AI is being described as the biggest economic shift since the industrial revolution. Yet when you look closely at how financial institutions, investment managers and advisors use it in financial modelling for energy, infrastructure and project finance, the revolution is so far conspicuously absent.
In May 2026 we hosted a roundtable in Singapore for senior leaders in project finance, to explore the role of AI in the industry now and in the future. Lenders, investors and advisors shared their hopes and fears, both personal and institutional.
Here are the main themes that came out of that discussion with project, energy and infrastructure finance leaders.
AI literacy in project finance: why trial and error is not a training strategy
The pace of development has outrun the industry’s ability to absorb it. With so many tools to choose from, it is hard to decide which is right. On top of that, we expect instant results from people who have not been trained to get the most out of AI.
A clear analogy comes to mind: expecting analysts to build a robust, industry-standard project finance model from scratch with no training. AI is far more powerful than Excel, and yet for many the strategy is to let individuals learn by trial and error. What they learn is often not shared across the team or the organisation, and it is lost. AI is also far more fallible than Excel. Trial and error is not a training strategy. It is the absence of one.
For teams that want to move beyond trial and error, our AI in Project Finance Modelling course provides structured workflows, a prompt library and a governance playbook designed for project finance teams.
What AI can and cannot do in project finance modelling today
In our own work, we have tested how far the current generation of tools can be pushed. The honest answer: they can draft, edit, analyse and model with real capability, but they cannot work at a transaction-ready level. They lack the context, the specialist judgement and the institutional memory that real deals demand.
We see a future where AI tools increase productivity, reduce risk and open up new opportunities in project finance. But a due diligence process run end to end by AI, or a financial model developed autonomously, will still go before a credit or investment committee of people. AI is a collaborator, not a replacement.
Shadow IT brings new risks that are hard to manage
We all know the concerns about data security, privacy and intellectual property when using tools like ChatGPT, Claude and Gemini. Where data goes, how it is processed and stored and who owns the work of AI are the main concerns for cybersecurity and legal teams across the industry.
That risk is real, but another is often overlooked: people using blacklisted AI tools at home or on their own devices, especially late at night against a tight deadline. And why not? If all you need is a dashboard summary for your financial model before the credit paper goes in, why not ask Claude to do it in minutes? The result looks great. Dig a little deeper, and many of the “insights” turn out to be works of fiction.
How teams manage the flow of data and the use of AI will need to keep developing. While we see more and more “AI slop” on channels like LinkedIn, the telltale signs are getting harder to spot and the output looks cleaner.
AI-assisted model review: the risk of bypassing financial model audit
As part of our own research into how AI could change project finance in the near future, we have been testing tools and ways of applying them to our work. One example was building a tool step by step with AI assistance: a model review tool to help check a financial model against industry best practice.
The result is impressive, but it took 32 iterations of code to reach an acceptable outcome. Many people would not persist to that level and would simply accept what the first or second round produced. Even then, how many models from different sources must it be tested on before you can trust what it says? Our internal work suggests that a production-ready solution needs far more work, skill and testing than most professionals are willing to put in. While our development continues, we still rely on a traditional process where quality can be 100% assured.
We see a real risk of false security when teams rely on review processes built in-house with AI and bypass external verification and validation through established, industry-accepted processes.
Will model audit change because of AI? Absolutely. Will the way we give assurance on a financial model change completely? Not overnight.
Read more about AI in financial modelling
Jerome Brice (opens in a new tab), global leader of financial model audit at Forvis Mazars, has published a series of posts on LinkedIn that go further into these topics:
- Who needs professionalism when intelligence is free? (opens in a new tab) Garicano’s bundle framework, Catalini’s asymmetry between the cost of generating and the cost of verifying, and the Jevons paradox, and what the three suggest about where professional judgement sits when intelligence gets cheaper.
- Better model, same constraints (opens in a new tab). On the release of GPT-5.5: benchmarks show clear gains in agentic tasks, knowledge work and reasoning across complex documents, and in some areas it is now competitive with Anthropic’s Claude.
- AI is generating a lot of things. One of them is gaps (opens in a new tab). These are not gaps in AI capability. They are gaps in our ability to trust what that capability means in practice.
- AI has tells in Excel too (opens in a new tab). We are all getting good at spotting AI-written prose: the em dash, the “uncomfortable truth“, the asyndetic tricolon and the closing line that gestures at a horizon nobody asked about.
In our financial modelling webinar series in February 2026 we covered the topic in How AI is used in project finance modelling. The webinar:
- Looked at where AI fits into everyday modelling work, from building and reviewing models to handling documents, comparing projects, version control and writing credit or investment papers
- Showed how AI helps not only the person building the model but everyone who works with it: reviewers, decision makers and other stakeholders
- Discussed which tools and platforms are emerging as the main options in the market
- Explored what is holding people and organisations back from scaling AI in financial modelling
- Walked through practical examples of where AI saves time, and where its limits still show, in both simple and complex workflows
The path forward for AI adoption in project finance
The path forward is not technical. The technology is, mostly, here. The path forward is institutional, professional and cultural, and it needs the industry to have honest, cross-functional conversations of the kind we had together over lunch. Lender expectations should be clear. Training should be deliberate. Verification should be non-negotiable. And one principle should govern all of it: quality should rise, not fall.
Our AI in Project Finance Modelling course in Sydney and Singapore is designed to provide exactly that: structured, hands-on training for project finance teams adopting AI.
About the author, Joshua Grimm
