Friday, 4 September 2026
Artificial intelligence is beginning to alter something more fundamental inside corporate finance than the speed at which invoices are processed or reports are prepared. As companies move from experimenting with individual AI tools towards deploying networks of specialised agents, the finance department itself may be heading towards a different operating model.
That was one of the central conclusions of the AI4 discussion “From Automation to Advantage: AI’s Impact on the Office of the CFO,” where finance and technology executives examined how AI is moving from personal productivity tools into forecasting, accounts payable, analysis, enterprise intelligence and ultimately workforce design.
The discussion included Mayank Sharma, Chief Financial Officer of EnergyX, Yoav Nové, Co-Founder and CEO of Reindeer, and Jonas Melton of RSM, whose work focuses on automation and technology transformation within accounting and finance.
One of the clearest changes described by the panel was not technological but behavioural. Sharma said members of his finance organisation had moved from initial reluctance about AI towards increasingly routine adoption. Some now treat AI almost as another member of the team.
That change in attitude matters because the next phase of AI adoption is likely to involve considerably more than employees occasionally using ChatGPT or an enterprise copilot. Companies are beginning to contemplate finance organisations in which dozens, or eventually hundreds, of agents carry out different functions alongside employees. Those agents could reconcile transactions, process invoices, prepare analysis, retrieve information, monitor exceptions or feed continuously changing financial forecasts.
That creates a management problem that barely existed a few years ago. Someone eventually has to supervise the digital workforce. The panel suggested that organisations may therefore need new functions responsible for selecting, training, monitoring and evaluating AI agents. Responsibility could sit with the CFO, CIO, transformation office or a newly created team, but companies deploying AI at scale will increasingly need someone accountable for its performance.
This would include measuring whether agents are producing sufficient value, whether their decisions remain accurate, how much they cost to operate and where human intervention remains necessary.
For finance executives, one of AI’s biggest effects may ultimately be the disappearance of the traditional reporting delay. Finance departments have historically spent considerable amounts of time assembling information before they can analyse it. A forecast might be updated monthly or quarterly because gathering the underlying numbers and producing commentary takes substantial effort.
AI changes that economics. Sharma described EnergyX moving towards a continuously updated forecasting process in which financial information can be examined at considerably greater detail, including down to individual invoices.
Instead of finance employees spending much of their time producing numbers, automated systems can perform more of the collection and preliminary analysis. Employees can then concentrate on understanding why something is happening and what management should do about it.
That moves finance closer to the point where business decisions are being made rather than explaining their financial consequences several weeks later. The potential advantage is therefore not simply reducing headcount. It is reducing the distance between an event happening inside a company and management understanding its financial implications.
The panel repeatedly cautioned against evaluating every AI project through immediate labour savings. Some finance processes involve relatively small teams. Eliminating several manual tasks may therefore produce only modest direct financial savings when compared with very large customer-service operations or software engineering teams.
The greater advantage could emerge once numerous processes become automated simultaneously. Accounts payable, accounts receivable, reconciliations, tax preparation, procurement information and forecasting can potentially feed a much broader financial intelligence system.
Instead of management receiving information weeks or months after transactions occur, AI agents could process new information continually and alert decision-makers to changes as they develop. That could make forecasting more responsive and potentially allow CFOs to identify deteriorating margins, unexpected spending, collection problems or project overruns earlier.
One of the more practical messages from the panel was that companies risk using AI where ordinary automation would work perfectly well. Melton described situations in which businesses considered introducing generative AI for processes that could be resolved through straightforward integration or robotic process automation.
That distinction could become increasingly important as boards pressure executives to demonstrate an AI strategy. Putting an AI model into a process does not automatically make that process better.
Companies first need to determine whether the underlying workflow makes sense, whether systems should be integrated, whether departments are unnecessarily performing the same task differently and whether conventional automation can solve the problem before adding an intelligent layer.
The most significant discussion concerned employment. Sharma gave an example from his own organisation where an accounts-payable employee left and the company decided not to replace the position because AI agents could handle much of the workload.
Humans remain responsible for supervision and final approval, but the example demonstrates how AI could gradually reduce demand for transaction-heavy finance positions.
That creates a potentially profound structural problem. Traditional accounting organisations resemble pyramids. Large numbers of junior employees perform reconciliations, payments, invoice processing and reporting work before gradually gaining enough experience to become controllers, finance directors and CFOs.
