You Cannot Automate a
Process Built for Humans
The first task is not to remove people. It is to determine why they are there.
Finance automation underperforms when technology is applied to a workflow whose data defects, policies and decision rights remain implicit. The first task is not to remove people. It is to determine why they are there.
Many finance automation programmes achieve technical success without delivering the expected operating result. The workflow is digitised, tasks are automated, and productivity improves in selected activities. The finance cycles during the year remain dependent on the same people, the cycle time changes marginally, and the anticipated capacity is not released.
The common explanation is that the technology was not sufficiently capable. The accurate explanation is that the process was never designed to operate without human intervention. Automation was applied to the visible tasks while the decisions which govern the process remained undocumented and distributed across several employees.
The practical test is straightforward: how many people must review, adjust or approve the output before the organisation is prepared to rely on it?
In a mature process, each intervention should have a clear purpose. In many finance processes, it does not exist. The intervention exists because an employee knows where the data is late, which rule is not represented in the system, or which exception must be treated differently.
Task Automation Does Not
Change the Operating Model
Most automation business cases start with hours, hand offs and transaction volumes. These measures are useful, but they describe work rather than decision rights. They identify where effort is spent, not why the effort is required.
A pilot can therefore automate a high proportion of tasks and still leave the operating model unchanged. If a person must review the result before it is used, the organisation has retained the governing intervention. The reported automation rate and the realised business return then diverge.
McKinsey's 2025 research tested 25 organisational attributes against EBIT impact from AI. Workflow redesign had the largest effect of any attribute tested. Only 21% of surveyed organisations reported that they had fundamentally redesigned a workflow. Technology deployment without process redesign is therefore not a temporary implementation gap. It is a materially weaker value proposition.
Human Intervention Is
Not a Single Category
A process cannot be redesigned effectively while every human intervention is classified as judgement. The term conceals three different conditions, and each requires a different response.
An employee adjusts the output because the source cannot be relied upon. The correct response is to fix the source or the reconciliation, not to automate the workaround.
A rule is applied consistently but has not been formally defined. The response is to document, approve, and encode the rule.
A commercial, accounting, or risk decision requires accountable judgement. That decision should remain with a person, but it should be located at a defined control point with explicit authority.
This distinction changes the business case. A process with five interventions may not contain five discretionary decisions. It may contain two data workarounds, two unwritten rules and one genuine decision. The first four can be addressed through process, data and control design. The final decision should remain with an accountable person.
Agents Require Approved Rules
and Explicit Decision Rights
An agent can interpret information and select an action. It cannot supply accountability for a rule which was never approved or a decision which was never assigned. A system may infer a pattern from historical behaviour, but an inferred rule cannot be defended to an auditor or regulator unless the organisation has decided that the rule is valid and authorised its use.
The objective is not to remove every person from the process. It is to remove unnecessary intervention, formalise repeatable policy and retain genuine discretion at a defined decision point. Once that has been done, the technology requirement becomes clearer and the control position becomes defensible.
Redesign One Material
Process Before Scaling
The work should begin with one process rather than an enterprise-wide automation programme. Select a process with material cost or delay, and where the executive owner is prepared to change the process rather than only the supporting system. Record every point at which a person reviews, adjusts or approves the output, and the stated reason for the intervention.
Classify each intervention as compensation for a defect, application of an undocumented rule or genuine discretion. Assign the remedy to data, policy, control or decision rights, and identify the accountable owner. Then test the redesigned process against the real exceptions which currently require experience, not only against clean or standard cases.
The difficult cases are the relevant test.
A pilot on clean input demonstrates technical capability. It does not demonstrate that the process can operate under the conditions which created the manual intervention in the first place.
Measure the Operating Result,
Not the Automation Rate
The business case should be established on the basis of what the redesign actually removes. The relevant measures are capacity released, cycle time reduced, costs avoided, and risk controlled. An automation percentage is not sufficient if the governing interventions remain unchanged.
The relevant executive decision is not whether finance should adopt agents. It is whether the organisation is prepared to redesign a priority process, resolve the data and policy defects within it, and make genuine decision rights explicit. If the answer is no, further automation will improve selected tasks without changing the operating result. If the answer is yes, the organisation can evaluate technology against a process which has been designed to use it.
Start with one material process. Record every human intervention, classify it and establish the economics of removing the interventions which should not be there.
Decision Inc. is an advisory-led cloud & AI partner that works with over 300 organisations annually to build and scale enterprise data platforms across financial services, retail, telecoms, manufacturing and the public sector.
Book a Process Redesign Diagnostic
Identify which interventions in your finance process are data workarounds, undocumented policy, or genuine discretion, and what it would take to remove them.