Building the baseline for AI transformation

The first step on our journey to transform the office of the CFO here at Airwallex was to define our baseline. Without understanding where time was being spent and how cross-functional processes (including month-end close) worked, we lacked a clear way to determine where to focus.

So we mapped all the distinct tasks across Finance & Corporate Development, accounting for hundreds of thousands of hours of work every year. Each department documented their processes, the frequency, time required, and current and target state. We supplemented all of this with a further 20 hours of interviews across the team to provide further qualitative insights from those closest to the work.

The end result is a database of nearly 600 tasks, of which c. 80% are recurring.

Controllership

Controllership accounts for the largest share of Finance hours and was the natural place to start for a deep dive. Our team is split between a centralised Shared Services Centre (SSC) and Regional teams. The SSC centralises finance processes like Procurement, AP and T&E. Much of this already runs through Airwallex, benefiting from our platform’s combination of infrastructure and agentic workflows.

The work of the regional teams is more ad hoc in nature. Regional controllers build relationships with local teams, regulators, and third parties. They are our local experts on regional matters and support the Group with matters relating to licensing and expansion. Today, around 40% of their work is ad-hoc or annual, and we would like that share to increase so they spend more of their time on judgement and local expertise. 

One factor that surfaced during interviews but was not obvious in the mapping exercise is the importance of throughput at month-end. The Controller role follows a monthly arc, and at month-end, the entire team is on high alert and operating at max capacity. The goal is to close the books and report numbers as quickly as possible after the month closes. This follows a strict process with multiple dependencies across teams, so any bottleneck here cascades and jams the queue. So throughput, not annual hours alone, is shaping our automation roadmap here.

Tax

Our tax team manages Airwallex’s compliance obligations across 25+ countries. While there is a lot of heterogeneity in the requirements of different tax authorities, we uncovered a surprising amount of commonality. However, we also found that the way financial data was gathered from Controllers and reconciled across regions varied, with different teams using different sources and processes. The team is now working to standardise these processes to create a consistent foundation for automation.

The team is building a centralised hub to automate the completion of different filing document types. By bringing more of the filing process in-house rather than outsourcing it, we expect to save approximately $1 million a year.

Treasury

Treasury is the lifeblood of our business. We found that 75% of the team’s tasks happen daily, including interest income forecasting, alerts, rebalancing, and FX. Many of these tasks are repetitive and relatively low complexity and the team has already made significant progress in automating them. The aim is to create more capacity for activities like forecasting and liquidity optimisation, where we need both human judgement and machine learning.

FP&A

FP&A is a team whose work involves a high level of ambiguity, multi-dimensional analysis and significant use of judgement. The consequence of that is that most of the tasks the team completes were classified as having high automation complexity.

Budgeting and forecasting are good examples. Much of the manual effort goes into validating budget submissions across the organisation, and reconciling different data formats. AI is helping automate that reconciliation work, bringing different inputs into a consistent format and unifying them in a single database. The team can then iterate on budgets and forecasts much faster, spending less time preparing and validating data and more time on the analysis and judgement that shape decisions.

Developing the roadmap

With the insights from our mapping exercise, we have developed a prioritised AI roadmap for each team within the Finance function. Around 100 projects sit within them, with many automation solutions being built by citizen developers within the teams. We can now prioritise the remainder against a common view of value, feasibility, and capacity.

We are undertaking an ambitious transformation programme, having established the baseline, identified the drivers of manual work, and built the roadmaps to act on them. I expect every member of my leadership team to drive the transformation of their own department. This means OKRs tied to the efficiencies we create and a regular view of progress. We will continue to share what we learn as we turn that plan into a new operating model for Finance.