What AI gets right (and wrong) in financial modeling

In financial modeling, AI should be seen as a time-saver, but not a decision-maker. Think of it as an exceptionally fast (but novice) analyst that can tackle grunt work, but shouldn't make high-stakes calls. AI lacks the nuanced business context required for complex modeling and needs constant oversight, but use it well, and you’ll free up more time for deep strategic thinking.

Ramsha Rizvi, CFA, is the Director of FP&A at Enduring Planet. We sat down with her to discuss where AI can be of use, where to avoid it, and why its rise means great financial modelers are more important than ever.

The work you can offload to AI

Too often, the financial models meant to simulate a company’s entire future are built on shaky ground. Two major avoidable flaws crop up again and again: 

• Poor adherence to model best practices: Minimal color-coding and hard-coded outputs make the model impossible for anyone but the author to follow.
• Not built for regular updates: Models aren't optimized for ETL (Extract, Transform, Load) processes. Actuals, sales pipelines, or operational data are keyed in by hand, meaning the insights become stale the moment the file is saved.

These are exactly the kinds of mistakes AI is adept at fixing. It can also automate the tasks that are taken for granted in FP&A, like writing complex formulas, balancing cash flow statements, or building efficient ETL systems. While creating dynamic and scalable formulas take significant cognitive function, instead they just want to know what the numbers mean for the next raise, business development hire or path to profitability, scalability or sustained liquidity.

How to use the time AI frees up

With the manual architecture in place, modelers free up time to answer the business questions that really matter.

This means you can: 

• Run dozens of "what-if" scenarios in the time it used to take to build one.
• Factor in messy, real-world inputs for variable analysis, from shifting market sentiment to the timing of a key hire.
• Map out not just where the company could be, but the specific drivers required to get there – and what happens when those drivers break.

The mission-critical tasks that are beyond AI’s reach

AI is great at pattern recognition and processing large datasets. But it has severe shortfalls when it comes to complex financial modeling and judgment calls – particularly in climate. From regulatory nuances to novel technologies, the unique context of this sector requires a much more specialized and human touch.

The elements of modeling that can’t be substituted by AI include: 

1. Complex entity structures
The more complex the structure you’re modeling, the more AI becomes unreliable. In carbon removal and sustainable infrastructure, for instance, companies often ring-fence individual projects into separate legal entities to de-risk capital for lenders. That means modeling isn't just about one P&L, but capturing each project's standalone economics as well as how they roll up – something AI struggles to handle.

2. Novel funding frameworks
A highly useful financial model is built at the intersection of finance, operational, and strategic insights. This often leads to unique challenges that demand innovative frameworks.

For instance, if your company is heavily reliant on grants, you’ll likely have to deal with the headache of managing different payment structures. This means you’ll need a model that’s flexible enough to handle each structure individually, but also consolidate them into a single view based on personnel time allocation. This approach requires creative thinking, something that AI – which is trained on existing data and patterns – struggles to pull off.

3. Waterfall representations
Developing accurate waterfall representations for climate fund models is both complex and high stakes – it’s completely out of AI’s wheelhouse. 

Enduring Planet once tested this by tasking Claude with building an Excel fund model featuring a waterfall structure designed to absorb any losses that exceeded the capacity of the most subordinated tranche. Despite several minutes and ten iterations, the model consistently failed to get it right.

4. Cash runway analysis
Cash runway analysis requires isolating key drivers, modeling them in ways that are both robust and flexible, and most importantly, understanding what levers you can actually pull to extend runway. This requires situational awareness that AI lacks. Unlike your team, it doesn’t know the difference between discretionary marketing spend and non-negotiable costs like core engineering. 

5. Budget variance analysis
A simple budget variance analysis might tell you what changed, but not why. To be effective, you need a granular breakdown by department, customer, or project lifecycle. This kind of complex, multi-step workflow demands high accuracy, and is exactly where AI falls short. In fact, Vals AI Benchmark puts the top finance agent at only 63% accuracy, compared to 78.8% for the top software engineering model.

6. Privacy considerations
Building financial models requires highly sensitive internal information, which means using AI comes with real risks around data privacy and information leakage. Only use an enterprise-grade LLM with strict controls and be cautious about what you hand it the keys to.

Ramsha brings a wealth of research and underwriting skills to Enduring Planet, having performed credit, financial analysis and deal structuring for dozens of companies and over $10M+ of investments over her career.

Prior to Enduring Planet, Ramsha played critical roles for SIMA Fund I, a $90M fund for decentralized solar and micro-finance companies in emerging markets, and Spark+ Fund, a $75M fund for the clean cooking sector, anchored by African Development Bank and European Commission. Ramsha is a CFA Charterholder and graduated with a Bachelors in Accounting and Finance in the top 10% from IBA Karachi (a top business school in Pakistan).