Agentic AI surged to the top of enterprise agendas in 2025. Yet while enthusiasm was sky‑high, many organizations struggled to bridge the gap between vendor vision and the practical progress business and technology leaders expected. At Forrester, we’ve documented the evolution from deterministic workflows to agent‑ish systems and, ultimately, to fully agentic architectures. Industries such as customer service and software development have moved fastest — but financial services has often been mischaracterized as slow to adopt agentic.

That perception no longer holds.

Financial services firms are building agentic capabilities tailored to the sector’s unique demands for precision, compliance, auditability, and reliability. Early deployments now demonstrate how agent‑based systems can augment — and eventually transform — high‑stakes analytical and decision‑support work.

Reliable “Agent‑ish” Systems Require Substantial Investment — and Bloomberg Shows How

Bloomberg’s new AskB agent represents an advanced example of agentic AI designed specifically for a regulated, data‑intensive domain. AskB serves as a conversational front‑door to the Bloomberg Terminal, reflecting a broader trend of enterprises consolidating complex workflows behind a unified agent interface. Under the hood, AskB is a multi‑agent system composed of domain‑specialized retrieval agents coordinated by an orchestrator. This architecture enables precise cross‑domain information access across structured and unstructured Bloomberg content. Additionally, this approach enables each data product sourced by AskB to be independently updated and domain-specific quality maintained.

Today, AskB remains primarily agent‑ish: it retrieves and synthesizes information but does not yet take autonomous actions. It does enable actions between humans for enhanced decision making on data. For example:

  • An equity analyst uses AskB to analyze a company.
  • AskB retrieves a relevant research contribution and identifies the analyst who authored it.
  • The system facilitates collaborative follow‑up between the user and the researcher — enabling clarifications, data validation, or deeper analysis within the workflow.

This is an example of an emerging model of human–agent collaboration in financial services.

AskB Is Not Alone  — The Financial Services Ecosystem Is Moving But Evaluation Remains A Bottleneck

AskB’s development also shows some of the limitations and fundamental challenges in developing agentic applications – namely the resources needed to ensure that the AI agents are behaving properly. Today there are fundamentally two approaches: human-driven evaluation, and using an LLM-as-a-judge. Bloomberg, like many others developing agentic AI, took the approach of primarily human expert-driven evaluation. While this approach can ensure achieving higher quality and accuracy, the cost of those evaluators in highly specialized domains like financial services, healthcare, or law can quickly become unsustainable – and human evaluations can take time. However, the alternative of using an LLM-as-a-judge introduces its own costs and constraints in aligning it to effectively and transparently judge outputs.

Most regulated enterprises will need hybrid evaluation strategies to balance accuracy, cost, and speed.

While Bloomberg’s AskB is a notable entrant into this space, other companies working with financial data are quickly introducing their own agentic capabilities. Experian has introduced their AI Assistant which has dedicated sub-agents that can be applied to workflows like credit risk and fraud detection. While Experian does use human experts to evaluate their AI agent accuracy, they also integrate an automated loop which continuously tests and evaluates responses from the agents – hybridizing between the human-driven and model-driven approaches to agentic evaluation. Mastercard is also advancing the agentic space for their B2B customers and users, introducing documentation for how to build agents leveraging MCP connections and LLM prompting within their platform.

What It Means: Financial Services Firms Must Design Agentic Systems for Reliability From Day One

As financial institutions explore agentic AI, success will hinge on:

  • Robust evaluation frameworks blending human review, automated testing, and LLM‑as‑judge approaches
  • Clear orchestration and role definitions for sub‑agents
  • Strong governance and oversight for any action‑taking capabilities
  • Controlled data-access pathways with logging, lineage, and auditability
  • Human‑agent interaction models that preserve accountability while accelerating insight

Ready to Build Your Agentic Strategy?

Forrester clients exploring agentic architectures — and the evaluation, governance, and orchestration patterns required to deploy them safely in financial services — can schedule a guidance session to dive deeper into emerging best practices.