AI Agents in Treasury: A 3-Year Roadmap to 70% Adoption by 2028

By 2028, roughly seven out of ten treasury departments are expected to run AI agents for cash forecasting, liquidity management and risk monitoring, according to a consensus of fintech analysts and leading treasury executives.

What Are AI Agents? A Beginner’s Primer for Treasury Teams

Key Takeaways

• AI agents differ from chatbots by acting autonomously and learning from data.
• Treasury processes are data-rich, making them ideal for AI-driven decision support.
• Early pilots show up to 40% reduction in manual reconciliation time.

An AI agent is a software entity that can perceive its environment, reason about the data it receives, and take actions without human prompting. Unlike generic chatbots that follow scripted dialogues, AI agents combine natural-language processing, predictive analytics and reinforcement learning to make independent decisions. In treasury, this means an agent can ingest real-time cash-position feeds, evaluate upcoming payment obligations, and suggest optimal funding strategies - all while continuously improving its models from historic patterns.

Cash forecasting, liquidity management and risk monitoring are uniquely suited to AI agents because they rely on massive, high-frequency data streams such as bank statements, ERP transactions and market rates. The repetitive, rule-based nature of these tasks creates a low-friction path for automation, while the strategic impact of better forecasts provides a strong business case. Moreover, treasury teams already operate under strict compliance frameworks, and AI agents can be built with compliance-by-design rules that flag anomalies before they become regulatory breaches.

Consider a scenario where an AI agent automatically reconciles invoices against bank statements. The agent matches payment dates, amounts and vendor identifiers, learns the typical lag between invoice receipt and payment, and flags outliers for review. In a parallel use case, the same agent runs a predictive cash-flow model that updates every hour, allowing treasurers to move surplus cash into short-term investments before market rates shift. Both examples cut manual effort dramatically, freeing analysts to focus on strategic initiatives rather than data entry.

Basware’s Breakthrough: How the First AI Agent Training Changes the Game

Basware has unveiled what it calls the industry-first AI agent training program, a modular framework that turns raw treasury data into a living, self-improving assistant. The program starts with a data ingestion layer that pulls transaction records, payment schedules and market feeds into a secure data lake. From there, scenario modeling engines generate thousands of what-if simulations, teaching the agent how to react to liquidity shocks, currency swings or sudden regulatory changes.

What sets Basware apart is the combination of modular skill sets and a compliance-focused architecture. Each skill - such as invoice matching, cash-flow prediction or hedge optimization - is packaged as a plug-in that can be activated or de-activated without rewriting code. The compliance layer embeds rules from IFRS, SOX and local treasury regulations, ensuring the agent’s recommendations are audit-ready. Integration is handled through open APIs that connect to ERP, treasury management systems and banking portals, allowing the agent to act within the existing tech stack rather than forcing a wholesale replacement.

For treasury analysts, the benefits are tangible. Decision cycles shrink from days to minutes because the agent surfaces the best funding option the moment a cash-outflow is recorded. Forecast accuracy improves by 15-20% in pilot programs, as the agent continuously refines its statistical models with new data. Compliance risk drops as the agent automatically flags transactions that breach policy thresholds, reducing the likelihood of costly fines. In short, Basware’s training transforms a static reporting function into a proactive, intelligence-driven hub.

Current Adoption Landscape: Where Treasury Teams Stand Today

"Only 22% of treasuries have moved beyond pilot projects to full-scale AI agent deployment, according to the 2023 Treasury Tech Survey."

Despite the hype, AI agents are still in the early-stage adoption curve for most finance groups. The 2023 Treasury Tech Survey, which polled 350 senior treasurers across North America and Europe, found that 48% of respondents are running limited pilots focused on automated liquidity reporting, while just 12% have integrated agents into daily cash-management workflows. The gap between pilot and production is largely driven by three interlocking barriers.

First, data quality remains a stubborn obstacle. Many treasury departments still rely on legacy ERP extracts that contain missing fields, inconsistent timestamps and duplicate records. Without clean, structured data, an AI agent cannot learn reliable patterns. Second, change-management fatigue hampers progress; finance teams are accustomed to strict control environments, and introducing an autonomous decision-maker triggers concerns around accountability. Finally, there is a shortage of specialized talent - data scientists who understand both treasury operations and AI ethics are rare, making it difficult for teams to build and maintain robust models.

Typical early-stage use cases reflect these limitations. Automated liquidity reporting agents pull daily bank balances, consolidate them with cash-forecast inputs, and generate a one-page dashboard. While useful, these agents rarely perform end-to-end functions such as dynamic funding allocation or real-time risk hedging. As a result, most treasuries are still experimenting, gathering proof points before committing larger budgets.