If AI removes much of that entry-level work, companies must develop a different way of training the senior finance professionals of the future.
A controller does not acquire judgement simply by reaching a certain age. Much of that judgement historically came from years spent understanding transactions, reconciliations, exceptions and financial controls.
The industry may therefore need to redesign junior roles rather than simply eliminate them. Future finance employees could spend less time manually processing transactions and considerably more time supervising automated systems, interpreting information, investigating exceptions and advising operating departments.
The result could be a substantially different organisational structure. Instead of large teams performing repetitive processing underneath progressively smaller layers of management, future finance departments could contain fewer people overseeing much larger amounts of automated activity.
Senior professionals may increasingly become orchestrators, deciding what agents should do, checking their performance and intervening where judgement or accountability is required.
At the same time, the boundary between finance professional and technologist may continue to weaken. The panel noted that modern AI tools already allow finance employees with limited traditional programming experience to build dashboards, automate workflows and create relatively sophisticated internal applications.
This may reduce dependence on software engineers for smaller finance projects. The valuable employee of the future could therefore combine accounting knowledge, business judgement and enough technical understanding to design and supervise AI-assisted processes.
Another notable disagreement with conventional transformation thinking concerned data. Companies have spent enormous sums creating data warehouses, standardising records and cleaning historic information in preparation for analytics and machine learning.
Nové argued that businesses should be cautious about turning data preparation into a multi-year prerequisite for AI adoption. Large language models can often be taught more like employees: given instructions, examples and current information, then improved as they encounter new situations.
That does not eliminate the need for reliable information. However, it suggests that companies may be able to introduce useful AI applications without first cleaning every historic record across the organisation.
Other panellists stressed that the difficulty often lies elsewhere. Financial information may be scattered between accounting platforms, procurement tools, payment systems, project-management applications, SharePoint sites and internal documents.
The problem then becomes connectivity rather than simply cleanliness.
Despite considerable enthusiasm for experimentation, the panel drew a sharp distinction between AI at the edge of finance and technology controlling the company’s official books.
CFOs appear much more willing to experiment with AI for analysis, commentary, forecasting and workflow automation than with the systems that record financial transactions.
Enterprise resource planning platforms such as SAP and NetSuite remain deeply embedded because their behaviour, controls, audit processes and limitations are well understood.
For a CFO, software generating a management insight can usually be switched off if it performs badly. Software posting transactions into the general ledger or moving company money carries an entirely different level of risk.
Sharma therefore advocated maintaining a stable financial core while experimenting more aggressively around its edges.
The distinction could become an important feature of enterprise AI architecture: established platforms remain the systems of record while flexible AI systems become the layer through which employees interact with those records and perform work.
The rapid creation of AI-focused financial software companies is presenting CFOs with another decision: wait for established software providers to introduce AI functionality or move faster with younger AI-native vendors.
The panel suggested the answer depends heavily on the risk attached to the particular process.
Young companies may innovate faster, but finance departments also have to consider whether those suppliers will exist several years later, whether their controls are mature and whether auditors and regulators understand their systems.
Melton said some AI-native accounting products remain narrower than established platforms, potentially requiring customers to combine several products to reproduce the functionality of a mature enterprise financial system.
For companies already operating major ERP platforms, AI alone may not provide sufficient justification for replacing them.
The opportunity may be greater among growing middle-market businesses that already need to upgrade their financial systems. Those companies can evaluate AI-native alternatives as part of a transition they would have needed to make anyway.
Despite different perspectives, the speakers ended with broadly similar advice for CFOs.
Melton recommended beginning with the desired business outcome rather than searching for AI use cases simply because other companies have adopted them. Nové argued that companies need someone specifically responsible for AI strategy because the technology and available applications are changing too quickly for ownership to remain informal.
Sharma encouraged finance leaders to begin with relatively small, repetitive processes where the impact can be understood and controlled.
But his final point captured the broader message of the discussion. Organisations should not simply automate work and record the resulting labour savings.
The strategic question is what companies do with the capacity AI releases.
If finance teams use it merely to operate with fewer people, AI will primarily become a cost-reduction programme. If they use it to provide faster forecasts, deeper analysis and greater involvement in commercial decisions, the change could be considerably larger.
The office of the CFO would move from documenting what happened inside a business towards helping determine what happens next.
Source: CIJ.World Research & Analysis Team
