Weekly

S&P Capital IQ Pro: ChatIQ enters Excel modelling and PowerPoint generation

ChatIQ Spreadsheets

An AI modelling assistant in Excel for building, reviewing and analysing models, with refreshable formulas

ChatIQ Presentations

Generates PowerPoint presentations using Capital IQ Pro data; currently in Labs

Portfolio Insights

Identifies portfolio-return drivers and links them to news and earnings events; currently in Beta

Other updates

New DCF analysis, customisable TEV calculations, private-fund benchmarking tools, credit-memo tools and more

S&P Global Market Intelligence says in its press release that ChatIQ brings company, market, credit and macroeconomic data together with financial documents, news and research in a single conversational interface. This update focuses on extending that interface into two files analysts open every day: Excel models and presentation decks.

The release provides no user counts, accuracy measures or customer case studies. ChatIQ Presentations remains in Labs, and Portfolio Insights remains in Beta. This is a vendor announcement, so the descriptions of functionality are the company’s claims. The release specifies that the Excel assistant uses refreshable Capital IQ Pro formulas, meaning that models remain connected to the underlying data. Institutions can use this as a basis for assessing traceability.

Further reading: S&P Global press release (English) ↗

Bloomberg Enterprise MCP: giving agents a guide to what data fields mean

Positioning

An AI access layer built on Data Licence Plus for financial institutions whose agents use licensed Bloomberg data

Coverage

More than 100 million securities and over 50,000 fields

Metadata

Includes qualifying information such as exchange, currency, price type, period and as-of date, with natural-language field search

The central point of the report is that access to data does not mean an agent knows what a number represents, how it was calculated or whether it is appropriate for the task. Bloomberg’s approach is to provide field definitions and qualifying conditions alongside the data.

This is a media report on a product launch. It does not provide information on customer usage or outcomes.

Further reading: cfotech report (English) ↗

PitchBook: Data MCP and AI agents that run on a schedule

PitchBook AI

Adds Skills and Agents that turn natural-language requests into shareable deliverables and run them on a recurring schedule

PitchBook Data MCP

Lets investors’ own agents access PitchBook data directly

Roadmap

Expansion into deal sourcing, deal execution, credit research and fund analysis in 2027

The press release was issued at TechCrunch Disrupt. The interesting feature is scheduled execution: research repeated every week can become a managed, recurring workflow. This is a vendor announcement; the functionality is described by the company, and no customer case studies are provided.

Further reading: PitchBook press release (English) ↗

South Korea’s NPS: AI agents for fund research, using public data only

Approach

A cloud-based proof of concept (PoC) for research agents, running for about 4 months from June to September

Three tasks

Analysis of central-bank monetary policy, company earnings presentations and integrated equity-market research; the first two completed validation

Self-reported results

Monetary-policy research fell from about 8 hours to about 10 minutes; earnings analysis had previously required several fund managers to spend about 180 hours each per quarter

Data boundary

Public data only, with no transfer of internal management information or non-public information outside the institution

NPS describes this as the first data-analysis system built with AI agents at a public institution. According to the report, it was scheduled to share its PoC results and plans for further application with AWS in Seoul on 16 October. The report does not state that the third task—integrated equity-market research—had been completed.

The time figures come from NPS itself. No evaluation methodology was published, nor was there an explanation of how accuracy was assessed, so these should be treated as institutional self-reporting. The boundary design is worth noting: starting a PoC with public data and excluding internal information can make the security review much simpler.

We have confirmed the potential for AI to streamline research in fund operations. We will gradually expand its use so that fund managers can focus on their core investment responsibilities.

Kim Sung-joo, Chair of NPS (as reported in Chosun Biz’s English-language coverage; translated from the supplied Chinese text, not independently checked against the original English wording)

Further reading: Chosun Biz report (English) ↗

Brokerages bring investment-research capabilities to third-party AI platforms

Tencent WorkBuddy

GF Securities was the first brokerage to join in early September, with 12 in-house Skills and 9 expert capabilities; CSC Financial launched its six “Qingting Skill” capabilities in late September

Alibaba Qwen

Four brokerage Bots joined the first cohort on 7 September: Guotai Haitong’s “Lingxi”, Industrial Securities, Dongwu’s “Xiucai” and CICC Wealth

Two routes

Professional tools are distributed as Skills; conversational services for retail investors are offered as Bots

Jiemian’s review notes that a year ago brokerages were competing to develop their own large models; now they are collectively bringing specialist AI capabilities to third-party platforms. Analysts interviewed for the article see expanding customer touchpoints as the main objective. In the reporter’s test, using one brokerage’s stock-selection Skill first required downloading that brokerage’s app to grant authorisation. Account opening and trading still took place on the brokerage’s own platform.

