For years, finance teams heard the same warning: do not automate a broken process. The phrase appeared in consulting reports, ERP playbooks, and countless transformation programs across finance and operations. It became a foundational belief among finance professionals.
The rule emerged during a time when enterprise software was expensive and difficult to change. Large ERP implementations often took years. Workflow engines inside these systems required detailed configuration and testing. Once deployed, altering the logic could trigger long development cycles and new rounds of validation.
Because of this, companies invested heavily in redesigning processes before any automation was introduced. Teams spent months mapping workflows. Consultants documented every approval step and exception case. Finance leaders reviewed process diagrams that stretched across entire conference room walls.
The goal was to make sure the process was clean before technology executed it. If automation ran a flawed process, it would simply execute the mistake faster and at greater scale.
For many years, this thinking was entirely rational. But the economics and capabilities of automation have changed so dramatically that the rule itself may now be outdated.
Why the Old Rule Made Sense
In the traditional world of finance automation, systems followed instructions exactly as they were written. Business Process Management platforms, ERP workflow engines, and Robotic Process Automation tools all operated in a similar way. They relied on fixed rules and carefully scripted steps.
This meant every workflow had to be carefully designed in advance. The goal was to remove ambiguity before the automation system went live. A simple example is invoice approvals. A finance team might decide that invoices under a certain threshold require one approval, while larger invoices require two or three. The rules are then embedded inside a workflow engine.
If those rules are poorly designed, automation does not correct the problem. Invoices are routed through inefficient approval chains faster than before. Bottlenecks appear earlier. Delays grow across the system.
In this scenario, redesigning the process before automation was not just good advice.
The Economics of Automation Have Changed
Artificial intelligence has now disrupted the economics of automation.
Traditionally, organizations treated automation as a large, carefully planned project. Experimentation was discouraged because each change carried real cost.
AI systems behave differently. Modern AI agents can be deployed quickly and updated continuously. Instead of relying entirely on predefined rules, they can analyze patterns in data and adjust their behavior over time. This shift creates a new contrast between old and new automation models.
Organizations no longer need perfect process designs before automation begins. They can start earlier and learn faster. AI becomes not only a tool for executing work, but also a tool for discovering how that work should evolve.
Automate First, Improve While Running
This shift suggests a different way to think about process improvement.
Organizations can begin by automating parts of the process while allowing the system to observe how work actually flows through the organization. As tasks move through the workflow, the AI collects signals that reveal inefficiencies.
For example, the system may detect repeated human overrides of automated suggestions. It may notice delays at certain approval stages. It may identify duplicate tasks across departments or repeated manual corrections to the same type of transaction.
In effect, AI can treat the process itself as a learning system. The system executes the work while simultaneously identifying where improvements should occur.
One important point that is often misunderstood is new technology adoption in finance will rarely happen as a “big bang.” For most mid-to-large enterprises, that kind of transformation is simply too risky. Finance operations cannot stop running while a new system replaces everything at once. Continuity of business always comes first.
This means organizations will not deploy one AI agent that suddenly takes over the entire month-end close or the entire accounts payable function. In reality, adoption will happen step by step.
The more practical way to start is by selecting specific pieces of a process. Finance leaders might begin with a few reconciliation categories, such as balance sheet reconciliations, intercompany transactions, prepaid accounts, or fixed assets. These areas are often complex, time-consuming, and prone to manual effort.
When AI begins assisting with these tasks, something interesting happens. Teams quickly see where time is actually being spent. They see which reconciliations require the most attention.
They see which approvals create delays. And they begin to understand where knowledge inside the process is concentrated in only one or two individuals.
Once those insights become visible, organizations can move to the next improvement. Another reconciliation can be automated. Another workflow can be assisted by AI. Over time, the finance operation becomes more intelligent without ever needing to pause the business.
In other words, AI adoption in finance is less like replacing an engine and more like upgrading parts while the machine is still running.
Why Many Organizations Hesitate
Despite these advantages, many organizations find this approach counterintuitive.
For decades, operational thinking has been shaped by Six Sigma methods and consulting frameworks that emphasize process perfection before execution. The idea of automating before fully redesigning the process feels risky. Many executives worry that automation might simply accelerate chaos.
On the other hand, AI systems allow rapid experimentation. In this environment, the greater risk may not be automating imperfect processes. The greater risk may be delaying experimentation and missing the opportunity to learn how the process truly works.
Instead of waiting for a perfect process redesign, organizations can begin with workflows where AI can observe real activity. The goal is not immediate efficiency. The goal is insight.
This creates a new operational layer above core systems. Some platforms describe this as a system of action, where AI continuously orchestrates tasks across finance systems, documents, and communication channels. Platforms like ChatFin are focusing on how AI can coordinate financial workflows rather than simply record transactions.
The New Rule of Automation
AI has changed the automation dynamics.
Experimentation is now inexpensive. Learning cycles are faster. Workflows can evolve continuously as new patterns emerge.
The new rule is beginning to replace the old one: Let AI run the process and show you what should change. The companies that succeed in the AI era will not necessarily be the ones with perfectly designed processes. They will be the ones willing to let intelligent systems observe, learn, and reveal the better process while the work is already happening.
Written by Ashok Manthena, an author & researcher on the Finance AI topic. He currently heads AI at ChatFin.ai
Ashok is the CFO & Chief of Finance AI at ChatFin, author and compelling speaker on AI in finance.
He is passionate about transforming how finance teams work by using the power of AI.
He works closely with finance leaders across functions like FP&A, Treasury, Tax, and Controllers, helping them streamline processes and unlock their full potential.
His mission is to make AI a game-changer for finance professionals, helping them become 10x more impactful in their roles by:
1. Helping CFOs bring AI into their finance processes.
2. Leading ChatFin, an advanced finance AI agent built for accounting and finance teams.
3. Staying hands-on by experimenting with the latest AI models and designing intuitive user experiences for AI.
4. Sharing his journey through articles, papers, and keynote talks.
