AI in Asset Management Accelerates: Singapore Survey Reveals Investor Team Leadership and the Rise of Vertical Tools

Overview

This report is based on the proceedings of Bernstein’s “Future Tech Conference” held in Singapore, focusing on the current state, challenges, and future trends of generative AI applications in the asset management industry. Through interviews with Jimin Choi, AI Strategist at J.P. Morgan Asset Management, and Jacob Choi, Co-founder of the vertical‑AI platform LinqAlpha, the report highlights regional differences—between Singapore and London/Hong Kong—in AI adoption pathways, illustrates how large institutions are scaling AI from proof‑of‑concept to production‑grade deployments, and examines the unique value of domain‑specific AI tools compared to general‑purpose large models. The key takeaway is that AI is evolving from a mere productivity enabler into an “agent” system capable of both augmenting and autonomously generating investment insights; however, at present, commercial returns remain largely confined to efficiency gains and risk‑management improvements.

Core views

Regional Differences Are Marked: Singapore Exhibits “Investment-Driven” Adoption and a Strong Preference for External Procurement. Survey data reveal that Singapore’s asset management firms follow a markedly different AI integration path compared to London and Hong Kong. First, the sources of driving force differ: in Singapore, 57% of AI projects are led by investment teams, whereas in London (61%) and Hong Kong (52%), technology teams take the lead. Second, maturity remains at an early pilot stage: 41% of Singaporean firms are still in the pilot phase, with only 24% having deployed use cases into production—lower than in London (51%) and Hong Kong (52%). Finally, deployment strategies diverge: Singaporean firms show a pronounced preference for purchasing third-party generative AI solutions (68%), contrasting sharply with the “hybrid self-built plus external procurement” approach commonly adopted in London and Hong Kong. Moreover, Singaporean users tend to employ fewer AI tools (54% utilize one or two), yet their preferences for mainstream tools align closely, with Copilot, ChatGPT, and Claude ranking as the top three. Large‑Scale Institutional Practices: JPMorgan Moves from Efficiency Optimization to Autonomous Agency. JPMorgan has unveiled its enterprise‑level AI strategy, the “LLM Suite,” which has been rolled out to over 250,000 employees worldwide, achieving daily usage rates exceeding 50%. The bank’s AI initiatives have narrowed from an initial portfolio of 600–700 ideas to roughly 50 high‑ROI strategic priorities. Currently operational high‑impact use cases include: (1) AI‑enhanced guidance, which automatically extracts and annotates client investment preferences; (2) Proxy IQ, which analyzes voting data from more than 3,000 annual meetings, replacing external proxy advisory services; (3) Research Assistant, leveraging vast internal and external datasets to compare market views and synthesize emerging themes; and (4) Intelligent Monitoring, delivering personalized daily investment briefings. Looking ahead, the architecture will likely converge toward a small number of centralized “agent layers” deeply embedded in organizational context, governed by central teams while allowing portfolio managers to customize workflows. The Value Proposition of Vertical AI: Why LinqAlpha Matters Beyond Claude Alone? According to LinqAlpha’s founder, general‑purpose large models like Claude face limitations when addressing finance‑specific use cases. Vertical platforms establish competitive moats through three key advantages: (1) Accuracy and Data Standardization: Financial data originate from multiple vendors—such as S&P and FactSet—with varying citation conventions and licensing constraints. Vertical platforms integrate these heterogeneous data sources into unified knowledge graphs, mitigating the confusion that arises when general‑purpose models query across asset classes; (2) Cost Control: Directly invoking APIs to build custom solutions can be prohibitively expensive and highly volatile due to token consumption. In contrast, vertical platforms absorb model‑related cost variability, offering more predictable enterprise‑level pricing; and (3) Internal Context Integration: These specialized tools can seamlessly incorporate a firm’s proprietary data—emails, internal notes, and more—enabling truly personalized insights beyond mere retrieval of publicly available information. Measuring ROI and Future Skills: From Productivity to Judgment. At present, it remains premature to quantify the direct alpha (excess returns) generated by AI; instead, ROI is primarily assessed through cost avoidance, risk reduction, productivity gains, and automation levels. As AI adoption deepens, the role of portfolio managers will evolve—from information consumers to architects of systems and agent frameworks—where core responsibilities shift to defining signals, managing uncertainty, and intervening when AI systems falter. Consequently, “judgment”—the ability to distinguish signal from noise, detect consensus errors, and seize opportune moments—will emerge as the most critical skill for investors in the AI era. Meanwhile, the industry is transitioning from passive querying to round‑the‑clock autonomous agency, making the design and collaboration of such agents essential for capturing future investment value.

