Charging for AI-assisted work — the ethics · AI for Accountants

A partner in a Lalitpur firm finishes a piece of work in ninety minutes that, before the AI tools came in, would have taken her six hours. She has done it well; the verification was thorough; the client will receive something at least as good as what they would have received last year. She now stares at a half-completed time sheet. Six hours is what the engagement letter expects. Ninety minutes is what the work took. She has to write a number in a box, and whichever number she writes is a quiet statement about what kind of profession she thinks accounting is, and what her firm is actually selling. This final section is about that decision. It does not pretend to give one right answer. It does insist that there are wrong ones, and that the moment of choice is here, not somewhere over the horizon.

The honest framing nobody wants to start with

Hourly billing rests on a simple promise: the client pays for time, the firm sells time, and the rate per hour is the negotiated meeting point. The promise made sense in 1985 and made sense in 2015. It does not, quite, make sense in 2026, because the unit being sold and the unit being bought have come apart. A task that took a competent senior six hours in 2015 takes a competent senior with a well-set-up AI workflow under two. The output is the same. The judgement that went into it is the same. The verification — done properly — is the same. The time, which is what the bill is denominated in, is a third.

If the senior writes six hours on the time sheet anyway, she is — let us be plain — telling a quiet lie. The lie is not enormous on any individual engagement. It compounds. Eventually a junior who joined in 2024 will notice that partners take six hours on tasks she finishes in ninety minutes with the same tools, and she will have a quiet question that the firm will not have a good answer for. Eventually a client whose nephew works in tech will notice the same pattern. The current way of charging cannot survive widespread AI competence at the staff level. The firms that pretend otherwise will be the ones whose staff leave first.

Three honest paths forward

There are essentially three responses that hold together under scrutiny. None of them is comfortable; that is partly the point.

1. Bill actual hours, accept the revenue cut. The cleanest path. The senior writes 1.5 hours on the time sheet because that is what the work took. The bill to the client reflects 1.5 hours at the agreed rate. The firm absorbs the revenue compression on tasks AI accelerates and tries to make up the gap by taking on more work, by raising rates for the human-judgement portion, or by accepting a smaller but cleaner top line. The advantage is moral clarity and the freedom from any conversation that begins with “actually, about that bill.” The cost is real revenue, and any partner who tells you otherwise has not done the arithmetic.

2. Move to value-based or fixed fees agreed up front. The middle path, and probably the right one for most engagements. Instead of billing time, the firm agrees a fee for an outcome — a monthly bookkeeping package, a tax return preparation, an audit — and the fee reflects what the work is worth to the client, not what it costs the firm in hours. AI productivity gains then accrue to the firm as margin, which is acceptable because the client got what they paid for at the price they agreed. This requires real conversations at the start of an engagement; it requires the firm to know what its work is worth; it requires the courage to walk away from clients who insist on hourly billing as a cost-control mechanism. The reward is a business that is not undone by every productivity improvement.

3. Carve out an “AI tooling” line item. The hybrid path. Time is billed at a reduced rate when AI did the heavy lifting, and the engagement letter shows a separate “technology and tooling” charge that recovers the firm’s investment in enterprise AI subscriptions, training, and the local model server in the back room. The client sees what they are paying for. The firm does not pretend the AI is free. This is the easiest path to explain to a long-standing client who is suspicious of any change to the billing model — and the easiest path to abuse, if the technology charge starts quietly outgrowing the value it represents.

Any of the three is defensible. The fourth path — billing six hours for ninety minutes of work, telling nobody, hoping nothing changes — is the one that ends badly.

The disclosure question

There is a related question that every firm will face from clients in the next two years: did you use AI on my engagement, and if so, where? The answer cannot be the marketing-flavoured non-answer of “we use a range of modern tools.” It has to be a real position.

The position that holds up is this: disclosure scales with risk. AI used for an internal first-draft note that a senior then rewrote? No client disclosure needed; the work product is the senior’s, and the AI is no more remarkable than the word processor. AI used to summarise a regulatory document that informed advice given verbally? Probably no formal disclosure, but the engagement file should record it. AI used to draft the actual text of advice that the client will rely on in a tax filing or a board paper? Yes — the file should record it, the partner should know it, and if the client asks, the firm should say so plainly. AI used in any way that affects a number that appears in a filing, a financial statement, or a return? Always recorded, with the verification step also recorded, because that is the workpaper trail a regulator or a successor auditor will look for.

The mistake to avoid is the opposite extreme — disclosing every keystroke of AI use as if to inoculate against future complaint. That tells the client they are paying for a process they no longer understand and invites exactly the conversation about pricing that you are trying not to have. Real disclosure is calibrated. It says, here are the places AI shaped the work, here is how we verified it, here is why we are confident in the result.

The market is already moving

It is tempting, in Nepali practice, to assume that the AI conversation is a few years away — that clients here will not notice, that competitors here will not change, that the existing way of working has another decade in it. This is wrong, and demonstrably so in 2026. The mid-tier firms in Kathmandu are already using enterprise tools. The juniors entering ICAN are already fluent in chat interfaces. The clients — particularly the SME tech businesses, the export-oriented manufacturers, the second-generation owners who studied abroad — already know what AI is and have a rough sense of what it can do.

What they do not have, yet, is a coherent story from their accountant about it. The firm that gets the story right first — “we use AI in these specific places, here is our verification protocol, here is why our fee reflects value not button-presses, here is what we will not use it for and why” — will look like the serious profession. The firm that has no story will look like the firm that did not notice. In any market with even moderate competition, that gap is decisive within three to five years.

The Nepali clients who matter most for a firm’s future — the growing businesses, the second-generation owners, the export-facing companies — will, in 2027 and 2028, be choosing accountants with this question in mind. Not “do you use AI” — they will assume you do. But “do you have a position on it that you can defend.” A firm without a defensible position will not be on the shortlist.

A short closing on professionalism

It is worth ending the chapter, and the course, on a more direct note than any of the practical sections so far.

The mechanical parts of accounting — the data entry, the schedule preparation, the routine reconciliations, the formatting of statements, the drafting of standard letters, the summarising of long documents — have, for almost the entire history of the profession, been most of the day. They were what juniors did to learn the work, and what mid-level staff did because there was simply nothing else to do with the hours. AI changes that. Not all of it, not overnight, but in the direction of less and less of the day being mechanical.

What is left — what AI cannot do and probably will not do for a long time — is what the profession was always claiming to be about anyway. The judgement to know when a tax position is aggressive enough to warrant a conversation with the client. The relationship to be the person the proprietor calls before he signs the loan agreement, not after. The trust that means the bank looks at your signature on the audit report and accepts the figures because it knows the name. These are not mechanical things. They are the things human beings, with reputations and licences and a stake in being the kind of professional their clients describe to others, are uniquely positioned to provide.

That is what you charge for. Not the typing. Not the calculation. Not the formatting. The judgement, the relationship, the trust. AI has, in a sense, done the profession a quiet favour. It has stripped away the mechanical work that was distracting from the actual thing. The accountants who notice — who reorganise their pricing and their practice around what AI cannot do, rather than competing on speed at what it can — will not just survive. They will look, ten years from now, like the people who first understood what accounting was always supposed to be.

What comes next

This was the final section of AI for Accountants. What remains is the course exam — ten questions drawn from across the full course, covering everything from where AI fits in a Nepali small firm, through bookkeeping and entry, tax and compliance, reports and analysis, client communication, and the safe-habits material of this chapter. A score of 80% is required to pass. The exam is here: /courses/ai-for-accountants/exam. Take it when you have a clear half hour and the course material fresh in mind.