Treasury has never lacked data. What it has historically lacked is the ability to turn that data into confident, forward-looking decisions at the scale and speed that modern organizations require. AI promises to close that gap by automating the complex, time-intensive calculations and surfacing insights that would otherwise take days or weeks to produce.
Realizing that promise, however, requires organizations to do the foundational work first. More often than not, this is the dividing line between interest in AI and actual adoption.
The AI Landscape: Generative AI vs AI Forecasting Models
Generative AI and AI forecasting models are both broadly referred to as “AI,” but the distinction is especially important in a treasury context.
AI forecasting models use machine learning (ML) — a branch of AI where systems learn patterns from historical data — to analyze historical transaction data, identify patterns, and generate predictions about cash flows, liquidity positions, and risk exposures. Many teams have already encountered ML capabilities embedded in their TMS or forecasting tools.
Where AI forecasting models interpret existing data, generative AI creates new content based on that data. Sitting on top of an ML analytical layer, generative AI makes output accessible and actionable for humans. It produces narrative summaries, flags anomalies in plain language, and surfaces liquidity risks in real time.
Both technologies have the potential to elevate treasury teams from operational administrators to strategic business partners. Both also depend entirely on well-governed underlying data and on the willingness of end users to trust and act on what the models produce.
The Current State of AI Maturity in Treasury
Treasury’s appetite for AI is growing steadily. According to PwC’s 2025 Global Treasury Survey, AI adoption in treasury is shifting from experimental to essential: 74% of respondents are either expanding or actively using AI, with a particular focus on machine learning (71%) and predictive analytics (64%).
Yet, the same survey reveals that only 26% of respondents consider their AI capabilities moderately or very mature, while 42% remain in an experimental or pilot phase.
EuroFinance’s AI in Treasury Deep Dive reinforces this pattern. While 46% of respondents are actively evaluating AI solutions, 47% report no plans for immediate adoption — even as they acknowledge AI as a strategic priority. Their hesitation reflects a lack of trust, driven by privacy and security concerns and persistent uncertainty about the reliability of AI-generated outputs.
Barriers to AI Adoption
The three most commonly cited barriers to AI adoption in treasury are data quality concerns, systems integration complexity, and a lack of expertise within treasury teams. Together, they explain why many teams remain stuck in pilot mode.
Data Quality
Data quality is a fundamental issue in AI adoption. Thirty-six percent of respondents to the EuroFinance study cited poor data quality as the biggest obstacle to trusting AI-generated cash forecasts. Because AI models are only as reliable as the data that feeds them, teams with incomplete or inconsistent data are more likely to automate existing errors than to correct them.
Integration Complexity
A typical treasury tech stack includes an ERP, TMS, payments hub, bank systems, and risk management tools, many of which don’t often communicate natively with one another. Data lives in silos and is shared through fragile connections that frequently break when source systems are updated. More than a third of PwC survey respondents rely on manual tools, meaning many are attempting to build AI capabilities on infrastructure that isn’t designed to support them.
Lack of Internal Expertise
As the treasurer’s role evolves from cash manager to strategic business partner, technical demands are growing faster than skills within the function. Many companies don’t have dedicated data science resources, and without AI literacy embedded in treasury teams, even well-designed systems are prone to underuse and misapplication. Unless AI capabilities are built into tools teams already use, adoption will remain inconsistent.
Bridging the Adoption Gap: Taking AI From Pilot to Production
Recognizing AI as a priority is not the same as being ready to adopt it. According to the EuroFinance study, only 22% of treasury teams consider themselves fully prepared for AI adoption. A separate survey of senior finance executives found that between 15% and 25% have scaled AI internally, with a quarter still in pilot mode.
The leap from proof of concept to production generally occurs when:
- The data is solid enough that outputs are consistently trustworthy
- The system is integrated into existing workflows rather than running as a parallel initiative
- The people using it understand when to accept outputs and when to challenge them
Treasury and finance leaders can close the gap from trial to adoption by investing in these key steps:
1. Assess Your Readiness
Before evaluating tools or vendors, take an honest inventory of where your organization currently stands:
- Data: How complete, clean, and accessible is your underlying data? Are there known gaps or inconsistencies in your ERP, TMS, or other source systems?
- Systems: How well does your tech stack communicate? Are integrations stable, or do feeds break regularly?
- People: Does your team have enough AI literacy to evaluate outputs critically, identify anomalies, and escalate concerns when the results appear incomplete?
- Processes: Are your existing workflows documented and consistent enough to support automation?
2. Build Systems You Trust
When beginning your AI journey, apply new capabilities to lower-stakes, high-frequency workflows first — for example, routine cash positioning or bank reconciliation. Establish confidence through repeated, verifiable performance before expanding to higher-stakes decisions. Prioritize tools that provide transparency into their reasoning, and build human-review checkpoints into every AI-driven workflow from the outset.
3. Invest in Ongoing Training
A single onboarding session isn’t sufficient to prepare treasury teams for the reality of working in tandem with AI models. Effective training covers:
- How the model works and what its limitations are
- When to act on outputs and when to override them
- How to recognize when a result warrants further scrutiny
Paired with ongoing training and open feedback loops, this approach provides a clear and sustainable pathway to team-wide proficiency.
4. Govern Your Data Continuously
High-quality data requires ongoing stewardship. Assign clear ownership to every data source feeding your AI systems, define and enforce quality standards, and review inputs on a regular cadence. A well-maintained data foundation gives your team the confidence to act decisively on model outputs and ensures that AI delivers compounding value as your organization scales.
Trust, Governance, and the Path to Scalable AI
The principle of garbage in, garbage out applies with full force in treasury. A model trained on inaccurate data will produce confident but incorrect forecasts. In cash and liquidity management, confident inaccuracy carries real financial consequences.
That is precisely why foundational investment matters as much as the sophistication of the algorithm. Strong data governance creates conditions under which AI can be trusted, and a robust control framework ensures AI outputs are validated before they drive decisions. Successful AI deployments are also underpinned by clear data residency and access control policies, ensuring that outputs are accurate and secure.
Treasury functions across the industry are already using AI across liquidity management, exposure management, and cash forecasting. It’s important to invest in the foundations as early as possible to capture the full benefit of these capabilities as the technology matures. It is equally important not to race towards the algorithm before addressing the data quality and skills gaps that undermine it.
Ready to see how TIS can support your treasury team’s AI readiness? Book a demo today.