Analysis framework

The research report employs an analytical approach that combines on-site fieldwork, expert interviews, and cross‑market benchmarking. 1. **Survey-Based Comparison**: By hosting similar conferences in Singapore, London, and Hong Kong and conducting real‑time on‑site voting, the study gathers quantitative data on AI adoption rates, driving factors, maturity levels, and tool preferences, thereby highlighting the structural differences across financial centers in their respective AI implementation pathways. 2. **In‑Depth Case Study**: Taking industry leader JPMorgan as a representative of large institutions, the report meticulously dissects its end‑to‑end deployment—from infrastructure development (LLM Suite) to specific use cases (Proxy IQ, Research Assistant)—illustrating the closed loop from ideation to production. 3. **Product Value Assessment**: Drawing insights from the perspective of LinqAlpha, a specialized AI startup, the analysis contrasts general‑purpose large models with domain‑specific financial AI in terms of data governance, cost structures, and contextual understanding, underscoring the necessity of vertical tools and addressing the market’s question: “Why are dedicated tools still needed when powerful general‑purpose models already exist?” 4. **Forward‑Looking Scenarios**: Based on current technological bottlenecks (such as hallucinations and data silos) and business needs (including alpha generation), the report projects that future AI architectures will evolve from fragmented applications toward a unified agent layer, and accordingly anticipates the reconfiguration of professionals’ skill sets.

Methodology notes

• Industry/Industrial Analysis FrameworkSupply-and-Demand Framework

  The adoption of AI in the asset management industry is constrained by the quality of data supply and the ability to integrate data, rather than by demand or model capabilities alone.

  The research report points out that the rigorous standardization (such as master data for assets and fiscal calendars) and orchestrated retrieval of public, licensed, and proprietary data are bottlenecks for AI applications. This underscores that, in the financial‑AI space, the supply of high‑quality structured data is even more critical than the models themselves, aligning with the industry adage that “data, not models, is the limiting factor.”

• Competition and Strategic FrameworkMoat / competitive advantage

  Vertical AI tools build competitive barriers through data standardization, licensing management, and internal context integration.

  Research reports suggest that general-purpose large models cannot address the challenges of multi-source, heterogeneous financial data and the complexities of regulatory compliance. In contrast, vertical platforms such as LinqAlpha—by leveraging a pre‑validated data layer and embedded compliance controls—deliver accuracy and security that are difficult for general models to replicate, thereby establishing a distinct competitive advantage.

• Company Fundamentals and Financial FrameworkFree cash flow analysis

  The ROI metrics for AI investments are diversified and extend beyond direct revenue growth.

  The research report notes that, at present, AI’s ROI is primarily assessed in terms of cost reduction or avoidance, risk mitigation or prevention, productivity gains, and automation—rather than by directly quantifying the alpha generated by AI. This suggests that, when evaluating the returns on AI investments, investors should focus on the implicit cash‑flow contributions stemming from improved operational efficiency, rather than solely on top-line revenue growth.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• J.P. Morgan (JPM)

  An industry benchmark, demonstrating how large asset management firms systematically deploy AI and transition it from pilot projects to full-scale production.

