What AI takes from your accountant, and what stays
On 24 August, August Aquila, a consultant who has advised accounting firms for four decades, published an assessment of which part of a firm's work AI already absorbs and which part stays with the license holder. This is one person's opinion, not a rule and not regulation, and the figures in it describe American companies with revenue between 10 and 70 million dollars. The split it proposes is still worth reading, because it holds at a much smaller scale.
The figures, and who they belong to
A typical finance team burns seven to 15 days closing a month. Putting AI into that process cuts several days off, and reduces the overall workload by 40% to 60%. In money that comes to annual labor savings of 85,000 to 195,000 dollars, plus a 20% to 50% drop in the volume of work bought from outside firms.
The worked example is a manufacturer with 40 million dollars in revenue. The saving there is 150 hours a month at 80 dollars an hour, so 12,000 dollars a month or 144,000 dollars a year. Setup cost 30,000 dollars, upkeep 24,000 dollars a year, which makes 54,000 dollars in the first year against a net benefit of roughly 90,000. Payback lands after four to five months.
At a 30 million dollar company, the assessment counts 300 hours a month that fall into this category: 120 hours of reporting, 80 of reconciliation, 60 of variance analysis and 40 spent preparing board material. Of those, 120 to 180 hours can be cut, at 60 to 90 dollars an hour.
A one-person business in Israel recognises itself in none of those numbers. What carries over is the separation behind them.
What moves to the machine
Aquila's list starts with the month-end close, where the system gathers data, flags errors and proposes journal entries. Then report writing, with AI producing a first draft instead of someone writing the explanations from scratch. Variance analysis that used to take hours finishes in minutes. Board packages, forecasts that account for history and seasonality, detection of duplicates and miscoded transactions, and ad hoc questions answered through a conversational interface.
What all of these share is that the work starts from data already sitting in the system and ends in output you can check against that same data.
What stays with the license holder
Preparing and signing a tax return, audit work, reviews and compilations of financial statements, tax strategy, and CFO-level advice. The reason given is the same each time: the work needs judgment, knowledge of legislation that keeps changing, and legal responsibility that someone has to carry.
There is one more clause that is easy to skip past. Someone still has to read the automated output, because an AI system does not always produce what you expected. The hours saved arrive after that review, not instead of it.
The shift worth noticing
Aquila's conclusion is that AI does not erase the accounting firm, it changes what you buy from it. Businesses move the routine work in-house to staff supported by AI, and leave the firm the parts that need a license and judgment.
For a self-employed person in Israel, the bill from an accountant or tax adviser breaks down along similar lines. One part is data handling: taking in invoices, reconciling, arranging the material ahead of a filing. The other part is judgment and signature. The first part is the one that shrinks, and it is also the part you can shorten today with no AI at all, simply by handing over material that already makes sense. Bookkeeping for the self-employed covers what that looks like in practice.
The assessment talks about four to five months to recover an investment at a company paying 80 dollars an hour for internal work. In a business where that hour is your own, the arithmetic looks nothing alike, and the first question is how many of your hours genuinely count as arranging data.