When Bloomberg unveiled its 2026 roadmap for ASKB on 16 April, the headline framing was the evolution of the conversational AI interface – now in beta – from a discovery tool into “a deeply integrated engine for institutional intelligence.” Beneath that framing sits a substantial set of integrations: ASKB will work directly with Portfolio & Risk Analytics (PORT), Research Management Solutions (RMS) Enterprise, alternative data including Bloomberg Second Measure, and expert intelligence content from Third Bridge, alongside an expanded ASKB Workflows automation layer.
Taken together, these integrations bind a firm’s proprietary workflows – portfolios, internal research, expert network entitlements, alternative data – into Bloomberg’s AI layer on the Terminal. The strategic effect is to make ASKB the primary query interface across a wider portion of the institutional investment process.
That positioning takes shape in an industry where the surrounding architecture is shifting quickly. Third-party data providers are adopting open protocols to make content reachable from any AI agent, in any environment. Bloomberg’s roadmap is a clear statement of where it sees its own advantage in that environment: in the depth of its data, the breadth of its integrations, and the workflows its users already run on the Terminal.The model question, clarified
One of the more interesting clarifications around the roadmap concerns BloombergGPT, the 50-billion-parameter finance-domain large language model unveiled in a March 2023 research paper. Public detail on the model’s role in Bloomberg products has been limited since then. In conversation with Market & Alt Data Insight, Wayne Barlow, Global Head of Terminal Products at Bloomberg, is clear on where BloombergGPT now sits in the ASKB architecture.
“BloombergGPT was a research model and is actually not used at all in any of our products,” he says. “Our products are built on a combination of models. We have some commercial models, some open-weight LLMs, some smaller, custom language models that we build ourselves.”
He continues: “Because we have an agentic system, different layers may use different models… each solving specific needs. Some of it relates to latency and speed, in which case you may need a smaller model. In other cases, it’s simply which model performs best for a specific task.”
The architectural choice is itself a useful signal for institutional data teams. Bloomberg has 400-plus AI engineers and published the BloombergGPT research paper in 2023; its conclusion that an orchestrated portfolio of external and in-house models is the right architecture suggests where the company sees durable advantage sitting. Not in a single proprietary model, but in the data, the entity linkages, the orchestration layer, and the distribution into existing workflows. The ASKB roadmap is, in large part, an argument that Bloomberg is well-positioned across those four dimensions.The alternative data play: Second Measure as canonical arc
Among the integrations announced, the alternative data story is one of the most operationally significant. ASKB users will be able to nowcast company KPIs using Bloomberg Second Measure credit card transaction data, with third-party alt data packages layered alongside through entitlement.
The Second Measure progression illustrates how Bloomberg has built an alt data product into Terminal-native infrastructure. Bloomberg acquired the San Mateo-based consumer transaction data startup in late 2020. In 2023, ALTD <GO> launched on the Terminal, bringing the data alongside market data, broker research, estimates, and news in a single function. In May 2024, Second Measure transaction data was made available to Enterprise customers via Bloomberg Data License. In September 2024, Bloomberg launched the Second Measure U.S. Consumer Spend Index via the ECAN function. In April 2026, the data became natively queryable through ASKB.
That is a five-year arc from acquired asset to Terminal function to Enterprise data product to branded economic index to agentic AI query primitive. For alt data vendors evaluating distribution through Bloomberg, it is the clearest available template of what end-to-end integration looks like – albeit a template for an asset Bloomberg owns outright.
One asymmetry is worth noting. Bloomberg publishes historical accuracy of Second Measure and other alternative datasets versus company-reported KPIs, accessible through ALTD documentation on the Terminal. The accuracy of the data layer is, in that sense, measured and disclosed. The accuracy of the AI layer that now consumes it is a separate question, addressed below.
Third Bridge and the question of distribution
The expert intelligence integration introduces a useful comparison.
The roadmap names Third Bridge – provider of more than 100,000 expert interview transcripts – as a showcase ASKB partner. Users will be able to surface excerpts from recent transcripts on geopolitical risks and catalysts for their holdings, alongside their other entitled content. Within the ASKB ecosystem, it is a meaningful early commitment from a major expert network.
In parallel, Third Bridge has been pursuing a broader distribution strategy. The company has built a Model Context Protocol server and describes its approach as moving “from walled gardens to open intelligence,” with content available to clients regardless of which LLM or platform they use, including ChatGPT, Claude, ModelML, and others. Through 2025 and into 2026, Third Bridge has joined a consortium with Aiera and made content available on Portrait Analytics, Hebbia, Model ML, Anthropic, Finster AI, Samaya AI, and Rogo.
In Third Bridge’s own framing, the strategic shift is a response to closed ecosystems that require clients to leave their preferred environments to access content. Bloomberg is not named in those materials, but the contrast in distribution philosophy is clear: a multi-channel approach on the content provider side, a deeply integrated single-environment approach on the platform side.
The result, for institutional clients, is genuine optionality. The same expert content is reachable through several channels: through ASKB on the Terminal; through the Aiera platform; through an in-house agent talking directly to Third Bridge’s MCP server; or through emerging research platforms such as Hebbia and Rogo. Which channel a firm prioritises will depend on where its primary research workflow sits – and that, more than the underlying data, is the strategic question the ASKB roadmap is engaging with.