  Strengths

  It boasts a robust internal data infrastructure, a unified LLM Suite platform, and a well-defined ROI measurement framework; it has successfully launched high-impact production‑grade applications such as Proxy IQ.

  Comparison

  Compared with smaller and mid-sized institutions, JPM enjoys advantages in resources and data scale, enabling it to shoulder the substantial upfront capital expenditures required for infrastructure.

• LinqAlpha

  A leading vertical AI tool that addresses the pain points of general-purpose large models in financial data processing.

  Strengths

  It offers multi-source data standardization, built-in compliance and permission management, and a cost‑controlled enterprise‑grade pricing model, while focusing on deep contextual integration within the financial sector.

  Comparison

  Compared with general-purpose models such as Claude, LinqAlpha offers distinct advantages in accuracy, data governance, and the seamless integration of internal context, making it well-suited for institutional investors with extremely high standards for data quality.

  Risks

  Market competition is intensifying, and major tech companies may launch comparable vertical‑specific solutions.

Key data

• Driving Forces Behind Singapore’s AI Initiatives57%Led by the investment team, this proportion is higher than that led by the technology team (33%).
• Singapore’s AI Solution Preferences68%Preference for purchasing third-party solutions is significantly higher than for in-house development or hybrid models.
• Proportion of AI Production Environments in Singapore24%Only 24% of respondents reported that use cases have already entered production, compared with 51% in London and 52% in Hong Kong.
• JPMorgan LLM Suite Coverage Rate>50%Among its more than 250,000 employees worldwide, over half use the platform on a daily basis.
• Contribution of Vertical AI Models30-40%The model itself accounts for only 30–40% of answer quality; the remaining 60–70% hinges on “harness” engineering, including data ingestion, retrieval, and evaluation.

Impact & implications

For the asset management industry, the pervasive adoption of AI is reshaping workflows and value‑creation processes. In the short term, AI primarily serves as an efficiency tool, reducing operational costs and freeing analysts to focus on higher‑value tasks by automating document processing, meeting transcription, and data aggregation. Over the medium term, as “unified agent‑layer” architectures mature, AI will become more deeply embedded in investment decision‑making—generating automated reports on investment thesis drift or event‑driven triggers—requiring institutions to strengthen their data governance capabilities and develop robust internal knowledge‑graph frameworks. For technology service providers, pure‑play general‑purpose large‑model vendors may face competition from specialized players in vertical domains, who, with a deep understanding of financial data standards, regulatory compliance, and workflow integration, are better positioned to meet institutional clients’ stringent demands for accuracy, security, and interpretability. As for practitioners, proficiency in using AI tools is merely foundational; the shift in core competencies means that “judgment” and “agent‑system design expertise” will increasingly distinguish top‑tier investors from their peers.

Risks

• The risk of hallucinations in AI-generated content, particularly when dealing with specific financial figures, necessitates validation through deterministic rule-based systems.
• Data privacy and leakage risks remain, and even with the adoption of retrieval-augmented generation (RAG) and dynamic context injection, stringent approval procedures are still required.
• Overcoming cultural resistance: AI adoption is often viewed as a cultural challenge rather than a technological one, and the transition from employee denial to acceptance can be a lengthy process.
• The uncertainty inherent in ROI measurement currently makes it difficult to directly quantify AI’s contribution to alpha generation, which may lead to volatility in long-term investment appetite.

What to watch

• The evolution of AI pricing models: progress from charging per token or seat to charging based on outcomes, such as per report or per signal.
• The extent of adoption of Autonomous Agents: Assess whether more institutions are shifting from passive query-based approaches to agent systems that provide round-the-clock monitoring and proactive signal delivery.
• The competitive‑cooperative relationship between vertical AI and general‑purpose large models: Pay close attention to whether major cloud providers will launch industry‑specific data layers or middleware tailored for the financial sector, thereby eroding the competitive edge of vertical‑focused startups.
• Regulatory policy changes: in particular, the regulatory classification of AI-generated content in investment advice and restrictions on cross-border data flows.