The report also identifies two hurdles. Investment-advice wording must avoid implying returns, and the responsibilities of the brokerage and platform must be clear. The other hurdle concerns data: how information uploaded by brokerages to the platform is authorised, accessed and stored. An interviewed academic proposes the principles of “minimum necessary access, tiered classification and end-to-end traceability”. These are media reporting and interviewees’ views; individual brokerages’ specific data arrangements were not disclosed.

Further reading: original Jiemian article (Chinese) ↗

Danske Bank: employees build their own agents, with safeguards supplied by the platform

Platform

“AI City”, including an Agent Developer Kit: users describe tasks in natural language, and a declarative engine generates agents

Example

At a June internal hackathon with 700 participants, a customer adviser who had never used a terminal built 5 agents in two days for everyday work

Self-reported results

The chief AI officer says this method of deploying agents is 10 times faster than employees building them from scratch

Governance

Agents automatically include security, guardrails and compliance capabilities, with an emphasis on traceability and avoiding “shadow AI”

In the report, Danske Chief AI Officer Kasper Tjørntved Davidsen describes the project as “democratising technology and building skills across the organisation”. Employees do not need to be software developers: they set out their preferences, and the system guides them through the process.

Both the 10-times-faster claim and the hackathon example come from bank executives. The report does not disclose how many agents are currently running in production or how outcomes are measured. The organisational significance lies in the division of work: business users define requirements and own their agents, while the platform team is responsible for safeguards and compliance.

Further reading: Ingeniøren report (English) ↗

TD: agents expand from mortgage pre-underwriting into funding and discharge

New stages

Mortgage funding and discharge

What the agents do

Coordinate multi-step document checks, reconcile information across systems, flag discrepancies and generate outputs for employee review

Responsibility

Employees review outputs and remain responsible for decisions that require judgement

Self-reported results

In the pre-underwriting stage launched in May, average processing time fell from 15 hours to under 3 minutes

TD says the new capabilities aim to double employee productivity in selected mortgage workflows. All these figures come from TD’s own press release and have not been independently verified.

Further reading: TD press release (English) ↗

Cathay Financial Holdings: piloting “digital colleagues” in three settings

At its 2026 technology conference, Cathay Financial Holdings disclosed AI-agent pilots in project management, technology governance and legal-contract review, built on the GAIA 2.0 generative-AI framework introduced in 2025. The company says agents can autonomously access enterprise knowledge and tools, while humans still review critical decisions and exceptions. Controls include identity and access management, knowledge-access permissions, system-integration safeguards, cybersecurity monitoring and audit trails.

The report also notes that the announcement gives neither a timetable for deployment at scale nor quantified efficiency gains. This is a disclosure of pilots, not evidence of achieved results.

Further reading: Global Fintech Edge report (English) ↗

HKMA: introducing “ART” and previewing a supervisory framework for agentic-AI governance

Framework

“ART”: Accountability, Resilience and Trust

Actions announced

Updated supervisory guidance on responsible AI adoption; an AI training-data sharing platform developed with Cyberport and the University of Hong Kong; continued testing of agent applications through GenA.I. Sandbox++

Forthcoming

A discussion paper on a supervisory framework for agentic-AI governance, exploring “intent-anchored accountability”, a register of material AI agents and effective intervention mechanisms

Sandbox results

Technical trials for the second cohort of the GenAI sandbox have been completed; more than 60% of use cases explored agentic AI, according to HKMA

HKMA notes that AI systems are becoming more autonomous and can access more data, tools and workflows. Its supervisory focus is therefore on accountability, operational resilience and trustworthiness. The discussion paper has not yet been published; the register and intervention mechanisms remain areas for exploration.

Read the original: Hong Kong Government press release (English) ↗

FSB: finalising 12 sound practices for the responsible adoption of AI

Content

12 practices in two groups: organisation-wide AI governance and risk management across the AI lifecycle

Governance

Board and senior-management oversight of AI’s alignment with business models and risk appetite; clear responsibilities; integration of AI risks into risk-management frameworks and retention of usage records; continuous learning

Lifecycle

Use-case risk assessment, technology selection, data governance, explainability, performance evaluation and human oversight, among other areas

Positioning

A practical toolkit adaptable to each institution’s circumstances, which the FSB encourages boards and senior management to consider

This finalises the consultation draft issued in June. One of its aims is to promote cross-jurisdiction coordination and information-sharing between financial institutions and supervisors.

Read the original: FSB press release (English) ↗