The UK’s Financial Conduct Authority (FCA) recently commissioned a study exploring how AI is likely to transform financial services in the next several years. The primary finding in the Mills Review centers around the shift from human-led, “episodic activity” towards AI-enabled, autonomous agents.
The findings underscore increasing regulatory, compliance, and governance processes that are top of mind across enterprises and larger governing bodies. The study cited that 81% of firms are adopting AI at some level, with 40% implementing more scaled or transformational stages.
The key takeaway for many investment firms that have already started scaling robust AI efforts: as autonomy grows, accountability becomes harder to trace. The need for grounded, traceable, auditable AI is greater than ever: “A useful answer is not enough if the basis for it cannot be reconstructed."
Fully verticalized solutions like AlphaSense unify enterprise content with domain-specific retrieval, a context graph, model orchestration, synthesis, and auditable outputs. Together, these components work together as one intelligence layer for trustworthy AI that investment firms leverage for decision-grade research.
Key Findings and Recommendations
The FCA study found that AI will have a “seismic” impact on financial services by 2030 with autonomous, agentic AI playing a prominent role in processing information, serving clients, and providing evidence outcomes. It is evident that firms are keen on this transformation for operational and competitive advantages.
At the same time, the review details the fundamental risks associated with mainstream adoption: hallucinations, model inconsistency, response driftage, and a failure to produce regulated, traceable responses. For financial service firms, this is a non-negotiable risk.
The study highlighted key findings and specific recommendations for the FCA Board to consider:
- Secure the Regulatory Perimeter: The study recommends the FCA launch an immediate review into the scale and impact of general-purpose Large Language Models (LLMs) such as Claude and ChatGPT that operate outside the current regulatory perimeter.
- Strengthen Coordination and Oversight: With no dedicated AI regulator in place, the study recommends the FCA should tighten coordination with national authorities and international partners among model, cloud infrastructure, and distribution providers.
- Monitor Autonomous Models: Provide clarity on accountability, governance, and consumer protection frameworks to promote dynamic model governance, monitoring and assurance, and end-to-end controls across the AI lifecycle.
- Scale the FCA AI Lab: The review calls for a structured capability within the existing AI Lab to assess AI models and systems before they become embedded in core operations, supporting "responsible growth" and addressing explainability challenges.
- Foundations for Agentic Finance: A focus on establishing the trust frameworks, permissions, and authentication needed for autonomous AI agents to operate safely.
- Agentic Supervisory Model: The adoption of an "agentic supervisory model," deploying its own AI-enabled tools for authorization, supervision, and real-time monitoring to detect system-wide risks that are not visible at the individual firm level.
- Public-Interest AI Service: Plans to convene the development of a free, inclusively designed AI-enabled service to provide consumers with reliable financial information and guidance.
The Challenge Firms Face
Enterprise AI integration is an inevitable reality. However, general model use is prone to hallucinations, insecure data storage and usability, and illogical conclusions. From the Mills Review: “General-purpose and frontier AI models can introduce challenges including opacity, model drift, data bias, hallucinations and emergent behaviours.”
As the study finds: “Firms will increasingly require more dynamic approaches to model governance, monitoring and assurance, supported by end-to-end controls across the AI lifecycle.”
According to broker research from AlphaSense, 87% of UK business leaders anticipate spending 10% or more of their total budget on AI over the next 12 months. The challenge for investment firms is no longer determining if AI infrastructure is the answer, but instead how to support enterprise integration that keeps pace with evolving technology and has the foundational protocols and compliance risk framework for highly regulated work.
The solution to this challenge is vertically integrated, decision-grade AI: intelligence that is rooted in accuracy, context, and traceability and is the gold standard for enterprises going forward.
The Gold Standard
As the gold standard for enterprise adoption, vertically integrated intelligence systems unite trusted content, retrieval capabilities, context, orchestration, and synthesis in one place. This is the standard that ensures outputs are verified, defendable, and citable, and protects against “black box” outputs identified in the FCA study.
Each component plays a critical role: Models need access to relevant information, tools, and institutional knowledge. Commonly, teams consult multiple systems, retrieving partial and potentially conflicting outputs that further complicate and fragment the process by placing the onus on model interpretation.
The context graph is built on a foundational knowledge graph that is informed on millions of documents a day, resolving entities, concepts, and the relationships between them into a structured layer beneath the corpus, before any query arrives.
An analyst querying Apple's supply chain exposure can distinguish between Apple and its relationship to other entities such as suppliers. Search is orchestrated on a structured, trustworthy foundation—a pre-assembled, verified view of Apple’s supply chain. Decision quality is therefore shaped not just by any retrieved available evidence, but the right evidence.
A vertically integrated intelligence system removes enterprise risk by bringing all these components together in one place. The result is not only a better answer, but a faster, simpler, and less costly path to that answer. And one that promotes token efficiency.
AlphaSense is built with a fully vertically integrated approach for this exact purpose, designed with high-stakes business-grade decisions in mind. It is built on the premise that trusted AI for market intelligence requires premium content, domain-specific retrieval, knowledge graph-driven context, model orchestration, synthesis, and auditable outputs working together as one intelligence layer.
A Trusted Path Forward
With AI becoming increasingly autonomous and agentic, firms are at a critical juncture on their integration journey. Supported by findings from the FCA commissioned study, the need for grounded, traceable, and auditable AI is greater than ever.
AlphaSense’s vertically integrated, end-to-end platform unites the trust and verifiability that financial services firms expect, along with the cutting-edge functionality that allows them to remain competitive in this next phase. This gold standard is the trusted path forward, one that the FCA has formally identified, that AlphaSense has devoted over a decade to building in-house. Because access was never the hard part, intelligence is.
With solutions for integrating standalone, sustainable, long-term AI adoption for high-stakes decisions, or for partnering with the leading agentic AI market intelligence platform to enhance their own bespoke AI build, the answer for trustworthy AI is clear.
