If we have selected the wrong experience for you, please change it above.
Authors:
Ask a CFO which function is measured most rigorously in their organization, and the answer is their own. Ask which is most automated, and the answer is often somebody else’s.
This gap between ambition and adoption remains striking. Deloitte’s most recent CFO Signals research found that 87% of CFOs expect AI to be extremely or very important to their finance department’s operations this year. Half rank finance digital transformation as a top priority, and more than half plan to embed AI agents directly into finance workflows.1 Yet across independent surveys, finance and accounting consistently ranks at or near the bottom of enterprise functions for AI in production.2 A large majority of teams remain in limited pilots.
This is often interpreted as conservatism. A more useful explanation is that finance automation must meet a higher standard of evidence before it can scale.
Marketing can run an experiment and discard it. Finance has far less room to do so. Every output the function produces—a balance, an accrual, a disclosure or a VAT return—will ultimately need to be defended to an auditor, regulator, or board. An automation that works 95% of the time and cannot explain the other 5% is not a productivity gain. It is an audit finding waiting to happen.
The barrier to finance automation isn't the technology. It's that every decision has to be auditable. Get the architecture right from the start, so you can scale it!
— Bettina Werner
This is not an argument for waiting. It is an argument for sequencing. Finance processes automate well when inputs are clear, rules are testable and review points are defined. These are also the characteristics of a well-controlled process. Done properly, automation strengthens control by standardizing treatment, time-stamping evidence, and removing undocumented spreadsheet steps.
Most finance functions have already invested in core technology. The ERP is in place, the consolidation tool is live, the planning platform has been upgraded. Yet manual effort has not disappeared; it has migrated into the connective tissue between those systems.
That is where the measurable value is. In accounts payable, benchmark studies put best-in-class processing at roughly €2–3 per invoice against €10–15 for everyone else, with cycle times of around three days versus more than two weeks. The differentiator is the touchless rate the share of invoices that travel from receipt to payment with no human keystroke. Average performers sit near a third. 3
What drives the gap is simple, proven mechanics: structured data capture, automated three‑way matching, and clear rules so only real exceptions go to a person.
The same pattern shows up everywhere: period close, intercompany reconciliation, cash application, and regulatory reporting. The work is not conceptually hard; it is high-volume, deadline-bound, and only visible when something goes wrong.
The technology conversation has moved beyond automating individual tasks. Rules-based tools handled repetitive keystrokes; document-processing tools reduced rekeying; Generative AI began drafting narratives. Agentic systems go a step further by coordinating a sequence of tasks, identifying cases outside defined tolerances, and escalating them with supporting evidence.
The most practical architecture is usually an additional intelligence layer around the accounting core, rather than wholesale replacement. It can read across operational and finance systems, detect variances, trace likely causes and propose treatments subject to appropriate access controls, logging, and human approval.
The winning design is not one clever agent. It is a small set of narrow, well-scoped agents, each with a defined job, a defined boundary, and a human who owns its output.
— Cédric Jadoul
Procure-to-pay: This is the clearest case with a fixed external deadline. From July 2030, mandatory e‑invoicing for intra‑EU B2B transactions (under the VAT in the Digital Age package, EN 16931 standard) will apply.4 Recent surveys suggest that while almost all organizations are aware of this, only about 20% have a concrete transition plan.5 Structured invoice data is not merely a compliance artifact; it is the foundation that makes all downstream processes automatable.
The financial close: Reconciliation preparation, flux analysis, and first-draft variance commentary can all be automated to a credible standard. The review does not move; the preparation does. Teams that do this well typically shorten the close by days, not hours, and more importantly, use the time saved for analysis rather than assembly.
Regulatory and sustainability reporting: This requires finance teams to collect data across entities, map it to reporting taxonomies, maintain supporting evidence and follow up on gaps. Much of this coordination can be automated.
Most failures fall into four patterns:
Automation is not a project you finish. It is a service you operate—and almost nobody budgets for the second part.
— Cédric Jadoul
Luxembourg’s finance functions carry a distinctive load: many entities, multiple GAAPs, cross-border reporting, and dense regulation, all handled by teams that are, by international comparison, small. That combination changes the argument entirely. Here, automation is rarely about reducing headcount. It is about creating capacity in teams that are already at full stretch, and coping with a reporting calendar that keeps tightening without more people.
It also changes work sequencing. In a five-person finance team, the constraint is not ambition; it is bandwidth to transform alongside the day job. That points to a focused approach: narrow scope, fast payback, and a first use case that produces reusable infrastructure rather than a one-off solution.
Finance is not going to stop closing the books, but the shape of the work is changing: less assembly, more judgment; less month-end scramble, more continuous visibility. Progress does not always require a platform decision or a five-year roadmap. It starts with one process where inputs are structured, rules are testable, and the evidence trail is defensible. Build that properly so the second process is cheaper and faster than the first.
The regulatory clock is already ticking. Teams that start with auditable, well‑controlled processes will discover, when the deadlines hit, that they have been getting ready the whole time.
The finance functions that win will not be the ones that automate the most, but the ones that can prove what they automated.
