Every day, the Bloomberg Terminal processes 450 billion data points and 1.1 million news items — the central nervous system of global finance. Now, Bloomberg is channelling that data advantage into ASKB, a conversational agentic AI built not for speed alone, but for the accuracy and trustworthiness that high-stakes investment decisions demand. Amanda Stent, Head of AI Strategy and Research at Bloomberg, explains how the firm is engineering ground truth into every AI response.
Humans at the Centre of Financial AI
Despite the automation surge sweeping through financial services, Amanda Stent is resolute: keeping humans in the loop is not a limitation — it is a competitive edge. AI excels at eliminating the drudgery of repetitive data retrieval, but it lacks the discernment that defines a veteran analyst.
As Stent explains, markets are shaped by data, incentives, regulation, and psychology. A model can detect a pattern — but it takes an experienced human to judge whether that pattern represents a meaningful signal or simply noise. The true value of AI, in her view, lies in how it empowers the person behind the screen: while AI handles retrieval and synthesis, financial professionals are freed to focus on interpretation and strategy.
The analyst of tomorrow, Stent argues, will be defined not by coding ability but by critical thinking — knowing which questions to ask, which anomalies to challenge, and which assumptions to stress-test. Bloomberg's AI systems are designed to support that decision-making process without replacing it, and are explicitly not designed to offer financial advice.
"A model can detect a pattern, but it takes an experienced human to discern if it's a meaningful signal or just noise."— Amanda Stent, Head of AI Strategy and Research, Bloomberg
Constructing a Ground Truth with ASKB
The most persistent problem with generative AI in professional environments is hallucination — models inventing facts with unwarranted confidence. Bloomberg's answer is ASKB, a conversational AI interface tethered entirely to Bloomberg's proprietary data fortress rather than relying on a general-purpose LLM's world knowledge.
Every news article, research document, and social media post ingested into ASKB is automatically enriched with metadata — tagging entities, sentiment, and salience — so the AI doesn't merely read text but understands context. The underlying challenge, as Stent frames it, is that standard large language models have no ground truth of their own. Bloomberg's role is to give them one: decades of structured analytics, curated company documents, and alternative data sources ranging from credit card spend to foot traffic.
To enforce this, Bloomberg has built a layered system of domain-specific validators and guardrails — deterministic checks that verify whether a calculation was called with the correct parameters, whether a paragraph summary faithfully reflects only the source material, and whether every output is accurate and timely before it reaches the user. These checks are continuously expanded with the input of Bloomberg's own subject matter experts.
Transparency is baked in at the output level too. ASKB provides full source attribution for every answer, including the Bloomberg Query Language code underpinning any data point — allowing users to export that exact logic into Excel or BQuant for independent verification.
Privacy by Design
In a market where a leaked document can move prices, data privacy is non-negotiable. Bloomberg has engineered ASKB with strict boundaries ensuring that client data never becomes training material for external AI models.
Bloomberg does not use client content to train or fine-tune generative AI models for purposes of generating, displaying, summarising, or reproducing such content without explicit additional consent. The system applies fine-grained role-based access controls across all content available to AI systems — including user-uploaded files, which remain accessible only to the individual who uploaded them within the specific ASKB thread where they were shared.
Where third-party model providers are involved, their access and handling of data are governed by contracted terms with strict restrictions on access, data usage, and retention — ensuring that neither client nor proprietary Bloomberg data is ever used to train or fine-tune external LLMs.
"With ASKB, Bloomberg is focused more on accuracy than speed — something our clients seem to appreciate, regardless of whether they're an equity analyst, portfolio manager, or credit analyst."— Amanda Stent, Head of AI Strategy and Research, Bloomberg
The Power of a Multi-Agent Architecture
Rather than relying on a single large, general-purpose model, ASKB deploys a multi-agent architecture — a coordinated team of specialised AI agents, each optimised for a distinct data type: news, filings, calculations, and more, working in parallel. This design enables hyper-specific functionality that a monolithic model simply cannot deliver.
Stent explains the logic clearly. When searching news, recency matters above all — blending recent and historical articles in a single model would dilute the latest breaking signals. When searching company documents, type and perspective both matter: a 10-K, a corporate presentation, and analyst coverage each offer complementary but distinct viewpoints that clients need to be able to separate and compare.
In response to a query, ASKB orchestrates the appropriate agents and tools across a library of more than 400 million documents, running discovery, synthesis, and calculation tasks in parallel. The ultimate goal is to eliminate the swivel-chair effect — the hours analysts spend jumping between applications to manually piece together research — replacing it with contextually accurate, sourced, and actionable answers that bridge the gap between a question and a fully formed investment thesis.