Finance has long been the function that explains what happened last quarter. Axel Demazy, the chief executive of spend management platform Spendesk, wants to change that — moving the CFO's office from a recorder of history into a real-time engine for decision-making.

In a wide-ranging conversation, Demazy lays out what it actually takes to build an AI-native finance function, why most companies' AI experiments stall before they reach production, and what the CFO's role looks like when agents handle the execution and humans own the governance.

What AI-native finance really means

For Demazy, the phrase "AI-native" is doing serious work — and it's frequently misused. Dropping a chatbot onto an existing workflow or generating reports through a copilot isn't the transformation; it's a surface-level upgrade that leaves the underlying structure untouched. Genuine AI-native finance means the function is rebuilt around continuous intelligence from the start.

In practice, that looks like autonomous agents that monitor spend as it happens, surface risks before they become problems, apply policy rules automatically, and propose — or in some cases initiate — corrective actions within clearly defined boundaries. The reporting cycle stops being a periodic ritual and becomes a live, always-on feed. Spendesk's own AI Connect product, launched at a major industry event this year, goes further still: it lets finance teams query their spend data in plain language through AI assistants, pulling answers in seconds rather than building reports from scratch.

Letting the whole company build

When Spendesk rolled out its internal AI programme in 2025, the decision was made to give every employee access from day one — no phased rollout, no top-down mandate dictating which teams got to experiment first. The logic was simple: the people closest to the work know where the friction lives.

The results confirmed the instinct. Over 200 agents were created across the business, and the majority graduated from experiment to production. The impact showed up in concrete places: the Customer Success team now completes quarterly business reviews in around three minutes, a task that previously took the better part of two hours. But Demazy suggests the more consequential change was cultural. A quarter of all employees have built their own agent. Teams now habitually ask whether an agent could handle a task before defaulting to a manual process. That shift in reflex, he argues, matters more in the long run than any individual use case.

The cultural barrier is bigger than the technical one

When companies struggle to adopt AI at scale, the obstacle is rarely the technology. More often it's a mindset problem: finance teams built around manual oversight find it deeply uncomfortable to cede control to an intelligent system, even when the system is faster and more consistent.

Demazy frames the necessary shift as moving from control through personal review to control through system design — and acknowledges it requires genuine trust, built up through transparency, clear guardrails, and the chance to experiment in low-stakes settings. He also points to an emerging job category he calls the Finance Engineer: someone who layers automation, workflow design and AI tooling onto traditional finance expertise. Analysts become orchestrators of AI pipelines; controllers become architects of policy; FP&A professionals become scenario strategists rather than variance reporters.

Organisations are experimenting with copilots and small automations without rethinking the underlying system. AI-native means rebuilding end-to-end, not adding on. — Axel Demazy, CEO, Spendesk

Why most AI projects don't survive the proof-of-concept

Recent industry research Demazy cites finds that only a small minority of agentic AI initiatives make it past the experimental stage into genuine operational change. He sees a clear pattern: businesses run a pilot — a chat interface, an automation for one narrow task — without questioning the underlying process the tool sits on. The result is a productivity bump, not a structural improvement, and the ROI never materialises at the scale originally hoped for.

The second gap is governance. Many organisations lack any coherent framework for AI accountability: who owns a decision an agent made, how is a bad outcome escalated, what does good look like and how do you measure it? Without that infrastructure, AI stays a local efficiency tool rather than becoming a strategic asset the CFO can depend on.

Rethinking control in an agent-driven model

The CFO's traditional lever of control — reviewing transactions, approving spend, signing off on reports — looks very different when agents are handling the execution. Demazy's view is that control doesn't disappear; it moves upstream. Instead of scrutinising individual actions, CFOs become the designers of the systems that govern autonomous agents: setting the rules, defining the escalation paths, maintaining the audit trails that make the whole thing accountable.

Risk management, in this model, becomes continuous by default. Agents flag anomalies the moment they appear rather than waiting for the monthly close. Human error drops — but only if the humans in charge have invested in the transparency mechanisms that let them see clearly what the agents are doing and why. The CFO's role, in Demazy's framing, evolves from functional manager to architect and steward of an intelligent financial ecosystem.

SP&A: the function built for what comes next

Demazy introduces a distinction between traditional Financial Planning & Analysis and what he calls Strategic Planning & Analysis — SP&A — which he sees as the natural successor in an AI-native environment. Where FP&A spends its energy on budgeting cycles, variance reports and historical reconciliation, SP&A operates in real time: running simulations, modelling strategic scenarios, and translating data into decisions that the wider business can act on quickly.

When AI handles data consolidation and routine forecasting, the human team is freed to focus on the harder questions — evaluating trade-offs, stress-testing assumptions, advising leadership on where to move. SP&A becomes what Demazy calls the strategic brain of the CFO's office: the function that bridges operations, finance and executive leadership and keeps the organisation moving with both speed and clarity.

Where to start if you want to get there

For finance leaders looking to begin the journey, Demazy's advice is grounded and sequential. Start by mapping workflows end to end, with honest attention to where decisions are actually being made — because that is where AI delivers its sharpest value, and where bolt-on tools consistently underperform. From there, pick a small number of high-impact, low-risk agents to build first: anomaly detection, policy enforcement, spend categorisation. Early wins matter because they build the organisational trust that larger changes require.

AI literacy, he argues, comes from doing rather than from training programmes. Build the first agent, let it run, learn from what breaks, and iterate. In parallel, establish the data foundations and cross-functional governance group that will keep things accountable as the system scales. Above all, he says, leaders have to own the mindset shift themselves — because no amount of tooling compensates for an organisation that hasn't yet decided it believes in the direction.