This issue only includes two publicly available materials that can be broken down into roles, inputs, review, and outputs. For other sections, cross-reference where possible; where independent evidence cannot be assembled, write “Data unavailable.” Founder posts are treated as public statements without third-party cross-verification.
Today’s Most Actionable (2 items)
1. Multi-Entity Finance: Ledgers Remain in ERP, Rules Move to Code Repository, Agents Segregated by Role
- Scenario: Mid-sized group with ~400 employees, 20+ entities, payroll in 5 countries, multi-currency consolidation. Founder publicly stated finance headcount reduced from 7 to 0; the last FM to spend one close cycle auditing rules before transferring roles.
- Actions: Do not reduce headcount first. Establish “system separation”: Sage Intacct (or existing ERP) as the official ledger only; GitHub for rules/exceptions/changes; Postgres for operational events. One agent each for AP / AR / Treasury / Controller / Internal Audit, permissions aligned with roles, no superuser accounts. Supplier, collections, bank four mailboxes unattended for attachments; new suppliers must pass chatbot approval + bank account verification, then written to ledger after 24 hours.
- Review Controls: Voucher-level permissions per bank-grade custody; changes via Git history; payments/new suppliers/close still require human sign-off. Author acknowledges pushing ERP migration twice in 3 years reduced team size and advocates “still need one person driving.” Do not treat “hiring a growth lead without finance knowledge to watch numbers” as an executable staffing plan.
- Deliverables: Role permission matrix, rules repository, close task board, read-only query scopes (department heads/shareholders see their own).
- Source: Founder public statement, exact date undisclosed. Gerald Neves post
2. Month-End Close Workspace: Documents, Suggested Entries, Open Issues, Unclosed Items in Four Sections, Not Another Chat Window
- Scenario: Firm/accounting team month-end close. Clients upload statements and attachments.
- Actions: System extracts transactions, flags missing items, identifies duplicates and intercompany transfers, generates classification and entry drafts based on client historical rules, with evidence attached to entries. Humans only handle four categories: unseen transactions, ambiguous accounting treatment, material adjustments, questions requiring client input. After issues closed, system reruns reconciliations and checks; humans review accounts and variance analysis, approve the period, then open to client workspace. After corrections, can propose “apply same treatment going forward,” but still requires human approval to write into rules.
- Review Controls: Ledger capabilities should not be replaced by agents. Balancing entries, opening balances, reconciliations, period locking, correction trails must have reliable underlying systems. Without these, the workspace only makes errors look prettier.
- Deliverables: Four-column workspace (Documents / Suggested Entries / Open Issues / Unclosed Items) + period approval log.
- Source: Practicing CPA public workflow description, exact date undisclosed. Charlie Barmore post
Accounting / Close / Controls
- Month-end close four-piece set: see Today’s Most Actionable item 2. This week select only one entity and one expense category for a trial run; Controller reviews only the “ambiguous + material” columns.
- AP mailbox and new supplier onboarding: see Today’s Most Actionable item 1. Can be extended to invoice extraction, but automatic payments remain disabled.
- Other independent close/reconciliation cases: Data unavailable.
FP&A / Planning / Reporting
- Scenario: Evaluating Excel-connected budgeting/forecasting tools or stress-testing existing “appears connected” models for breakage in month 6.
- Actions: On demo day do not stop at “bring your own spreadsheet, refresh once.” After refresh perform two additional tests: insert a row, rename a sheet, add an account, then refresh again and ask whether mapping is anchored to cell coordinates, named ranges, or account codes; request a second login to verify whether role permissions and audit trails exist only in the demo tenant.
- Review Controls: Cell-coordinate binding means every future table change requires finance to remap. Failure to surface permissions under a second account should be treated as absent governance capability.
- Deliverables: One-page procurement/acceptance checklist (structural change test + second-account test + mapping anchor answers).
- Source: Vendor materials, page shows 2026-08-25 update. Datarails: Two Excel Connectivity Tests
- Independent variance commentary / board package cases: Data unavailable.
Treasury / Cash / Risk
See Today’s Most Actionable item 1: author frames Treasury, collections, cash, FX, payments, and dunning as fixed-role agents and stresses that payment egress must not receive superuser permissions. No additional verifiable 13-week cash forecast or bank-statement reconciliation workflow identified this period.
Data unavailable. Apart from the cross-references above, no independently actionable cash forecasting/DSO cases were found.
Tax / Compliance / Audit
Data unavailable. No new AI implementation cases or practical methods for tax research, SOX/internal control, or audit evidence management within the last 365 days were identified.
(The Internal Audit role and “last FM spends one close cycle auditing rules” mentioned in item 1 represent organizational intent only; no working-paper fields, sampling rules, or evidence-pack structure are provided, therefore not recorded in this section.)
CFO / Leader Team Building Experience
- See Today’s Most Actionable item 1. Only three points can be treated as executable staffing experiments: role-based segregation (AP/AR/Treasury/Controller/IA do not cross permissions); close rules audited through one full cycle before headcount reduction; author explicitly states one person must still drive. Other claims such as “zero-person finance” or “hire a growth lead to run finance” are viewed as high-risk and not recommended for adoption.
- Public sharing of training, owner division of duties, or ROI metrics by CFOs/Controllers at well-known enterprises: Data unavailable.
- LinkedIn public post data unavailable; job title/post summaries are not treated as factual cases.
Open Source / AI Engineering Reference
- Reusable architecture: see Today’s Most Actionable item 1 — separate official ledger (ERP) from operational database (Postgres); move rules and exceptions into Git; isolate runtime and credentials by agent; MCP/read-only queries authorized per person. Suitable for data-flow diagramming first; not suitable for migrating the ledger out of ERP.
- Independent GitHub / n8n / OCR repositories or runnable templates: Data unavailable.
This Week’s Small Experiments
-
New Supplier Onboarding (AP owner, Controller final approval) Connect only one invoice receipt mailbox. Extract supplier name, amount, and account; prohibit automatic payments. Output: extraction table + manual approval log. Passing standard: extraction accuracy rate and missed-extraction rate tracked for one week.
-
Month-End Close Four-Column Table (Accounting owner, Controller sign-off) Select one expense category. Columns: Documents, Suggested Entries, Open Issues, Unclosed Items. Humans handle only ambiguous and material items. Output: current-period workpaper. Passing standard: 100% of open issues must carry a sign-off before “close period” can be clicked.
-
Excel Connectivity Acceptance (FP&A owner) Take the budget model currently in use and run the following against the existing or candidate tool: insert row / rename sheet then refresh + second-account login. Output: one-page test record. Any failure disqualifies the tool from the next quarterly budget cycle.
-
One-Page Role Permission Matrix (CFO approval, IA co-sign) By AP / AR / Treasury / Controller list: what can be read, what can be drafted, what must never be done (especially payments and new suppliers). Constrain humans first, then constrain agents. Output: permission matrix. Without this document, do not pilot the multi-agent setup described in item 1.