AI in Treasury Operations: Cash, Exposure and Control

Treasury operations begin each day with an exercise in reconciliation: assembling an accurate cash position from statements that arrive in different formats, at different times, with different conventions. The judgement that follows — funding, investment, hedging — is the job, and the assembly consumes the morning. This guide covers where AI helps, drawing on FISTA Solutions' AI agents work in finance operations. It complements ai guide for cfos and how to build a board reporting assistant. This article is general guidance, not financial or legal advice.

Why does position consolidation take so long?

Because banks report differently. Statement formats vary, as do delivery timing, value dating conventions, reference field usage, and how intraday activity is represented. Assembling a single accurate position across a dozen relationships in several currencies is reconciliation work, done daily, under deadline.

Automating it returns the morning to treasury and, more importantly, makes the position available earlier in the day when decisions can still be acted upon.

Activity	Automatable	Human required
Statement ingestion and normalisation	Yes	—
Position consolidation	Yes	Review of exceptions
Forecast variance decomposition	Yes	Interpretation
Bank fee comparison	Yes	Dispute decisions
Exposure monitoring	Yes	Action on breach
Payment initiation and authorisation	No	Segregated humans

Why explain variance rather than improve forecasts?

Because the forecast will be wrong and what treasury needs is to know why. Decomposing variance into timing effects, volume effects, rate effects, and one-offs tells them what actually changed and whether it will persist.

That decomposition is arithmetic over data treasury already holds, and it is considerably more actionable than a marginally better point forecast. It also builds the understanding that improves forecasting over time.

What does bank fee review recover?

Regularly more than the cost of doing it. Fee schedules are complex, tiered, and negotiated, and billing errors occur. Nobody checks every charge against the agreement because the volume makes it impractical by hand.

Systematic comparison of charges against the agreed schedule, month by month, finds discrepancies that are recoverable. It is unglamorous and it is one of the clearest returns available in the function.

How should exposure be monitored?

Continuously, against policy limits, with alerts before breach. Currency exposure, counterparty concentration, and investment limits all move with transaction flow, and a position checked weekly can move through a limit and back without anyone observing it.

Continuous monitoring with pre-breach alerting converts a control that is periodically verified into one that is actually enforced.

What must stay under human control?

Payment initiation and authorisation, absolutely, along with any change to banking details or beneficiary records. Segregation of duties and dual authorisation are core financial controls, and they exist because payment fraud is a live and well-organised threat.

Any system touching payments should make those controls stronger rather than more convenient, and automation that streamlines authorisation is moving in the wrong direction.

What about beneficiary change requests?

They deserve particular attention, because business email compromise targets exactly this. A system that assembles verification evidence — prior banking details, the request's provenance, callback verification status — helps the human control operate well. A system that processes the change is a vulnerability.

How does this affect reporting?

Substantially. Once positions and flows are structured consistently, treasury reporting to finance leadership and the board becomes a query rather than an assembly exercise, and it can be produced on demand rather than monthly.

That also makes variance explanation available at the moment someone asks, which is when it changes a decision. See how to build a board reporting assistant.

What about hedging decisions?

They remain human. Hedging reflects policy, risk appetite, market judgement, and accounting treatment, and a recommendation from a system that does not hold those considerations is not useful. What helps is accurate, timely exposure data, which is the input the decision has often lacked.

How is it evaluated?

Time from day start to position availability, forecast variance explained rather than unexplained, fees recovered, exposure limit breaches detected before they occur, and treasury hours returned to analysis. Reports generated is not a measure.

What goes wrong?

Normalising statements without handling value dating correctly, which produces confidently wrong positions. Variance reporting without decomposition, which restates the problem. Automating any part of payment authorisation. And treating exposure monitoring as a periodic report rather than a continuous control.

What does it cost to run?

Low relative to the recoveries, since the work is ingestion, comparison, and scheduled analysis. The investment is in bank format handling and the fee schedule reference, both of which are one-off with modest maintenance as relationships change.

What should you do first?

Time how long the position takes each morning and how late in the day it becomes reliable. Those two numbers are the case for the work, and the second one — when treasury can actually act on it — is usually the more persuasive.

How FISTA Solutions helps

FISTA Solutions builds treasury operations systems with multi-bank statement normalisation and early position availability, variance decomposition into real drivers, systematic fee comparison against agreements, and continuous exposure monitoring, while keeping payment initiation and authorisation under segregated human control, through AI agents, AI enablement, and forward deployed engineers. The record behind the approach is 150+ projects for 50+ companies with 47% efficiency gains.

To get your cash position earlier and your fees checked, message FISTA on WhatsApp, or read how to build a board reporting assistant.