The Finance Process Automation trap: Moving faster in the wrong direction

The trap, illustrated

Automation can turn a process that looks fine on paper into an unmanageable one at scale. A collections process with sound reminder logic generated more call actions than the team could handle once automated, and a missing €0.23 threshold led to formal reminders for amounts too small to justify the cost of sending them. A procure-to-pay case made the same point from a different angle: auto-posting rules ran on incomplete master data (missing cost centers) and copied-over approval thresholds that didn't distinguish between major and routine purchases, exposing gaps that manual workarounds had been quietly covering. In each case, automation didn't cause the problem; it revealed and scaled one that already existed.

Why it happens

None of this is due to bad intentions. It happens because people tend to rush straight into the solution before fully defining the improvement potential. Lander is direct about the pattern he sees the most: "It most of the time starts with: we need a solution, we should automate or use a tool that will fix everything." Vendors sell finance automation as plug-and-play, and under budget and leadership pressure to show results quickly, that pitch is appealing. The unglamorous work of mapping and improving the process first gets skipped.

AI raises the stakes here rather than lowering them. You don’t always need to purchase software. And on top, when you were self-developing with traditional RPA in finance, there were natural speed bumps: you needed IT architecture, a dedicated bot user, technical setup before anything ran. Those prerequisites forced a pause, and often a conversation. AI removes most of that friction. According to Lander, the barrier to "just try something" is now a prompt, not a development project, which is powerful, but also means "the risk of stepping into the trap is much higher," because fewer people are in the room to say let's think this through first.

Here is a second quieter driver: organizations frequently confuse efficiency with effectiveness. Efficiency is doing the process faster and cheaper. Effectiveness is whether the process delivers the right outcome in the first place. AI in finance is extraordinarily good at improving efficiency. But it cannot, on its own, tell you whether the process was worth running at full speed to begin with.

And there is a third trap: adding AI functionality to an existing process instead of redesigning that process with both AI capabilities and AI limitations in mind. To counter these limitations, it’s essential to embed the “human factor” at critical steps to validate AI output.

Building genuine AI readiness means checking process, data, and ownership together, rather than assuming the technology will absorb the gaps. It’s a mindset shift: AI works when the foundations are solid, not when it’s bolted onto whatever happens to exist today.

What "process-first" looks like

The fix isn't a multi-month transformation program. A rigorous first pass can happen in weeks, following the best practice sequence: “Understand, improve, and control the process, and only then automate it,” says Johan.

Understanding starts with mapping the process end-to-end: what's the actual input, what happens today, what's the real output, not the process management assumes is happening, but what's actually occurring on the floor. Lander has seen this gap firsthand at client sites, where "one team thinks it's performed like this, and then you talk to the business and they say no, never happened like this."

One of the most useful diagnostics here is what Olivier calls the first-time-right ratio, the percentage of cases that move through the process correctly on the first attempt, with no rework. It's often lower than organizations expect, and a low ratio is the clearest signal that a process isn't automation-ready. "If you have a lot of rework and you're going to automate the process," Olivier warns, "it means that you simply are going to automate the rework leading to an increase of rework.." The fix doesn't require a deep-dive study; a few days of close observation is usually enough to surface the quick wins.

"If you have a lot of rework and you're going to automate the process, it means that you simply are going to automate the rework as well." Olivier De Boeck, on the first-time-right ratio as the clearest sign a process isn't automation-ready

Improving means stripping out what shouldn't be there, adding what is missing and changing what is not working, before it gets encoded into a workflow. Think about unnecessary approval layers, reports nobody reads, changing handovers, reducing or adding steps that don't add value. Only once those inefficiencies are gone, the team has run the improved process manually for a while to confirm it holds up in practice, and KPIs are in place to monitor efficiency and effectiveness, does automation actually deliver the time savings it promises.

Automation is more than processes as well. Sometimes it is deeply embedded within the finance operating model of an organization. In companies that have grown organically, task ownership is often scattered. High‑touch processes and excessive handoffs become the norm: work passes from one person to another, sometimes with multiple people involved in creating a single invoice, and no clear process owner to turn to when something goes wrong. The fragmentation makes it harder to get the full picture. And because of that, there is a risk of wrongly automating the process.

Where AI changes the calculus

None of this is an argument against automation. Quite the opposite. Once a process has been understood, tightened, and proven, AI can apply to it faster and more flexibly than RPA ever could, and finance is fertile ground: it runs on repetitive, data‑heavy, largely predictable transactions. But AI is not the answer to everything. Finance runs on a large amount of structured data, and for structured data, rule‑based automation (RPA) is often the more reliable fit, faster, simpler, and without the hallucination risk that still exists with AI. AI shines when the data is unstructured: emails, documents, exceptions, conversations, anomalies. The real power comes from combining the two: RPA handling the structured, rule‑based steps, and AI managing the unstructured inputs and surfacing the issues that need human judgment.

The pattern holds across all of these: AI doesn’t replace the diagnostic work, it reveals why you need it. Use AI to surface the issues, map the real process, and highlight inconsistencies. But if you skip the process check, AI will still run. It will just run the wrong process faster, and with more confidence.

Change mindset as the missing ingredient and the big ‘Why’

Ultimately, this isn’t a people problem versus a process problem, it’s a people‑and‑process problem. “The process is as strong as the people who do it, review it, improve it, measure it,” says Lander. The traps appear because of how teams think about their work, not because the workflow itself is flawed.

That’s why mindset change isn’t an add‑on once automation goes live, it’s part of the improvement project from the start. When people understand the purpose, feel ownership, and trust the redesign, the process becomes stronger, and the automation that follows becomes far more resilient.

Many automation initiatives start with a technology question: What can we automate? Which tool should we use? Where can AI help?

We believe the better starting point is much more basic: Why does this process exist, what outcome should it deliver, and how well does it perform today?

That “big why” matters because management’s view of a process is often different from the reality on the floor. A process may look perfectly logical in a workshop, yet behave very differently when hundreds or thousands of transactions run through it.

"The process is as strong as the people that do it, review it, improve it, measure it." Lander Coene

Slow down before you speed up

Automation can create enormous value in finance. But the sequence matters: Understand. Improve. Control. Then automate.

That initial discipline does not have to take months. A focused assessment can often identify the most important issues within weeks. And time spent upfront is usually recovered later through fewer defects, less rework, smoother implementation and better adoption.

So before approving the next automation initiative, ask one question: Are we about to accelerate a good process, or simply make a bad one faster? Because automation rewards speed. But only after you have chosen the right direction.