Agentic treasury runs on searchable, observable context

BNY’s article makes an important distinction: Treasury does not necessarily suffer from a lack of data. It suffers from the time and effort required to turn that data into a decision.

Treasury teams already have access to balances, transactions, forecasts, payment activity, liquidity positions, counterparty information, and market signals. The problem is that this information often remains fragmented across enterprise resource planning systems, treasury management systems, banking platforms, dashboards, reports, and spreadsheets.

According to BNY, this creates decision latency. The information may exist, but people still have to find it, reconcile it, interpret it, and determine what to do next. More dashboards can improve visibility without necessarily accelerating that process.

That observation resonated with me because it extends well beyond treasury and because I have seen the issue from several sides: within a global financial institution, within one of the world’s largest enterprise software companies, and now through the lens of search, observability, and AI at Elastic.

Across financial services, institutions have spent decades building highly reliable systems of record. Those systems are indispensable. They provide the controls, transaction processing, accounting integrity, and governance upon which the industry depends.

My years at SAP gave me a deep appreciation for the role ERP systems play in creating a trusted operational and financial foundation. CFOs and treasury leaders depend on them to manage core processes, maintain financial integrity, and establish a consistent view of the enterprise.

But even the most capable ERP or treasury platform operates within a much larger technology environment. Financial institutions also rely on banking platforms, payment systems, market-data services, risk applications, data pipelines, documents, communications, and an expanding range of cloud and AI services.

The challenge is therefore not that systems of record have failed. It is that they were not necessarily designed to interpret changing conditions across this entire environment in real time.

BNY proposes an intelligence layer that sits above existing ERP, treasury management, and banking platforms. Rather than replacing those investments, this layer would connect information across them, apply context, recognize changing patterns, and help treasury teams move from reporting what happened to anticipating what may happen next.

That approach is important. Having spent years in the ERP space, I am skeptical of any technology narrative that begins by suggesting institutions should discard the foundational systems on which their businesses run. The more credible opportunity is to extend those systems by making their information and the information surrounding them more accessible, contextual, and actionable.

This is where I began thinking about the BNY article through an Elastic lens.

An intelligence layer can only be as effective as the context available to it. Before an AI application or agent can recommend an action, it needs to retrieve relevant information from across a highly distributed environment.

That context may include:

• Account balances and transaction histories

• Payment messages and settlement activity

• Liquidity forecasts and positions

• Accounts payable and receivable data

• Market and foreign exchange signals

• Counterparty documentation

• Treasury policies and approval requirements

• Application events, logs, metrics, and alerts

Some of this information is structured. Some is unstructured. Some arrives as a continuous stream of time-sensitive events. Some may be represented as vectors to support semantic retrieval.

Elasticsearch is built to store and search structured, unstructured, vector, geospatial, and time-series data in real time. It can provide hybrid retrieval across enterprise information, combining traditional lexical search, semantic search, filters, and analytical queries to surface relevant context for applications, employees, and AI agents.

That makes search more than a tool for finding documents. In this context, search becomes part of the decision architecture.

Consider a treasury team that sees an unexpected deterioration in its projected cash position. The immediate question is not simply, “What is the current balance?”

The team may also need to understand:

• What changed since the previous forecast?

• Which expected payments have not arrived?

• Did customer behavior change?

• Was a settlement delayed?

• Did a source system fail to deliver current data?

• Is the projected shortfall real, or is the underlying information incomplete?

• What internal policy governs the proposed response?

• Has the institution encountered a similar situation before?

Answering those questions requires retrieving context across ERP data, treasury records, financial transactions, operational systems, historical events, and enterprise knowledge.

That is where a search and analytics platform can help provide the foundation beneath the intelligence layer BNY describes.

The article also prompted me to think about observability.

Observability is usually discussed as an IT discipline: monitoring infrastructure and applications, investigating failures, analyzing logs, tracing transactions, and maintaining system reliability.

But in an increasingly automated treasury environment, operational telemetry becomes part of the business context.

Imagine that a treasury intelligence application identifies a potential cash shortfall and recommends moving funds. Before acting, the treasurer needs confidence that the signal reflects an actual financial condition.

What if the apparent shortfall was caused by:

• A delayed data pipeline

• A failed banking API

• An incomplete payment feed

• An ERP synchronization issue

• A stale market-data source

• A problem with the application’s retrieval workflow

• A failed call between an AI agent and one of its tools

The difference between a financial anomaly and a technology failure can be material.

This is where the boundaries between search and observability begin to converge. The institution must be able to examine both the financial signal and the health of the systems that produced it.

The question is no longer only whether the application is operating properly. It is also if I trust the business decision this application is asking me to make.

Elastic Observability brings together logs, metrics, traces, application performance data, infrastructure information, and other telemetry to help teams identify abnormal behavior and investigate its cause. In an AI-enabled treasury architecture, that visibility can help institutions understand whether a recommendation was affected by missing data, system degradation, application latency, or a failed dependency.

BNY describes agentic treasury as the next stage of this evolution.

In this model, AI agents may continuously monitor liquidity and payments, identify anomalies, interpret changes, recommend responses, and in selected circumstances, execute predefined actions within established governance frameworks. BNY is careful to position this as an augmentation of treasury professionals rather than a removal of human accountability.

That distinction is critical.

The closer AI moves to consequential financial action, the more important it becomes to understand how a recommendation was produced.

Institutions will need evidence, such as:

• Which data sources were consulted

• Whether those sources were current and complete

• What information the agent retrieved

• Which models, tools, and workflows were involved

• Whether any application or data-processing errors occurred

• Which policies or thresholds were applied

• When human approval was requested or granted

• What action was ultimately taken

The governance framework for agentic treasury will extend across treasury policy, model risk management, identity and access controls, approval processes, regulatory requirements, and institution-specific business logic.

Search and observability can provide important elements of the supporting foundation: relevant context, operational evidence, searchable telemetry, anomaly detection, and a traceable record of how the underlying technology behaved.

One of the strongest points in BNY’s article is that institutions do not have to move immediately from dashboards to autonomous execution.

The journey can be progressive.

A treasury team might begin by unifying search and analytics across fragmented information. It might then introduce AI-assisted investigation followed by recommendations supported by relevant evidence. Human-approved workflows could come next with selective automation introduced only for clearly defined actions operating within established controls.

That is a much more credible path than treating agentic AI as a switch that can simply be turned on.

In financial services, trust is built by proving that systems behave reliably, decisions can be explained, and people retain appropriate control. It is the familiar principle of “trust, but verify” made essential in an agentic world.

My Elastic take on BNY’s vision is this: Agentic treasury will require a real-time context layer, but it will also require an observable decision architecture.

ERP, treasury management, and banking platforms remain essential systems of record. Search can help AI applications and treasury professionals access and connect the relevant information within and around those systems to understand changing conditions.

Observability can help establish whether the applications, data pipelines, models, and services producing those insights are operating as intended.

Together, they can help institutions move not only from data to decisions, but also from decision latency to decision confidence. BNY is right that the future of treasury is not simply greater visibility. The real opportunity is continuous, contextual, and governed decision-making. But before institutions allow AI agents to act, they must be able to answer a foundational question: Can we see and trust how the decision was made?

Get in touch to learn more about how Elastic can support your autonomous treasury journey.