---
title: "Bloomberg's $500M AI Secret: Start With Problems, Not Tech"
description: "5 source-backed AI claims from Bloomberg's $500M AI Secret: Start With Problems, Not Tech, with key statistics, context, and the original AI Adopters Club…"
url: "https://kbanc.com/claims-library/bloombergs-500m-ai-secret-start-with-problems-not-tech"
source: "https://aiadopters.club/p/bloombergs-500m-ai-secret-start-with"
date: "2025-04-03"
topics: ["strategy", "business", "implementation"]
generated: "2026-08-31"
---

# Bloomberg's $500M AI Secret: Start With Problems, Not Tech

By Kamil Banc | April 3, 2025

## Claims

1. **Early AI Adoption Since 2009** (source summary): Bloomberg began developing practical AI applications as early as 2009, including a Federal Reserve sentiment model.
2. **99% Extraction Accuracy** (source summary): Bloomberg automated data extraction from financial documents, achieving over 99% accuracy in the process.
3. **24 Hours to One Minute** (source summary): The automated extraction system cut document ingestion time from 24 hours to under one minute.
4. **BloombergGPT Model Specs** (source summary): BloombergGPT is a 50-billion parameter LLM trained on 700 billion tokens of financial data.
5. **Problems Before Technology** (source summary): Bloomberg's strategy identified critical information needs first, then built technology to solve those problems.

## Evidence

### Quote
> "They didn't start by asking 'how do we use AI?' They asked 'what information do financial professionals desperately need that's currently impossible to get?'" - Kamil Banc

### Key Statistics
- **Over 99% accuracy**: Accuracy achieved by Bloomberg's automated data extraction from financial documents
- **24 hours to under a minute**: Reduction in document ingestion time after Bloomberg automated data extraction
- **50 billion parameters**: Size of the BloombergGPT large language model
- **700 billion tokens**: Amount of financial data used to train BloombergGPT

## Context
The article presents Bloomberg's AI journey as a case study in problem-first adoption, drawing on publicly reported milestones such as the 2009 Federal Reserve sentiment model and the later BloombergGPT release. The author, Kamil Banc, frames these facts as evidence that sustained domain expertise and deliberate data collection outperform trend-chasing AI strategies. Practitioners can apply the core lesson by auditing which information needs in their organization are currently unmet before selecting any AI tool. The piece also previews a framework for deciding what to automate versus augment, though the full playbook is gated behind a paid subscription. Readers should note the analysis is interpretive commentary built on Bloomberg's disclosed achievements rather than an independent audit.

## Source
- Original: [Bloomberg's $500M AI Secret: Start With Problems, Not Tech](https://aiadopters.club/p/bloombergs-500m-ai-secret-start-with)
- Cite: kbanc.com/claims-library/bloombergs-500m-ai-secret-start-with-problems-not-tech