Accounting & Finance: the AI discount now has a number
Accounting and finance is the field where AI work got a public price. Since Era Haus last looked in late July, a US advisory firm priced agent-run assessments at 30 to 50 percent below the human version, accounting platforms pushed their agents into the work nobody bills for, and the EU's AI transparency duties came into force. The shift for a small practice: routine work is now priced in public by firms running agents, and Europe's compliance clock has started.
The discount got a number
On 1 September 2026 Cherry Hill Advisory, a US firm that reviews other companies' internal-audit departments, announced assessments run mostly by two AI agents, priced 30 to 50 percent below its traditional engagements and roughly 50 percent faster, per Accounting Today. That is professional-standards work, one firm certifying that another's internal audit meets the profession's rules, priced until now on the assessor's reputation. A qualified person still confirms every step, the firm says, and the figures are its own. Public is what matters: a discount one competitor advertises becomes the number a client quotes back at you.
Capital moved the same way. On 29 July Grant Thornton's US arm announced it would buy CBIZ, another US accounting and advisory firm, for $5 billion, the biggest deal of its kind in more than 25 years, tied in the announcement to a $1 billion AI investment. In Bloomberg Tax on 24 August the accounting academic Andrew Belnap named the mechanism: AI raises the minimum size at which a firm can afford to stay current.
Software went after the work nobody bills for
The vendor news points one way: agents took the chasing and checking; judgment stayed with people.
Xero, the accounting platform with about five million customers, mainly in Australia, the UK and the US, pushed its agent into that work at its 19 August 2026 conference. The agent, called JAX, is being extended to email a client for a missing receipt, chase it, and match it to the transaction when it arrives; Xero says that part is coming soon (Accounting Today, 24 August 2026). It times payment chasing to how each customer actually pays, and inspects every bill before payment, flagging an unusual amount, changed details, or a supplier never paid before. Sage, the UK-based software group, put the same fraud check into Intacct, its finance system for mid-sized companies. It shipped in August, per CPA Practice Advisor on 25 August.
RSM, a large US audit firm, has an AI control-testing platform in live use with its own auditors (Accounting Today, 28 August 2026): it reads the evidence, tests each control and writes a review-ready work paper, flagging anything ambiguous for a person. Preparing that work paper was what a junior spent two years learning; it is becoming a review task.
Latin America moved on its own software. Alegra, an accounting platform used in eight countries, mostly in Latin America, followed June's connector for outside AI assistants with an agent inside the software on 29 July: an accountant asks in plain Spanish which invoices are overdue and gets the answer from the live books (La Opinión, 29 July 2026). Colombia went first, three more countries from 5 August, and on 11 August the company ran an AI summit for accountants in Mexico City with the tax authority on the program.
Europe's rules went live; the US stayed on guidance
In July's piece, Accounting & Finance: AI moved into the close, and the rules followed, the EU's rules were due in August. They arrived, and they split. On 2 August 2026 the EU AI Act became generally applicable, so its transparency duties are live: an EU practice with an AI assistant on its client portal must tell clients they are dealing with a machine, and text an AI writes for publication on a matter of public interest must be marked unless a person reviewed it and holds editorial responsibility (European Commission, 31 July 2026). The high-risk regime, covering credit and employment decisions, moved to 2 December 2027, a date now fixed.
In the United States no new audit or ethics rule was written. The Public Company Accounting Oversight Board, the US audit regulator, asked where to set standards next; in comment letters due 7 August the four largest firms said that on AI they wanted guidance rather than rules (the accounting news site Going Concern, 12 August 2026). On 31 August the Journal of Accountancy confirmed there is no AI-specific ethics rule: competence, due care and client confidentiality already cover it. Proving an AI-assisted procedure was adequate is on you, and on the evidence you kept.
What it means for you
Pricing: a public 30 to 50 percent discount for agent-run work exists, and a client who has read about it will ask why your fee is the same. Staffing: the tasks a first-year learned on, chasing documents and preparing the work paper, are moving into the software, so a junior's first two years need rebuilding around review and client contact. Compliance: in the EU the disclosure duty is live today; in the US and Latin America it is the one you already had, proof that a person checked the machine.
What to do about it
Settle your pricing answer before a client quotes the discount at you: a fixed fee for the reviewed, signed result, and stop billing the agent's preparation by the hour. Switch on the checks you already pay for: invoice-fraud flags are already in the platforms above; document chasing is arriving. If you run an AI assistant on your client portal in the EU, add the one-line disclosure this week.
Watch rather than act on: an agent running an engagement end to end with no named reviewer. Cherry Hill kept a qualified person on every step of its cheaper product; that is the bar.
- Jul 28, 2026Accounting & Finance: AI moved into the close, and the rules followed
- Jun 03, 2026Accounting & Finance: AI is now inside the books