The CFO's Office in 2030: What AI Changes in a Nepali Finance Team, and What It Does Not

of finance functions surveyed by Gartner in 2025 use AI, against 37% in 2023

59%

use it for knowledge management, the most common finance use

49%

of employees using AI rely on its output without checking, per KPMG's 47-country study

66%

Walk into the accounts department of a typical mid-sized company in Kathmandu, Birgunj or Pokhara and the scene has barely changed in fifteen years. A row of desks. Files of purchase bills waiting to be entered. Someone matching a bank statement to the cash book line by line. Someone preparing the VAT return near the 25th. An accounts head who spends the first three weeks of every month finishing the previous month and the last week answering the auditor, the bank or the owner. The software is newer. The work is the same.

That scene will not survive the decade. The reason is not that a robot will sit at those desks. It is that the three activities that fill them, entering, matching and compiling, are precisely the activities at which current AI is already competent. What follows is a practical view of how the finance function of a Nepali company is likely to change between now and 2030, where the limits are, and how an owner or finance head should prepare.

This is the sixth article in our series on artificial intelligence for business in Nepal, and the first of four on finance: the others cover the close, forecasting and controls over AI-generated numbers.

Where finance functions stand today

Gartner's survey of finance leaders, published in November 2025, found 59 per cent of finance functions using AI, almost unchanged from 58 per cent in 2024 after a jump from 37 per cent in 2023. The most common uses were knowledge management (49 per cent), accounts payable automation (37 per cent) and error and anomaly detection (34 per cent). KPMG's global study of more than a thousand finance leaders, as reported, found about seven in ten saying AI was meeting or exceeding their return expectations.

Two things stand out. Adoption has plateaued at a little over half, which suggests the easy uses have been taken and the next ones are harder. And the common uses are modest: finding information, processing invoices, spotting errors. Nobody is running a finance function without people.

There is no equivalent survey for Nepal. The observable position is that banks and financial institutions, larger corporate groups and export-oriented IT firms are using AI features in their systems, while the typical mid-sized company uses AI informally: a junior with a chat tool, an accounting package that reads a bill from a photograph. The formal finance process is untouched. That leaves most of the change still ahead.

A finance function is a set of tasks

"Will AI replace accountants?" is the wrong question because an accountant is not one thing. The useful unit is the task. A finance function performs perhaps forty recurring tasks. Each can be placed on a scale from fully automatable to irreducibly human.

The pattern is consistent. The further a task is from a rule applied to a document, and the closer to a judgement someone must answer for, the less it changes.

What changes

Transaction processing becomes exception handling

Today a purchase bill is received on paper or as a photograph, typed into the system, coded, checked against an order if there is one, approved and paid. By 2030 the normal bill will be read, coded by rule and history, matched to the order and receipt, and queued for payment without anyone touching it. People will see only the bills that fail: no order, a price variance, an unknown vendor, a PAN that does not validate.

Nepal has a specific accelerant. Large taxpayers and prescribed sectors already issue invoices through approved software connected to the Central Billing Monitoring System, and the threshold is expected to come down in phases (e-billing and CBMS requirements). As structured electronic invoices replace paper, the reading problem disappears and the buyer's system can ingest the seller's invoice directly. The effort shifts from capture to control.

The close shortens and then dissolves

When entry and matching happen daily, the month-end is no longer a backlog to be cleared. A close on day 5 becomes normal, and for many balances a daily position is available. We treat this in detail in redesigning the close.

Reporting writes its own first draft

The management pack, with its tables, charts and a first draft of the variance commentary, can be produced from the ledger the day the books close. The finance head's job moves from assembling the pack to interrogating it: is this variance real, what caused it, what should the owner do? A draft that says "sales fell 8 per cent against budget, driven by lower volume in the eastern region" is a start. Knowing that the eastern distributor is in dispute over a credit note is finance work.

Forecasting becomes continuous

A cash forecast that updates weekly from actual receivables, payables, loan schedules and payroll replaces the annual budget as the working document. Models are better than people at the routine part, and worse at the breaks (forecasting under uncertainty).

