacy is a major concern, especially in a tightly regulated industry such as project finance. Any AI tools must incorporate strict privacy and control measures for data governance and model training permissions.
These considerations are part of an ongoing conversation at Banyan Infrastructure. We believe that responsible AI design requires not only thoughtful internal policies but continuous customer and industry dialogue. We created our AI Advisory Board for this very purpose: to collaborate with project finance leaders on their strategic goals and safety requirements, ensuring our platform minimizes risk while delivering the usability that project finance teams need.
Purpose-built models
There is an ongoing debate about the value of AI tools built by project finance professionals versus generic platforms like ChatGPT or Gemini. Generic systems can summarize language or extract numbers, but they miss nuance in loan structures, covenants and compliance triggers that require domain-specific context, which is why Deloitte recommends fine-tuned, vertical models for finance rather than one-size-fits-all large language models.
To put it simply, purpose-built models work better for project finance because they “speak our language.” They are trained on clean, well-organized deal data, so they can understand tricky terms and covenant logic more accurately and make fewer mistakes in their answers.
A roadmap for adoption
Managing organizational change and new technology adoption is no small task. Here are the strategies that successful operations professionals use to get their teams on board and up to speed with new, innovative technology:
- Define your objectives: Identify business processes where AI can deliver measurable benefits, such as faster diligence reviews or automated portfolio monitoring.
- Select a partner: Choose technology vendors with deep experience in project finance, transparent data practices and teams that are leading AI evolution so that you do not have to go it alone.
- Test before scaling: Run pilot programs to gain fast, iterative feedback to mitigate risk before expanding into daily operations.
- Expand with governance: Scale use only when oversight processes and human review are fully embedded.
AI does not replace professional judgment. It strengthens it by reducing manual work and improving access to insight. With disciplined design and rigorous validation, AI can help the project finance sector move faster, smarter and with greater confidence.
Amanda Li is the COO of Banyan Infrastructure.
Guest posts on ImpactAlpha represent the opinions of their authors and do not necessarily reflect the views of ImpactAlpha.