EY India says agentic AI could lift treasury forecasts

EY India has published a playbook on the use of agentic AI in corporate treasury, arguing that treasury teams could improve liquidity forecast accuracy to as much as 90%.

The study presents treasury as one of the clearest examples of how companies are struggling to move artificial intelligence projects beyond the pilot stage. It identifies fragmented data, spreadsheet-based workflows and weak governance as the main barriers, rather than limits in the underlying technology.

Treasury departments continue to devote a large share of their time to manual work. The findings show that teams still spend 60% to 70% of their bandwidth on tasks such as data aggregation, validation and reporting, leaving less time for strategic planning and risk management.

Spreadsheet use remains deeply embedded in the function. A mature treasury operation may run between 50 and 100 connected spreadsheets across cash positioning, foreign exchange exposure, investments and regulatory reporting.

That dependence has consequences for financial planning. In spreadsheet-led treasury environments, forecast variance often exceeds 20%, affecting liquidity decisions and the timing of funding needs.

Manual burden

More than half of corporates globally still rely on manual reconciliation processes, according to the report. It links that reliance to slower workflows, higher operational risk and weaker visibility over cash positions.

The analysis identifies workflow transformation as the first step in broader AI adoption. Treasury functions that have introduced digital breaks and workflow automation are already recording auto-match rates of 80% to 90% in reconciliation processes.

The report presents agentic AI as a tool for handling specific treasury tasks where data flows are repetitive and decision rules can be defined. It highlights cash forecasting, reconciliation, and KYC and AML exception handling as the most practical starting points for early adoption.

Among those areas, cash forecasting is described as the use case with the greatest business impact. AI-enabled treasury models can reach up to 90% forecasting accuracy across 30-day, 60-day and 90-day liquidity horizons when backed by stronger data foundations.

KYC and AML exception handling is another area the report identifies as suitable for automation. AI agents can manage 70% to 80% of routine cases with full auditability, allowing treasury and risk teams to focus on more complex issues.

Data foundation

A recurring theme in the playbook is that AI projects fail to scale when underlying data remains inconsistent or poorly governed. EY India argues that organisations need a reliable architecture before introducing more automated decision-making into treasury operations.

Its recommended model centres on a treasury data lake as a single source of truth. The structure would bring together structured and unstructured information from ERP systems, banking platforms, contracts, emails and market data in a common environment.

The report also calls for clearer workflow design and stronger governance controls. In this view, value comes less from adding another software layer and more from reducing fragmentation in how treasury data is collected, reconciled and used.

Economic uncertainty is another factor driving interest in the area. Companies are under greater pressure to manage liquidity risk more closely while meeting demands from finance leaders for near real-time visibility into cash and exposures.

That has raised treasury's profile within finance departments. Rather than serving mainly as a back-office control function, treasury is increasingly expected to provide faster insight on funding, risk and short-term balance sheet decisions.

Operating model

To support longer-term change, EY India advocates setting up a Treasury Centre of Excellence. The proposed unit would oversee data lake pipelines, workflow libraries and governance frameworks across the treasury operation.

The report draws on EY India's work with treasury functions in manufacturing, financial services and infrastructure. It combines observations on operating models, process inefficiencies and patterns in early AI adoption to outline a practical route for chief financial officers and treasurers.

Hemal Shah, Partner, Risk Consulting, EY India, said many treasury teams still rely heavily on spreadsheet-based processes at a time when organisations are seeking greater visibility, agility and control.

"Agentic AI presents an opportunity to move treasury from a reactive function to a predictive and intelligent operating model. However, realising this potential will require strong data foundations, robust governance and clearly defined workflows," Shah said.