Treadstone Associates
Article · Client reporting

AI for management reporting to clients

The drafting is automatable. The judgment, the basis of accounting and the sign-off are not — and one of them decides which standard you are working under.

Treadstone Associates · Updated 2026

Key takeaways

  • • AI drafts commentary and query lists; the ledger produces the numbers.
  • • CSRS 4200 scope turns on whether a communication is attached to the financial information.
  • • Every figure quoted in commentary must be traced back to the pack before it goes out.
  • • This is a practice serving a book of clients, not a company closing its own month.

The short answer, and the boundary that decides it

AI can draft the narrative around client management reports — variance commentary, the covering note, the query list, the consistency checks between this month and last. It cannot decide the basis of accounting, decide an adjusting entry, or take responsibility for what the numbers say. And the moment your firm issues a communication associated with financial information it prepared, you may have moved from an internal reporting service into a compilation engagement, which is a different standard with different documentation.

That boundary is the whole article. Get it right and AI is a straightforward productivity gain across a reporting book. Get it wrong and you have automated your way into a standards problem.

This is not the client’s month-end close

Worth separating two things that get called the same name. A company automating its own reconciliations, AP and AR is doing internal finance work, and we cover that on our AI in accounting operations page. A practice producing monthly or quarterly reporting packs for thirty different clients is doing something else: same task repeated across thirty charts of accounts, thirty sets of expectations, and thirty relationships where the commentary is the deliverable.

The second is where AI pays, because the repetition is across clients rather than across months, and because the expensive part — writing something intelligent about the numbers — is the part currently done last, badly, at eleven at night.

Where CSRS 4200 comes into it

The Auditing and Assurance Standards Board’s own account of the new compilation engagements standard is the place to start. CSRS 4200 replaced Section 9200 and the related assurance guideline, and among its stated key features is a scope that clarifies which services are compilation engagements: a bookkeeping service may result in system-generated financial information, and such information is excluded from the scope of the standard if no communication is included or attached to it.

Read that twice before you let a tool start attaching your firm’s letterhead to an automated monthly pack. The trigger is not the sophistication of the report. It is whether a communication is included or attached to the financial information you produced.

The standard takes effect for compiled financial information for periods ending on or after December 14, 2021, with early application permitted. If you are inside it, the AASB flags three things that bear directly on an automated reporting workflow: specific engagement acceptance considerations that apply when the compiled financial information is intended to be used by a third party; specific required work effort and documentation; and a requirement that the compiled financial information includes a note describing the basis of accounting applied. The old Notice to Reader is replaced by a compilation engagement report.

Those are documentation obligations, and documentation obligations are the ones automation genuinely helps with — provided the tool records what happened rather than generating a plausible account of what usually happens. the Auditing and Assurance Standards Board covers audit, review, compilation and other assurance engagements, and the reporting language differs across them.

What AI should actually be doing in a reporting pack

Variance commentary, first draft. Given this period, the comparative and the budget, produce a paragraph naming the three largest movements and the account-level drivers. Reviewer keeps the causal claims and deletes the invented ones.

The query list. Transactions that break the client’s own pattern — a recurring expense that stopped, a new vendor above a threshold, a coding change — assembled into questions for the client instead of found in week three.

Consistency checks across periods. Did a classification change without an explanation? Did a note say “seasonal” last quarter and “one-off” this one? Prose consistency is tedious for humans and trivial for software.

The covering note. The two paragraphs the owner-manager actually reads, drafted from the same inputs each month so the report stops arriving with a different structure every time.

Not the numbers. Extraction and calculation should come from the ledger and the reporting layer, not from a language model reading a PDF. Tools such as Xero’s practice software hold client data in one database with workpapers and query handling attached; that is where the figures come from. The model writes about them.

The sign-off that has to survive

Two review steps are non-negotiable. First, someone traces every number quoted in the commentary back to the report it came from, because a fluent paragraph citing a figure that is not in the pack is the single most damaging output this workflow can produce. Second, a named person signs the pack, and the file records who reviewed what.

Using software does not move professional exposure. Professional negligence claims against accountants are assessed on what the firm did, not on what the firm’s tooling did, and a client who relied on a commentary paragraph will not be interested in how it was generated.

A worked example: thirty owner-managed clients, monthly

A four-person practice produces monthly packs for thirty owner-managed companies. Historically the numbers were out by day eight and the commentary by day fifteen, which meant half the clients read commentary about a month they had stopped thinking about.

The change: the reporting layer produces the statements as before. A drafting step then generates variance commentary and a client query list for each entity from the same data, into a review queue. The manager reviews thirty drafts in one sitting rather than writing thirty from cold, keeping what is true and rewriting what is guessed. Packs where the standard applies keep their existing acceptance documentation and basis-of-accounting note unchanged.

Measure two things: days from period end to pack delivery, and the number of client questions that arrive after delivery. The second is the honest quality signal. Faster packs that generate the same questions have not improved anything except your calendar.

Questions we get asked

Does an AI-drafted commentary change our engagement type?
The drafting method does not. What matters is whether a communication is included or attached to the financial information your firm produced, which is the scope question the standard addresses. Decide that first, then choose tooling.

Can AI decide the basis of accounting?
No. The compiled financial information must carry a note describing the basis of accounting applied, and settling what that basis is remains a professional discussion with a client rather than an output.

Where does a small practice start?
With the client query list. It is generated from data you already hold, it is checked simply by looking at the transactions, and it earns time back in the same month rather than the next quarter.

See where AI pays off first in your firm.

A 30-minute call is enough to tell you whether AI pays for itself here.