24/7 Treasury: From static thresholds to programmatic rules at Airwallex

Treasury never sleeps. Can AI keep pace with every account, currency, and time zone?

Treasury is one of the most operational, yet strategic, high-frequency parts of finance. Every day, my team manages 40+ currencies and hundreds of millions of dollars across the globe. Funds move between accounts, FX gets executed, alerts come in and need a response, and customer funds must remain safeguarded at all times, in every time zone, without any room for mistakes.

It is a highly rule-based function, which is exactly why it’s such fertile ground for AI. With AI, we start to question how many of these constant, high-volume movements really need a person in the middle, and how much we can confidently hand over to AI.

I'll admit my first instinct was caution. Treasury professionals are naturally conservative, and for good reason. When you're responsible for liquidity and customer funds, you don't experiment casually. What shifted my thinking was realising how much the answer depends on how the treasury function is structured underneath. So that's where we started.

What the baseline told us

As part of a baseline mapping exercise led by our CFO, Pranav Sood, across the whole of Finance, we mapped every task in Treasury: who does it, how often, how long it takes, and how automated it already is. It gave us a clear starting point for deciding where to focus next.

Two findings stood out.

75% of our mapped tasks occur daily, spanning alert handling, rebalancing, FX, safeguarding and interest income management. 17% of tasks mapped were classified as highly easy to automate, and the team had already made significant progress in systemising these. This exercise helped us distinguish between areas where automation was already well advanced and those where meaningful opportunities remained, giving us a clearer view of where to focus our next wave of automation investment.

Forecasting is the hardest win. It's the most complex process in Treasury, and it lags behind the rest of the function in automation maturity. Rebalancing and cash movement follow clear rules, so they're easier to crack. Forecasting depends on external, less-structured data, has more requirements, and may require human judgement. Trying to automate it end-to-end is much more challenging, but that makes it more fun too. We now identify the individual pieces within the forecasting process that AI can take on, so the process as a whole gets stronger.

That shaped our priority. We want to remove the administrative actions the team handles every day, so there's more capacity for forward-looking work like forecasting and liquidity optimisation, where human judgement and machine learning work best together.

Foundations first, then a 24/7 mindset

None of this works without a solid base. My team rebuilt our entire bank account landscape, examining which accounts exist, how they connect to one another, and the workflows for opening, closing, and maintaining them. That included the API connectivity with our banking and asset management partners. How you design your bank account workflows and your cash positions determines what you can build on top of them, and that’s the true enabler.

This strong foundation brings a level of accuracy that has moved us away from the traditional finance rhythm and into a 24/7 treasury mindset. The core of treasury is cash management, it’s knowing how much cash you have. Traditionally, that meant importing yesterday's bank statement to work out today's position. We wanted to go further and understand our liquidity at any moment in time, not just as of yesterday's close.

The first step was visibility. The team built dashboards and apps that gave us a continuous view of our cash positions, and those became the starting point for everything that followed.

Where AI is making the difference in Treasury 

With that foundation in place, we can structure things much more cleanly. Using AI tools, we can now pull data and run analysis from relatively simple prompts. Two areas have changed the most.

Cash rebalancing now runs on programmatic, signal-based rules 

Moving funds is the essence of a lot of what we do in Treasury, and it happens daily. Rebalancing has relied on static thresholds, along the lines of "if a balance exceeds US$100,000, move funds". That approach made sense for the tools we had. Now we can set programmatic rules to move cash based on a range of signals, which makes us far more precise. This is what I think of as programmatic treasury: rules that respond to what's actually happening, rather than to a fixed number someone set months ago.

Dedicated tools continuously monitor safeguarding 

As a Treasurer in one of the largest global fintechs, safeguarding is my biggest priority. It means customer funds are held safely and kept fully separate from our corporate funds under all circumstances. We're fully committed to this and always work to the highest level of diligence. 

Much of our recent work has gone into dashboards and apps that help us monitor safeguarding continuously and report on it. One team member, for example, built a fully live AI-powered liquidity insights machine that factors in safeguarding and buffer constraints, flags large movements and identifies any possible FX mismatches.

The AI skills we've built have allowed us to be much more sophisticated and go deeper into the details than we could before, and that's changing both how we think and how we execute across our platforms and data. The real value of using AI lies in better decisions and safer execution across liquidity, FX, payments, and controls.

Thinking AI-first

Even after building many dashboards and apps, I still catch myself wondering whether I'm limited by my own assumptions. It's easy to default to thinking that something can't be automated when, in fact, it can.

My advice to the team, and to myself, is simple. Whenever something feels repetitive or frustrating, the instinct should be to ask why we don't automate it or why the report doesn't generate itself, rather than treating it as someone else's job.

We also need to think beyond automating individual tasks as they exist today. The bigger opportunity comes from stepping back, looking at the whole ecosystem, and asking where things could be streamlined end-to-end. In Treasury, nothing sits on its own. A change in one account or currency touches everything else, which is why fixing the foundation paid off across the whole system.

Compared with a year ago, we're able to take on far more sophisticated work with the same team. Much of that comes from how our tools have evolved, and just as much from how our own mindset has shifted.

The next priorities

We see this as a continuous evolution, and we're focused on three areas.

Forecasting

Forecasting remains the one area where AI hasn't fully caught up, largely because it relies on more external, less structured data. We've already identified parts of the process we can automate to support forecasting as a whole, and that's where we'll focus.

Safeguarding 

We've built a strong system of controls and structure. The next step is moving from today's dashboards and apps towards triggered, end-to-end workflows, where more of the infrastructure is fully optimised and automated.

Reporting

Treasury produces a lot of management information, from counterparty risk to FX risk and risk management more broadly. We want that reporting to be more sophisticated and more automated, so we spend fewer hours on creation of reports and more on the decisions it informs.