Compliance preparation is largely automated, and more exposed

VAT returns, TDS statements and annual return schedules are rule-based compilations. Software will prepare them. At the same time the Inland Revenue Department is building its own analytical capability: the 2083/84 budget announced an AI-driven e-assessment system for risk-based investigation and audit, and the department's strategy describes automated case selection and data matching. Compliance becomes less about preparing the return and more about being sure the return agrees with everything else the state can see (the data-driven tax authority).

Questions get answered from the ledger

An owner will be able to ask, in plain Nepali or English, "which ten customers owe us most beyond 90 days and what did we sell them last quarter?" and get an answer in seconds. This is Gartner's "knowledge management" use. Its value depends entirely on the ledger being right.

Those who examine the company change too

Auditors are adopting tools that test entire populations of transactions instead of samples. Lenders are moving toward data-driven credit assessment under Nepal Rastra Bank's guidelines. Expect requests for complete ledgers and transaction files in electronic form, and expect anomalies to be found that a sample would have missed (continuous assurance, how lenders will read your business).

What does not change

It is easy to overstate the case. Five things stay.

Accountability. Financial statements are approved by directors. Returns are filed under a person's responsibility. An audit opinion is signed by a member of the Institute of Chartered Accountants of Nepal. No tool carries liability, and no regulator accepts "the system produced it" as an answer. The more work a machine does, the more the signature matters, because the signer is vouching for a process they did not perform by hand.

Judgement on uncertain matters. How much of this receivable will be collected? Is this expenditure capital or revenue? Is this provision adequate? Does this arrangement have substance? Nepal Financial Reporting Standards are full of estimates that depend on facts about the particular business. A model can inform the estimate. A person makes it and defends it.

Relationships. Credit in Nepal still turns substantially on the banker's confidence in the promoter and the finance head. A tax assessment is handled across a table. A supplier extends terms because of trust built over years. These are human dealings.

Context. A model sees the transaction. The finance head knows that the customer is the owner's brother-in-law, that the stock in the Birgunj godown was damaged in the monsoon, that the price increase was announced but not yet applied.

Ethics. Pressure to present a better picture to a bank, or a worse one to the tax office, arrives through people and must be resisted by people. A tool does what it is asked.

How the working week changes

An illustration of how a finance team's time might shift. The percentages are our estimate for a typical mid-sized company, not survey data.

Seventy per cent of today's time is in the first three rows. In 2030 the same proportion sits in the last four. The team has not disappeared. Its work has moved up.

The team of 2030

What does this mean for people and structure?

Fewer entry-level processing roles. A company that would have employed five juniors on entry and matching will need one or two handling exceptions. In a growing company this usually means not hiring, not dismissing.

A stronger middle. The scarce people will be those who can review machine output critically, investigate anomalies and explain numbers. Today's competent senior accountant, retrained, is this person.

New responsibilities, not necessarily new posts. Someone must own the master data. Someone must maintain the rules and prompts that drive automation. Someone must test controls. In a mid-sized company these are parts of existing jobs.

A finance head who faces the business. Freed from compilation, the head of finance can spend time on pricing, working capital, financing and investment decisions. This is the difference between an accountant and a chief financial officer (CFO versus accountant). Many mid-sized Nepali companies have never had the second. The tools make it affordable to have one, in-house or through a virtual CFO.

The skills that matter

The apprenticeship problem

There is a real danger here, and it is particular to how accountants are formed. In Nepal a chartered accountancy student serves three years of articleship, much of it spent vouching, reconciling and preparing schedules. A junior in a company learns the business by entering its bills. That routine work is how judgement is built. It is also the work that is being automated.

If juniors never do the basics, where do the reviewers of 2035 come from? The KPMG and University of Melbourne study found two thirds of employees who use AI relying on output without checking. A generation trained only to accept output would be a serious loss.

The answer is deliberate training. Have juniors perform reconciliations and returns by hand for a period before they supervise the automated version. Rotate them through exception handling, which is where the interesting cases are. Ask them to find the errors seeded in a machine-prepared schedule. Professional bodies and training firms will need to rethink articleship along the same lines.

