AI AND THE FUTURE OF TREASURY
The human element in scaling treasury management with AI
Realizing AI's potential starts with the people implementing it.
Key Takeaways
AI adoption in Treasury is scaling faster than most organizations can keep up with. Here’s what leaders need to understand before they jump in.
- Successful AI adoption starts with reimagining a workflow, not automating the one you already have.
- Mapping a process step by step, before applying AI, exposes legacy inefficiencies that might otherwise stay hidden.
- The strongest early use cases share four traits: high transaction volume, clear rules, reliable data and measurable outcomes.
- As AI takes on more tasks, critical thinking, judgment and adaptability matter more, not less.
- Governance and accountability don’t disappear with automation. Guardrails have to be built into the workflow, and the final call still rests with people.
This biggest change AI is prompting is how we think about the work itself
The teams successfully scaling AI reimagine workflows before applying the technology.
As AI moves inside treasury management, it's reshaping how leaders forecast cash, manage risk, surface insights and make decisions. So far, much of the conversation has focused on “efficiency”: reducing manual work, accelerating processes, cutting time spent on routine tasks.
But the bigger shift is just getting started, and will be much more transformative: entirely reimagined workflows and new approaches that impact the entire organizational ecosystem. To win in this period of massive change, leaders need to harness, of all things, the power of human judgment.
Recent data on AI adoption highlights this opportunity. Nearly 90 percent of organizations use AI in some capacity, but only a third have successfully scaled it.[1] In our experience the teams scaling effectively, before reaching for a tool, did the upfront, big-picture work first: understanding what a process actually does, why it works the way it does, and whether it should be automated at all.
That upfront work is where human judgment has become essential.
Legacy workflows can mask operational baggage
Examining processes thoroughly before applying AI can expose costly inefficiencies.
Many treasury functions run on years of accumulated processes, controls and institutional knowledge. Each adjustment probably made sense at the time, but the result is often a collection of legacy bolt-ons that quietly slow everything down. AI has a way of exposing that.
“You have to completely reimagine the work before you apply the technology,” said Heather Manuel, SVP, Strategy Director, Treasury Management at PNC. “In this way, the value of the technology can be unlocked before it ever arrives.”
It’s advice she gives clients, and it’s the question she returns to with her own team: Is this a process we built for yesterday, or one we’re building for tomorrow? Could a 10-step process be three?
That kind of rethinking is harder than it sounds. Most people, asked to walk through their own job step by step, leave things out — not deliberately, but because the nuance of how work actually gets done is rarely written down.
Diagramming a process thoroughly, before automating it, typically reveals steps no one remembers otherwise.
3 strategic questions for AI treasury workflow audits
Clear and measurable use cases build momentum for AI deployment
Small, well-chosen processes with reliable data are good starting points.
Not every scenario is equally suited for AI, particularly in the early stages of adoption. It’s important for teams to have ways of clearly measuring the value of the technology.
“Treasury leaders need to pre-determine which challenges are appropriate for AI, and where automation can create meaningful value,” says Manuel. “Determine how success should be measured before pushing the button."
Criteria for Initial AI Use Cases | Strategic Treasury Impact |
High transaction volumes | Higher volume gives AI more data to learn from, sharpening forecasts and outlier detection and scaling cost and efficiency benefits. |
Clearly defined rules and workflows | Rules-based processes are the safest to automate with confidence, since AI output can be tested against known standards. |
Reliable, accessible data | Clean, integrated data keeps AI from amplifying errors or making erroneous connections. |
Outcomes that can be measured and validated | Clear, measurable results help build a case for organizational buy-in and justify scaling AI into higher-stakes activities. |
Critical thinking and adaptability are becoming even more valuable
The ability to question, assess and learn is as important as technical expertise.
Earlier waves of technology adoption (such as traditional machine learning and the broader digital transformation) required a specialized technical team before anyone saw value. Generative AI works differently, and more democratically. Teams can start testing use cases and generating value with the skills they already have, before any technical build-out begins.
This increases the need for sharp human judgment. While technical expertise still matters, organizations can find greater value in employees who can critically evaluate AI-generated outputs through a generalist’s business lens.
Treasury teams should prioritize employees with capabilities, such as:
- Critical thinking
- Analytical reasoning
- Clear communication
- Business judgement
- Curiosity and open-mindedness
- The ability to evaluate and question AI-generated outputs
- High accountability
“The same growth-minded people you hired before are the same ones you need today,” says Michael Wolfe, SVP, Head of Emerging Products, Treasury Management. “People who are unafraid to do new things, to change processes — that’s the skill set.”
Sometimes that adaptability starts small. One treasury leader described coaching a colleague who felt completely disconnected from AI until she used an AI tool to help draft a simple thank-you note. The result sounded like her, only sharper. This low-stakes moment did more to build trust than a formal training session could have.
Governance is now part of the workflow
Guardrails and other controls should be built directly into how AI is used.
AI already touches many treasury decisions, from flagging anomalies to shaping cash positions. As its role expands and evolves, the way those decisions are governed needs to evolve as well.
Many treasury groups are still applying traditional oversight models to a fundamentally new set of tools. In an AI-enabled environment, governance has to live inside the system, not sit around it.
“Elements of governance can be baked into the use of the tool," says Wolfe. “Guardrails are a form of governance. Agent definitions, skills and workflows are all a form of governance.”
Traditional governance frameworks are still required, like clear decision rights, escalation procedures and accountability structures. The most effective models can increasingly combine such traditional human controls with AI-specific mechanisms designed to build trust and confidence over time.
Treasury Governance: How It’s Changing with AI
Decision rights and approval chains | Same decision rights, but also written explicitly into the AI agent's configuration |
Escalation procedures for exceptions | Exceptions are flagged automatically by the AI agent, but still resolved by human judgment |
Accountability for outcomes | Unchanged — still held by treasury leaders, not the model |
Audits and reviews | No longer limited to periodic; can be continuous by building them into agent workflows |
More automation doesn’t mean less human accountability
AI might handle certain tasks, but strategy and outcomes remain with treasury leaders.
Even though the role of AI continues to permeate treasury workflows, leaders still remain responsible for the outcomes of the decisions they make, whether informed by AI or not.
Clear expectations need to be established around how AI-generated recommendations are evaluated, when human review is required, where decision-making authority resides and how exceptions are escalated and addressed.
As organizations gain more confidence in AI, they might become more comfortable allowing technology to handle certain tactical decisions within clearly defined parameters. But strategy, accountability and responsibility for outcomes remain firmly in human hands.
Trust is one of the biggest barriers to adoption. Organizations can scale AI once leaders, teams and users all trust the outputs are reliable and the guardrails work as intended.
That human trust is built through human use - by watching AI work, testing its outputs and learning where it adds value and where it doesn't. “Everybody has to move along the trust journey,” says Wolfe.
