AI that draws conclusions from many documents at once has to rank news and company filings differently. Recency matters most for news, document type matters more for filings, and perspective matters across both. Bloomberg’s answer is ASKB, a conversational AI interface for the Terminal built on specialized agents rather than one massive model.
In an interview with Hedge Fund Alpha, Wayne Barlow, Global Head of Terminal Products at Bloomberg, explained how that works: attribution down to the excerpt, guardrails against hallucinations, and where his team is deliberately holding agents back.
The ASKB (Beta) interface on the Bloomberg Terminal. Source: Bloomberg.
Building up to ASKB
Barlow has worked in finance and technology for more than 30 years, including the past 20 at Bloomberg. The Terminal Products team has looked at how agentic AI can transform its products, and Bloomberg has rolled out incremental tools along the way: AI-powered earnings and news summaries, and a Document Insights tool that answers questions about a document.
“We developed multi-document analysis, in a way that we knew would build up to ASKB, the conversational AI interface for the Bloomberg Terminal released in beta back in the beginning of this year,” Barlow said. “Our conversational interface brings together Bloomberg’s data analytics and other features into a common user experience”
Wayne Barlow, Global Head of Terminal Products at Bloomberg. Source: Bloomberg (used with permission).
Transparency and attribution
Barlow said one feature users have consistently asked for is transparent attribution for every answer.
“We want to make sure that we are grounding our AI in things like transparency, so that when you ask a question, those answers have citations. Whether it’s citations to structured data by giving users access to the Bloomberg Query Language (BQL) formula they can use to get that data, or citations to the original unstructured document source like a news story or broker research note,” he explained. “That gives our users the confidence that they can look at a statement that shows up in our interface, click into the underlying document, and see highlights to the relevant excerpt.”
Bloomberg is also extending that connectivity to sell-side analysts whose research is posted on the Terminal. Many of them are Terminal subscribers, and buy-side readers can message them directly through Instant Bloomberg with questions about their research.
Client data and Custom Fundamentals
With the Terminal pulling in so many documents and data sources, a natural question is whether Bloomberg has to get into the data-cleaning business. Barlow said it does not. Customers can upload files through Bloomberg’s data environment, where the company says its AI handles unstructured content well, and client material is securely segregated from the Terminal’s existing data.
Login required to continue reading.
Setup a free account to get access to this article (no credit card required).