Case Study: Financial Sentiment Analysis with GenAI & Google Cloud

Q: What was the fundamental business problem?

Manually analyzing thousands of daily financial reports took hours and was too slow to allow for quick, data-driven investment decisions.

Q: What technology was used to understand the documents?

We built a custom data pipeline in Google Cloud, using Vertex AI Gemini 2.5 to automatically extract relevant signals and sentiment from the unstructured text of the PDF reports.

Q: How fast is the new automated process?

The system processes over 60 pages of daily financial analysis in under 5 minutes, compared to the hours required for manual review.

Q: How accurate is the generated sentiment score?

The AI-calculated score has a high correlation (r = 0.79) with the standard CNN Fear & Greed Index, which validates its accuracy in reflecting the market.

Case Studies Software Development Process Automation & AI Agents Digital Transformation

INTRO

A European asset manager needed a more systematic way to tune their portfolios based on the tone and recommendations found in daily macroeconomic reports published by investment banks and financial consultants. Over four years, they had collected nearly 3,800 PDF reports, roughly 240,000 pages, with new material arriving every day

Their research team could only skim a few fresh reports each morning, often reacting too late to shifts in market sentiment.

In the summer of 2025, OPTI implemented a Google Cloud-based solution.

CHALLENGES

Transforming nearly four years of dense financial reports into a reliable, daily investment sentiment score, fast enough to act on, and accurate enough to trust. Therefore, we formulated specific objectives:

1. Process Large Volumes in Minutes

Automate the extraction of insights from 60+ pages of daily financial analysis in under 5 minutes, replacing manual workflows that took hours.

2. Extract Granular Sentiment Signals

Identify and weigh key market drivers, such as regime shifts, liquidity, and inflation, directly from unstructured text.

3. Achieve Quant-Grade Rigor

Deliver a sentiment score that aligns with established benchmarks like CNN’s Fear & Greed Index, while incorporating the client’s proprietary insights

"It is amazing that you rediscovered the CNN Fear & Greed index independently via AI. This proves AI can bring precision to financial research."

Investor and asset manager

SOLUTION

FROM IDEA TO DELIVERY IN 6 WEEKS

To address these challenges, OPTI implemented a custom AI pipeline using Google Cloud’s Gemini 2.5, purpose-built to analyze large financial documents at scale. Key components included:

End-to-End ETL Pipeline
- Automated ingestion of daily reports using Cloud Storage triggers and Cloud Functions, with OCR fallback for scanned pages.
AI-Powered Document Understanding
- Gemini 2.5 used strict prompt engineering and a map-reduce strategy to extract sentiment-relevant sections, financial tables, and key signals from unstructured text.
Custom Sentiment Scoring Model
- Insights were weighted across several market drivers (Regime, Liquidity, Positioning, Policy/Fiscal, etc), then normalized into a 0–100 daily sentiment score via an automated workflow.
Live Dashboard & Tactical Insights
- Scores were synced to Looker Studio / Google Sheets, enabling real-time monitoring and supporting tactical allocation decisions (Equity, Bonds, etc.).

This project shows how business intelligence (BI) can be revolutionized with AI. The final dashboard doesn't just present data; it provides actionable insights. It allows the management team to make faster, more informed investment decisions based on an objective analysis of market sentiment

RESULTS (30 DAYS LATER)

210h development + 30h PM/testing, Delivered in 6 Weeks

100% Report Coverage

All 3,800 initial reports + daily incoming documents now processed automatically.

< 5 minutes processing time

Down from 2–3 hours of manual reading per day.

High correlation with CNN Fear & Greed Index (r = 0.79)

Aligned sentiment trajectory during major events (e.g., Powell speeches).

Dashboard for daily allocation decisions

Enables data-driven shifts between asset classes, based on in-depth report analysis.

TECHNOLOGIES

Google Cloud Run
Cloud Functions
Vertex AI Gemini 2.5
BigQuery
Looker Studio
Google Sheets

The CNN Fear & Greed Index (top) vs. the results of our AI-based sentiment analysis (bottom). Its responses closely follow the well-known CNN index, showing that our AI can reliably interpret complex market sentiment.

References

It is amazing that you rediscovered the CNN Fear & Greed index independently via AI. This proves AI can bring precision to financial research

Investor and Fund Manager

It is amazing that you rediscovered the CNN Fear & Greed index independently via AI. This proves AI can bring precision to financial research

Investor and Fund Manager

Quick Questions

What was the fundamental business problem?

Manually analyzing thousands of daily financial reports took hours and was too slow to allow for quick, data-driven investment decisions.

What technology was used to understand the documents?

We built a custom data pipeline in Google Cloud, using Vertex AI Gemini 2.5 to automatically extract relevant signals and sentiment from the unstructured text of the PDF reports.

How fast is the new automated process?

The system processes over 60 pages of daily financial analysis in under 5 minutes, compared to the hours required for manual review.

How accurate is the generated sentiment score?

The AI-calculated score has a high correlation (r = 0.79) with the standard CNN Fear & Greed Index, which validates its accuracy in reflecting the market.

What is the TLDR (conclusion)?

What technologies and methodologies are involved?

Technologies: Google Cloud Run, Cloud Functions, Vertex AI Gemini 2.5, Google BigQuery, Looker Studio, Google Sheets
Methodologies: ETL Pipeline, Prompt Engineering, Map-Reduce strategy, Sentiment modeling, Business Intelligence (BI).