For October 26–November 2, the most useful developments concern close preparation, approval controls, and the ability to trace changes made by AI. This issue distinguishes pilot access from vendor-documented releases and automation claims from measured close outcomes. The sources reviewed are vendor announcements, product documentation, and release notes; they do not provide independent implementation benchmarks for the workflows discussed. No hands-on product tests were performed for this issue.
Intuit: Books Close beta and Agentic Books Close pilot are different offers
What changed. Intuit’s fall release describes Books Close beta in Core and Accelerate plans, with review templates, statuses, collaboration, and anomaly detection. Separately, Agentic Books Close is a sign-up pilot for a limited group of firms. Its proposed scope includes reconciliation, anomaly review, accrual schedules and entries, and closing reports such as variance analysis and AP/AR aging.
Evidence status
Vendor-documented beta availability and a limited pilot, not general availability of autonomous close. The footnote requires review and approval of every AI-drafted entry before posting, warns of possible inaccuracies, and says continued access will require a paid add-on after the pilot and beta. Intuit Intelligence chat is currently admin-only, with capacity depending on plan. Confirm actual eligibility and enabled features before planning an implementation.
Controller’s lens. Editorial assessment: evaluate preparation time and reviewer workload separately. Before a pilot, define the transaction population, cutoff, capitalization policy, allocation conventions, and approval matrix. A draft prepaid or fixed-asset schedule is not evidence that capitalization is appropriate. Agree on the accounting conclusion independently, then test whether the tool applies it and retains review evidence. Do not budget for broadly available agentic close on the basis of a pilot announcement.
Rillet: examine the approval design behind the zero-day headline
What changed. Rillet’s October 27 announcement promotes continuous close and claims that its architecture posts 99.7% of transactions automatically. Its October product notes describe a Command Center for approvals and integration failures, expanded expense approval controls, and Aura support for deferred-cost creation and amortization lookups. The monthly notes describe October changes; they do not establish the exact ship date of each feature.
Evidence status
The 99.7% figure is a vendor claim. The announcement does not define its denominator, measurement period, error rate, or independent validation. Aura documentation states that data-changing actions require approval, with records linked to their originating conversation or agent. These statements concern different scopes: overall transaction automation and Aura action approval. Neither proves that every material account is reconciled, estimates are complete, or reviewers have finished at period-end.
Controller’s lens. Editorial assessment: ask for automation rates by transaction type and value, alongside correction rates and unresolved material balances. In a sandbox, verify whether the proposal, approval, executed result, and source support can be reconstructed. Test integration failures as well as successful postings. Faster deferred-cost setup is useful only when the approved policy, amount, start date, term, and subsequent changes carry through correctly. Reporting speed and posting coverage should not substitute for close completion criteria.
Balances: actor attribution and reconciliation steps become more visible
What changed. Balances’ October 30 changelog adds a Made by activity filter distinguishing its assistant, connected apps, and people, with an Any AI view. It expands recorded events such as attachment changes, Stripe refunds, payouts, and disconnected feeds. Bank statements can open a prefilled reconciliation screen from the Inbox. Its Xero import progress screen reports balance tie-outs, but the notes exclude certain source books, including those using invoices, bills or payments, sales tax on bank transactions, or another currency.
Evidence status
These are vendor-documented shipped changes. An activity filter does not establish that history is immutable, complete, exportable, or adequate for a particular audit. A displayed migration tie-out does not establish that all required source records and supporting documents migrated.
Controller’s lens. Editorial assessment: actor attribution makes investigation easier, but verify the before-and-after values, timestamps, source links, reviewer decisions, and retention behavior. Treat import eligibility as an initial system-selection gate. A reconciliation screen prefilled from a statement still needs an independent check of the statement date, ending balance, outstanding items, and source completeness.
Experiments worth running
Use synthetic or appropriately anonymized data. These are proposed tests, not completed experiments.
Close pilot acceptance test: separate prep from judgment
Test: Use a sandbox with synthetic transactions covering a prepaid, a proposed capital purchase, an omitted invoice, a duplicate, and an unresolved estimate. A qualified reviewer sets expected treatments first. Compare a manual baseline with the assisted workflow.
Evidence to retain: Keep the approved policy, expected answers, population totals, draft entries, approvals, final ledger, exceptions, and separate preparation and review times. Record product version, plan, pilot status, and enabled features.
Success criteria: Every planted exception is surfaced; no AI-drafted entry posts before the required approval; final entries match the approved treatments; unresolved estimates remain visible. Report any time saving only after including correction and review time.
Automation denominator test: what does the percentage leave out?
Test: Create a synthetic close scorecard separating transaction count, transaction value, recurring entries, estimates, reconciliations, and review sign-offs. Include a high-value manual adjustment and a failed integration. Use it to structure a vendor demonstration or authorized sandbox trial.
Evidence to retain: Retain the population definitions, counts and values, excluded items, automation logs, corrections, failed-job queue, and reviewer sign-offs. Label a demonstration as vendor-run and a sandbox as tested only after it is actually run.
Success criteria: Every automation percentage has a defined population and period; counts reconcile to totals; high-value exceptions remain visible; failed integrations block the relevant completion status. A high posting rate cannot by itself produce a completed-close rating.
Change-history drill: can another reviewer reconstruct the result?
Test: In an authorized sandbox, have a human and an AI make separate approved coding changes to synthetic records, replace an attachment, and reverse one transaction. Ask a second reviewer to reconstruct the sequence using the product’s history and available exports.
Evidence to retain: Keep known starting values, intended changes, record IDs, actor and approver identities, timestamps, original and replacement support, reversals, and exported history where available. Document unavailable fields or export functions.
Success criteria: The reviewer can trace each final balance to the original record and every intervening change without relying on chat recollection. Missing history, ambiguous actors, or unsupported retention and immutability claims are recorded as limitations.
Start with the release status, then define the accounting acceptance test. Pilot access is not broad availability; automatic posting is not a completed close; actor attribution is not proof of an adequate audit trail. A useful Rebuild Accounting write-up would publish the population definitions, review effort, exceptions, and evidence gaps—not just a completion percentage. The experiments above are proposals and have not been run. Prepared with AI assistance from sources checked on November 2, 2026; no personal testing, endorsement, or review by Samantha Ye is implied.
Prepared with AI assistance for Rebuild Accounting. Sources were checked for this issue; product availability and terms can change. Editorial assessments and proposed tests are identified separately from vendor documentation. This issue does not claim hands-on testing or personal endorsement.
← Back to all issues