Financial Modelling with AI: Part 3
Generative AI tools like ChatGPT and Copilot can support the testing and implementation phases of financial model development. While AI cannot yet fully automate model building, it offers valuable assistance once the model is constructed.
This article focuses on the last two stages of model development:
- Test: Verifying the model's accuracy by checking results and reviewing formulas.
- Implement: Ensuring users understand how to operate the model, supported by clear documentation and training.
AI can help with scenario analysis, what-if and goal-seeking analysis, benchmarking against industry metrics, generating explanations for documentation, and creating visualisations.
Scenario Analysis
Both ChatGPT and Copilot performed well when tasked with generating scenario results by varying inputs. For example, they accurately reproduced nine scenario outcomes combining cost assumptions and tax rates. The results matched manual calculations, though the AI outputs were values only, without showing formulas or methodology.
This indicates AI tools can handle straightforward scenario analyses effectively, but they provide limited transparency on how results are derived.
What-If Analysis and Goal Seeking
When asked to adjust assumptions to reach a target net profit after tax (NPAT), both AI tools produced accurate values. They accounted for dependent calculations like cost of goods sold and operating expenses.
However, AI often outputs numbers without explaining how those results were calculated. This lack of context can be a challenge for users who want to understand or validate the results.
More complex iterative calculations over multiple periods were also handled correctly by both tools, but again without detailed explanations. AI managed multi-dimensional what-if scenarios involving simultaneous changes in indirect costs and tax rates, though minor discrepancies appeared with Copilot under heavier computational demands.
The takeaway: AI can assist with test calculations, but independent verification remains essential.
Metrics and Industry Benchmarking
After building a model, benchmarking financial ratios and performance metrics against industry standards is common practice. AI tools can generate these metrics from financial statements quickly.
ChatGPT provided a range of key metrics for the final year of a model, including profitability and efficiency ratios. Copilot also produced relevant metrics but fewer in number.
However, when asked to compare these metrics to real companies, AI struggled. Both ChatGPT and Copilot produced inaccurate references or fabricated data sources. They sometimes made misleading or illogical comparisons, such as equating vastly different cash flow figures.
Use AI benchmarking outputs as a starting point, but always cross-check against reliable data.
Providing Explanations and Documentation
AI excels at generating documentation and training materials. It can explain model logic, formulas, and workflows in conversational language. This interactive approach is useful for end users who prefer concise, on-demand help over lengthy manuals.
Visualisations and Charting
When it comes to presenting results visually, ChatGPT outperformed Copilot. Given financial statements, ChatGPT suggested relevant charts and produced multi-line graphs showing profit trends and expenditure breakdowns.
Initial charts had some axis and labeling issues, but with iterative prompts, the output improved significantly. Copilot struggled to create accurate charts, often producing incorrect values or labels.
Visual output by AI can enhance model presentation, but expect to refine and validate charts carefully.
Final Thoughts
Responses from AI tools can vary widely even with the same prompts. Still, ChatGPT and Copilot remain valuable helpers during the testing and implementation phases of financial models, especially for those with limited modelling experience.
Verification is crucial—never accept AI-generated results without checking. ChatGPT generally delivers stronger qualitative and quantitative support across modelling tasks.
For finance professionals looking to boost their AI skills, exploring targeted courses on using AI in Excel and financial analysis can be a practical next step. You might find useful resources at Complete AI Training.
Your membership also unlocks:
