The road to autonomous finance starts with solid foundations
For years the phrase “autonomous finance” has circulated CFO forums, conferences and vendor decks. The vision is appealing: a finance function that closes the books in minutes, predicts cash positions months ahead, and flags strategic risks before an analyst has even opened a spreadsheet. The gap between that vision and what most Australian organisations can implement is worth deconstructing.
The usual analogy is the autonomous vehicle. Both promise a future where human error no longer threatens complex, high-stakes processes, but that comparison only stretches so far.
Finance is more like a train on a track than a car navigating unpredictable roads. The rules, controls, and workflows are knowable, and the structure already exists. The variables that need to be considered are largely internal and bounded.
That reframes what “autonomous” should mean here. Car automation chases driverless operation. It’s a monumental challenge to take the human out of the driving equation, but it is nonetheless the intent of bringing automation to cars. In finance, however, the goal should never be that and instead needs to be to automate everything that does not require CFO judgment, so that judgment can be applied where it counts.
The Maturity Gap
Most finance functions operate between Level 1 and Level 3 on a five-level scale. Level 3 means automated workflows, invoice matching, and reconciliations running without manual intervention. Level 5 means a system ingesting macro-economic signals, combining them with internal data, modelling forward scenarios, and producing board-ready insight with confidence.
This is borne out in practice. In a recent poll of finance leaders at SAP Connect for Finance, Thirty-seven placed their organisations across the lower tiers and nowhere else. Seventeen sat at Level 2, with standardised core processes and some automation in place. Eleven were at Level 1, still spreadsheet-driven and siloed. Nine had reached Level 3. Perhaps most tellingly, the higher intelligent and autonomous tiers sat empty.
Asked where they aimed to be within 18 months, most respondents selected Level 3 and Level 4 as their targets , each chosen by 16 respondents, with six reaching for full autonomy. The appetite is clear, but as the survey shows, the starting line sits further back than the destination implies.
Put bluntly, Level 5 is not happening at scale within a year or two. AI project failure rates remain high, with some estimates putting successful enterprise adoption at eight to ten per cent. The most common cause is trust. When outputs cannot be traced to clean, accurate, coherent data, finance leaders stop relying on them.
The Clean Core is Non-Negotiable
Progress requires what the SAP ecosystem calls a clean core: systems running standard, out-of-the-box processes, free of the custom modifications and workarounds most legacy ERP environments carry. This is as much a discipline question as a technology one.
An AI engine is only as good as the data it consumes. If that data has been exported, manipulated in Excel, and reloaded elsewhere, the outputs reflect the degradation. The engine produces confident-looking results regardless, with no way of knowing its inputs are compromised.
Finance is the end of the line. Every upstream data problem, from inventory to HR to CRM, eventually shows up here and gets crystallised into a number. Better algorithms cannot compensate for poorly maintained upstream systems.
The Excel Problem
A simple diagnostic: how much of your reporting runs through Excel? The honest answer, in most organisations, is more than leadership realises. Reports are generated in one system, exported, adjusted to add context or make corrections, and fed into another to produce the final output.
The people who do this are the glue holding finance together. They know which numbers need adjusting and which workaround has been in place since the last migration. That knowledge is valuable, and it is also a structural barrier to automation. When the logic connecting systems lives in someone’s head and runs through a spreadsheet, there is no clean core.
Connectivity is the Real Constraint
Genuine predictive insight requires more than financial data. Where people represent 80% to 85% of the cost base, HR data is fundamental to forecasting churn, skills gaps, and retention risk.
For manufacturers, supply chain connectivity is the constraint. Delivery on time and in full is a financial metric as much as an operational one. The same logic extends to CRM data, asset registers, and compliance systems, with finance sitting downstream of all of it.
Where the Easy Wins Are
Organisations should not wait for perfect conditions. The easy wins are real and they compound. Level 3 covers automated invoice processing, purchase order matching, autonomous reconciliations, and workflow approvals with defined tolerances. This is achievable now, without resolving every integration challenge first. These clerical tasks consume real capacity, and the engines that replace them run continuously and scale without adding headcount.
The effect is what practitioners call slowing down to go fast. A team that spent seven to twelve days on month-end close can perform a soft close in hours, meaning that the recovered time gets redirected, and leaders trapped in execution can begin operating as genuine business partners.
The Human Factor Remains
An autonomous system can model the commercial logic of closing a plant. Weighing community impact, social licence, and the long-term effect on employee trust requires human judgement. So does reading a macro environment where the variables are genuinely novel and no historical dataset applies. What it cannot do is carry the weight of judgement that leadership uniquely requires.
Audit current maturity honestly, identify the friction points automation can address today, and begin building the data discipline and connectivity that higher levels will eventually require.
That instinct is widely shared. Asked what support would help most right now, the same finance leaders most often pointed to peer benchmarking: understanding where they stand against industry. Defining the roadmap and selecting the right technology followed closely behind. You cannot plot a credible path to higher maturity without an honest read of your current position.
Starting today means taking the clean core seriously. The first step is to map where Excel is running the business, and identify which upstream systems need drawing into the finance data ecosystem before the engine can be trusted to model the organisation accurately.
It would surprise many organisations that believe themselves to be on the pathway to automation to find just how prevalent Excel still is in parts of the business. Success is not a question of whether autonomous finance is valid, because it clearly is. Rather, it comes down to understanding what “autonomous” means in this context, and the ones who get there will be the ones who started cleaning up the track.