Accuracy and the survey context
The roadmap was unveiled at Bloomberg’s AI in Finance Summit on 16 April, the same day the company released findings from a live audience poll of more than 100 senior decision-makers from UK financial services firms. Half of respondents (50%) identified hallucinated facts and numerical errors as their primary barrier to AI adoption; a further 27% pointed to a lack of explainability. ASKB’s design directly engages these concerns through grounding in trusted data, attribution to source documents, and transparent surfacing of the underlying Bloomberg Query Language.
Barlow describes a layered approach to accuracy that goes beyond surface attribution. “Because we have that transparency, we can actually run post-processing on any results to cut down on hallucinations,” he says. “We can check whether a number appearing in a summary actually came from an attributed source, down to a particular bullet point.”
Customer feedback is tracked with explicit categorisation. “We do track explicit feedback to make sure we’re not seeing an uptick in something that’s an error of fact, as opposed to an error of presentation or an error of relevance,” Barlow says.
Bloomberg does not currently publish accuracy benchmarks for ASKB itself. “We have internal evaluation criteria that we always aim to achieve,” Barlow says. “When a new model comes out, we check and make sure it meets that target.” But the targets and the results are not externally reported. For institutional buyers comparing ASKB against internal benchmarks or alternative platforms, that means relying on Bloomberg’s own assurances and on observed performance in pilot use, rather than on disclosed metrics. Given the asymmetry with the ALTD-disclosed data layer accuracy figures, this is one of the points in the roadmap most likely to attract follow-up from buy-side data leads.
The accuracy question matters most where ASKB Workflows is concerned. The roadmap describes scheduled or trigger-based execution: customised morning briefs, weekly thesis health checks, pre- and post-earnings reports running automatically. Asked where Bloomberg draws the line on autonomous action, Barlow describes a current scope that is deliberately bounded.
“We can imagine a world where there will be additional things these systems can do, but we expect the human to be the one directing,” he says. “Right now, those tasks are very much around gathering data, synthesising data, and bringing that data back to the user to make decisions.” The current Workflows templates, he adds, are “designed around generating something more like a report or preparation materials – pre-earnings, credit analysis. These are very specific repetitive tasks that might be multi-step.”
Data residency, data handling, and proprietary content
The PORT and RMS Enterprise integrations bring proprietary firm content – security lists, internal research notes, financial models – into ASKB’s query layer, raising practical questions for chief data officers and compliance leads about where inference happens, how that content is handled, and how data residency is managed.
Within ASKB, users opt in explicitly when bringing their own content into the platform. In those opt-in cases, that content may be used for evaluation purposes and to develop the product. Client content is not used to train or fine-tune generative AI models for the purpose of generating, displaying, summarising, or reproducing that content without additional consent.
On infrastructure, Barlow confirms a detail Bloomberg has rarely discussed publicly. “We do use multi-vendor, multi-cloud,” he says. “We have those capabilities within the AI systems to make sure we can continue to serve users over time in a rapidly changing environment.” The underlying architecture is designed to serve clients globally across regions and hyperscalers, with data residency obligations handled at product level rather than treated as AI-specific.
Buyers will still need to negotiate the specifics directly with Bloomberg: which jurisdictions are covered, where the line falls between evaluation use and product development, and how consent works for opt-in content.
Where do the agents live?
Underneath the specific features, the ASKB roadmap engages with a broader architectural question now facing institutional finance. As firms build their own AI agents – increasingly on frontier models from OpenAI, Anthropic, and Google, or on open-weight stacks they host themselves – and as data providers from expert networks to alt data vendors adopt MCP and similar protocols, the strategic question becomes where the portfolio manager’s primary AI workflow will live.
If it lives on the Terminal, ASKB is positioned to capture it. The integrations announced this month – PORT, RMS, alt data, Third Bridge, Workflows – are designed precisely to support that scenario. If primary research workflows live elsewhere, in an in-house framework or a third-party research platform, ASKB becomes one of several places a firm consults, and the value proposition shifts to the depth and quality of Bloomberg’s data and analytics within those workflows.
Bloomberg’s public position, set out across the roadmap and supporting materials, is that ASKB will be most valuable to clients who run their primary research workflows on the Terminal, where the depth of integration is greatest.
For institutional data teams, the practical implication is that the choice is not binary. A firm can entitle ASKB and use it for the workflows where Terminal-native context is decisive – portfolio analytics, BQL queries, deep document Q&A – while routing other workflows through their own agents and external data via MCP and similar protocols. The roadmap announced this month gives institutional clients new reasons to expand the first category, and the firms that benefit most will be those that have already invested in integrating their proprietary content with the Terminal.
The deeper point is that the next phase of competition in institutional AI is shaping up to be less about the underlying models and more about the data layer those models consume – its breadth, its curation, its entity linkages, and the workflows in which it is consumed. The ASKB roadmap is Bloomberg’s clearest statement to date of how it intends to compete on those terms.
Subscribe to our newsletter
