Forecasting using AI Driver Based Modeling
Use AI reasoning to model detailed line item metrics, growth trends and all forward looking and guidance statements into a set of synthesized financial forecasts
Last updated: 12 November 2025
Objective:
Synthesize an updated and source referenced set of financial forecasts to compare against consensus from any company’s earnings presentations, transcripts and result releases.
Use AI’s reasoning and analysis capabilities to identify, extract and incorporate all recent trends and forward looking statements in arriving at an unbiased and comprehensive set of financial forecasts
Explanation:
Management guidance often comes in subtle and fragmented forms, spread across slide decks, transcripts, and regulatory filings using language that can be opaque, forward-looking, and qualified by corporate jargon. Forecast signals may be embedded in segment-specific commentary, relative performance descriptors, or conditional statements that are easily overlooked or misunderstood by investors with pre-existing biases or heuristics.
This prompt enables AI to systematically surface and interpret all such forward-looking clues, stripping out narrative noise and identifying the quantitative and qualitative anchors that actually drive revenue, margin, cost and earnings trajectories. By parsing both structured data (tables, guidance ranges, segment metrics) and unstructured text (verbal statements, tone, and directional indicators), the model can generate a comprehensive, internally consistent set of financial forecasts that can be free from the distortions of selective perception or confirmation bias.
Because generative models can integrate natural-language understanding with quantitative trend analysis, they can bridge the gap between narrative insight and financial modeling. This allows the AI to recognize when management implies moderation rather than acceleration, or when shifts in pricing mix or cost base signal margin changes not explicitly stated. The resulting forecasts, synthesized from multi-modal evidence, save analysts hours of manual cross-referencing while providing a rigorous foundation for discussion.
Importantly, the process does not end there: these initial AI-generated forecasts can be refined interactively through follow-on prompting, where the user adjusts assumptions, stress-tests sensitivities, or adds new data points—creating a dynamic human-AI forecasting loop that continuously improves analytical accuracy and insight.
As always, be aware that models can make mistakes. At each step, examine the response and challenge information or conclusions that appear erroneous before proceeding to any subsequent steps. If in doubt use a second model with the same prompt to verify the information and generate challenge questions and answers (CoVe process) to correct interpretations of data.
Link to blog post explanation:
N/A
Preferred Model(s):
Gemini 2.5+ Deep Research with attached source documents
Important Execution Notes:
Specify Company, Ticker and Exchange where indicated
Attached source documents (I have tested this with up to 8 attachments incl last full year and a further 3 quarterly presentations and result press releases)
Set the model running - Deep Research can take 5-10 minutes.
Perform follow-on prompting to sensitize forecasts for nuances that may not have been picked up or additional metrics required for specific companies.
Treat the first set of forecasts as an initial first take for fine tuning if necessary.
The model can be prompted to show sensitvities (eg the effect on EPS of an additional 1% revenue growth)
Sample Output:
Copy/Paste Prompt Set:
Important note: Subscribers can use this prompt set for their own analysis. However, the prompt is copyrighted by The Inferential Investor, paywalled, and must not be shared without permission.