The economics for a Nepali company

The labour arithmetic is different from that in rich economies, as discussed in where the value is. Finance salaries in Nepal are low relative to dollar-priced software, so the payback from automating finance is seldom the salary saved. It comes from four other places:

For smaller companies the same logic favours outsourcing. A finance function run by an outside team with good tools and a part-time CFO can deliver more than three in-house staff with none. This is one reason the outsourced model is growing (outsourcing and virtual CFO).

The regulatory frame

Several rules bear on how far and how fast finance can automate.

Responsibility for accounts. Under the Companies Act 2063 directors are responsible for keeping proper books and presenting accounts that give a true and fair view. Automation changes how the books are kept, not who answers for them.

Record retention. Tax law requires records to be kept and produced. Electronic records are recognised under the Electronic Transactions Act 2063, and a company moving to digital documents should confirm that its retention, backup and retrieval meet what the Inland Revenue Department and auditors expect.

Personal data. Payroll and employee records are personal information under the Individual Privacy Act 2075. Putting them into external tools needs care.

Sector regulation. Banks and financial institutions are subject to Nepal Rastra Bank's AI Guidelines, which, as reported, require risk classification, governance, board accountability and human oversight.

Professional standards. Members of ICAN remain bound by their code of ethics, including competence, due care and confidentiality, whatever tools they use. A professional who relies on a tool's output without verification has not exercised due care.

National policy. The National AI Policy 2082 signals risk-based regulation to come. Nothing in it restricts ordinary use of AI in a finance function today.

A roadmap to 2030

Horizon 1, the next 12 months: foundations and assistance. Clean the masters and establish a monthly close (data before algorithms). Switch on the AI features already in the accounting software. Introduce a usage rule. Automate invoice capture and bank matching. Train the team in using and checking the tools.

Horizon 2, years two and three: redesign. Move to exception-based processing with revised approval limits. Bring the close to day 5. Introduce a rolling cash forecast. Automate the compilation of the management pack and compliance schedules. Redefine roles: reviewers, analysts, a data owner. Put controls over automated processes in writing and test them.

Horizon 3, years four and five: continuous finance. Daily positions for cash, receivables and stock. Continuous monitoring of transactions for anomalies. Forecasts that feed directly into purchasing and financing decisions. A finance head who spends most of the week with the business.

Few companies will move in a straight line. The order matters more than the speed: foundations before automation, automation before redesign, control throughout.

The role of consulting and advisory firms

The large professional firms are both commentators on this change and subjects of it. The Big Four have announced investments running to billions of dollars in AI, built internal assistants for audit and tax staff, and begun embedding AI in audit methodology. Their surveys of finance leaders are the main source of data on adoption. Their own experience shows the hazards too: one of them refunded part of a fee in 2025 after apparent AI-generated errors appeared in a government report.

For clients this has two consequences. Audits will become more data-intensive, and a company with clean, complete electronic records will have a smoother and probably cheaper audit than one that hands over files. And advice on finance transformation is widely available, at prices set for large corporations.

For a mid-sized Nepali company, the adviser's role in this transition is practical:

This is our own field. We provide bookkeeping, virtual CFO services, process automation, management reporting and audit support, and we use these tools in that work with the review discipline described in this series. We do not perform statutory audits.

The next 90 days

Days 1 to 30. List the recurring tasks of the finance team and the hours each takes. Mark each by exposure using the table above. Identify the three with the most hours and highest exposure.

Days 31 to 60. Automate the first of the three using existing software features. Record the before and after. Decide what the freed time will be used for, and say so to the team.

Days 61 to 90. Write the review procedure for the automated task: who checks, what, how often. Run a training session on checking machine output. Present the board with a two-year plan for the function: tasks, roles, close date and controls.

Frequently asked questions

*This article is general guidance, not professional advice. Time allocations and exposure ratings are our estimates. Survey figures are from the published sources listed and reflect mainly large companies outside Nepal.*