First draft speed
A well-structured prompt now generates a three-statement model with scenario cases in about 15 minutes - work that used to take an analyst roughly 20 hours.
Published September 2026 by Amo Veritas
The State of Play in AI for Project Finance
And what it isn't, yet.
Project finance has always run on hyper-complex Excel models and mountains of paperwork - toll roads, fibre, oil rigs. Long-lived assets, zero tolerance for modelling error, financed by an industry that has treated new technology with justified suspicion for decades.
2026 is the year that starts to change - not because AI has “arrived,” most of the tooling is still early, but because large firms are now actually adopting these tools internally, not just piloting them. Click through the stack below to see who is doing what.
90,000people in AI-related roles across the 50 banks tracked by Evident's AI Index
32/50of those banks now disclose an AI use case with measurable impact, up from 26 a year earlier [Evident]
15 minto draft a three-statement model that used to take an analyst 20 hours
01
The three enterprise stacks
Each company sits at a different point in this market. Read each column top to bottom: a banking platform, the Excel plugin underneath it, and the coding agent the same company builds. Click any block for the detail and source. Hatched blocks are layers that company doesn't have.
02
How teams actually work the loop
Not a org chart of who owns what. This is the repeatable cycle a deal actually moves through once these three enterprise stacks are in the room, and where a human still has to step back in.
Human Judgement
Speed changes. The review step doesn't.
1. Draft
Pitchbooks, research, first-pass workflows, the platform tier
2. Model
Build and edit the spreadsheet itself, the Excel plugin
3. Automate
Dashboards, pipelines, reconciliation scripts, the coding agent
4. Verify
Cross-check formulas, citations, and the data room
5. Sign off
A human still owns the credit decision, then the loop starts again
1. Draft
Pitchbooks, research, first-pass workflows, the platform tier
2. Model
Build and edit the spreadsheet itself, the Excel plugin
3. Automate
Dashboards, pipelines, reconciliation scripts, the coding agent
4. Verify
Cross-check formulas, citations, and the data room
5. Sign off
A human still owns the credit decision, then the loop starts again
03
Adjacent & specialist tools
Outside the three stacks sits a second layer of the market - narrower tools built for one job rather than a whole banking relationship. Grouped below by what they actually do.
04
Where it earns trust today
Speed has arrived. Trust hasn't - not for the part of the work that actually carries risk.
Genuinely useful
Research
Running retrieval-augmented generation over a data room to extract clauses, payment terms, and risk boundaries from a due diligence pack is solid and well established across major banks.
Not yet
Complex modelling
Every tool in the Wall Street Prep benchmark was tested on general corporate models, not project finance structures - and every evaluation flags the same failure: none of them correctly handle the interest and debt circularity a real project finance schedule requires. [2]
Unsupervised trust
Build time has collapsed, but review hasn't gone away - it's moved. No tool here, plugin or platform, has made the leap from useful copilot to something a credit committee will sign off on without a human rebuilding the checks by hand.
05
If you're building here
Skip the enterprise-chatbot pitch. Nobody here needs another assistant. Three things actually clear the bar.
I.
Instrument Excel, don't replace it
This world will never move bespoke structuring into a black-box web app. Plug-ins that audit formulas and trace exactly how a number was derived - that's the shape that wins, and it's already what the usage data shows.
II.
Deterministic verification over generative flair
Zero tolerance for hallucination means the highest-value role for AI here is structural verifier: cross-reference the data room against the model, flag where a clause conflicts with a covenant, produce an audit trail a credit committee will actually trust.
III.
Fix the fragmented data problem
Every deal lives in its own isolated structure - scattered bank accounts, engineering portals. Whoever builds the connector layer for this owns a genuinely painful, recurring problem - this is where Pretium Constructum comes in, Amo Veritas's ontology for energy transition financial modelling and cost estimation.
The honest read
AI inside project finance houses is real but earlier than the hype suggests. Genuine tools exist - but nobody has yet made the leap from useful copilot to structural verifier a credit committee signs off on. That gap is the opportunity.