The 2024-2028 Adoption Forecast: Experts Say 70% of Treasuries Will Deploy AI Agents

The 70% adoption forecast is anchored in a blended methodology that combines trend extrapolation, market sizing and a Delphi-style consensus among 15 leading fintech analysts. Analysts tracked AI agent spend across 200 treasury departments from 2020 to 2023, noting a compound annual growth rate of 38%. They then projected that the total addressable market for AI-enabled treasury solutions will reach $4.2 billion by 2028, enough to support widespread deployment.

Breaking the timeline into phases helps illustrate the path to saturation. In 2024-2025, we expect a wave of pilot expansions as vendors like Basware, Kyriba and SAP release more plug-and-play modules. By 2026-2027, early adopters will move into full-scale roll-outs, integrating agents into cash-management, FX hedging and working-capital optimization. Finally, 2028 should see market saturation as legacy systems are retired and AI agents become the default decision-support layer for treasury.

Three key drivers accelerate this trajectory. Cost savings are the most compelling; early pilots report up to 30% reductions in labor hours and a 0.5% improvement in cash-conversion cycles, translating to millions in hidden revenue. Regulatory pressure also pushes firms toward real-time monitoring, a capability that AI agents excel at. Finally, the maturation of training platforms - exemplified by Basware’s modular program - lowers the technical barrier, allowing even mid-size treasuries to launch sophisticated agents without building in-house AI teams.

Opportunities & Risks: What the Future Means for Treasury Analysts and Strategists

As AI agents become mainstream, treasury professionals will need to acquire a new skill set. Data literacy moves to the front line; analysts must understand how to interpret model outputs, assess data quality and communicate insights to senior leadership. AI ethics training becomes essential to recognize bias in predictive models, especially when agents suggest funding strategies that could disadvantage certain business units. Scenario-planning expertise will also grow in importance, as agents will generate multiple “what-if” paths that humans must evaluate for strategic fit.

With autonomy comes new risk-management challenges. Model bias can creep in if training data reflects historic financing preferences that no longer align with corporate strategy. Over-reliance on agents may erode human judgment, making it harder to intervene when market conditions shift abruptly. Auditability is another concern; regulators will demand transparent logs of how an agent arrived at a recommendation, requiring robust version-control and explainability frameworks.

Despite the risks, the upside is substantial. Early adopters report a 20% boost in forecast accuracy and a 15% reduction in working-capital costs within the first twelve months. A multinational consumer goods company quantified an ROI of $4.5 million after automating invoice reconciliation and cash-flow prediction with an AI agent. Those figures illustrate that, when governed properly, AI agents can deliver both cost savings and strategic advantage.

Action Plan: How Your Treasury Team Can Get Ahead of the AI Wave

Getting ahead starts with a disciplined, step-by-step roadmap. First, assess data readiness by cataloguing all cash-related feeds, cleaning inconsistencies, and establishing a single source of truth. Second, select a vendor whose platform aligns with your compliance needs; Basware’s modular training program is a strong candidate for firms seeking a compliance-by-design approach. Third, pilot a small, high-impact use case - such as automated invoice reconciliation - to prove value quickly. Finally, scale up by adding complementary modules like cash-flow prediction and dynamic funding allocation, while instituting a governance board to oversee AI performance.

Training resources are abundant for beginners. Online courses from Coursera and edX cover data science fundamentals for finance, while Basware offers a free “AI Agent Fundamentals” module that walks users through data ingestion and scenario modeling. Industry workshops hosted by the Association for Financial Professionals (AFP) provide hands-on labs and peer networking opportunities.

Change-management success hinges on three best practices. Align stakeholders early by presenting clear business cases and risk-mitigation plans. Establish a governance framework that defines decision-rights, model-audit procedures and escalation paths. Finally, monitor AI performance continuously - track forecast error rates, compliance alerts and user adoption metrics - to ensure the agent remains a trusted partner rather than a black box.

What is the difference between an AI agent and a chatbot?

An AI agent acts autonomously, learns from data and can make decisions without human prompts, while a chatbot follows predefined scripts and requires user interaction for each task.

How can a treasury team prepare its data for AI agents?

Start by consolidating all cash-related feeds into a single data lake, standardize formats, eliminate duplicates, and establish data-quality metrics that can be monitored over time.

What are the biggest risks of deploying AI agents in treasury?

Key risks include model bias, over-reliance on automated decisions, and auditability challenges. Robust governance and explainable-AI practices are essential to mitigate these risks.

When is the right time to scale an AI agent pilot?

Scale after the pilot delivers measurable ROI - such as a 20% reduction in manual reconciliation time - and after governance structures are in place to monitor performance and compliance.

Which skill sets will treasury analysts need in an AI-driven environment?

Analysts should develop data literacy, understand AI ethics, and become comfortable with scenario planning tools that integrate AI-generated insights.