finance-ops

Use when setting up or reviewing the company's own financial operations (expense approval, invoicing, budgeting). Building a payment or ledger product is business/fintech-compliance.

SKILL.md

GitHub

finance-ops

Business layer (business/README.md). Distinct from business/fintech-compliance: that block is engineering invariants for a product that moves other people's money (a ledger, a payment flow); this one is the structural discipline around the company's own financial operations — who can approve what, how invoicing gets chased, how a budget gets tracked. Same honesty posture as the rest of this layer: not accounting or tax advice.

When

Setting up or reviewing who can approve an expense or a payment, how invoices/receivables are tracked, or how spend against a budget is monitored.

Steps

1. Separation of duties: the core control

  1. The person who requests or benefits from a spend never also approves it. This is the single highest-value control against both error and fraud — not because any specific person is suspected, but because a control that depends on one person's honesty rather than on a structural check is a control that fails exactly when honesty does.
  2. The person who approves a transaction is not the same person who reconciles or reviews it afterwards. Approval and review are two different checks; collapsing them into one person removes the second check entirely.
  3. A small team can't fully separate duties, and that's expected — use compensating controls instead of pretending the separation exists. A founder/owner personally reviewing bank and card statements monthly, requiring a second signature above a stated amount, or rotating who handles a given process periodically all substitute for headcount the team doesn't have. Don't skip the control because the textbook version needs more people than exist.
  4. Detection is faster with the control in place than a post-hoc audit relies on: organisations with real separation-of-duties controls catch discrepancies materially faster than those without, which directly caps how large a problem gets before it's noticed. Re-read before relying: the fraud survey that would support this figure could not be read on 2026-10-20 (the file exceeded the fetch limit), so this sentence is unsourced.
  5. Separate authority, custody and accounting, and expect limits. The US federal internal-control standard (Green Book, paragraphs 10.12 to 10.14, read 2026-10-20) says management considers splitting authority, custody and accounting so no one person controls a whole transaction; where that is not practical because of limited staff, it designs alternative controls; and separation reduces but cannot remove the risk of managers overriding controls, nor of collusion between two people. Symptom: one person who approves, pays and reconciles. Fix: a named alternative control (for example an owner reviewing statements, a second signature above an amount), and say that it is weaker than separation. A US federal reference, used for the principle only.

2. Invoicing and receivables

  1. An invoice has a stated due date and payment terms from the moment it's sent, not an implicit "whenever." Ambiguity here is what turns into a slow-payer problem later. Jurisdiction example only: the French business-service page read 2026-10-20 says a French invoice must show the payment due date, the early-payment discount terms or a statement that there are none, the late-payment penalty rate, and the fixed 40 euro recovery indemnity owed by a professional customer who pays late, with penalties due without a reminder; the maximum payment delay was not found on the pages read. Other countries differ: ask whoever owns finance.
  2. Overdue receivables are tracked and chased on a schedule, not noticed when cashflow gets tight. A simple aging view (current / 30 / 60 / 90+ days overdue) surfaces the problem while it's still small.
  3. A recurring late payer is a decision to make explicitly (stricter terms, upfront deposit, involving someone senior) rather than an accepted pattern nobody has actually decided to accept.

3. Budgeting and spend visibility

  1. A budget is a tracked commitment, not a document written once and revisited at year-end. Spend against it reviewed on a cadence (monthly is common) catches an overrun early enough to act on it, rather than as a year-end surprise.
  2. Every recurring spend (subscriptions, contracts) has a named owner who can say why it exists and whether it's still needed — an unowned recurring charge is how a company keeps paying for a tool nobody uses.
  3. A one-off spend above a stated threshold gets the same "requester ≠ approver" check as routine expenses — a threshold exists precisely so a large exception doesn't slip through because it felt too unusual for the normal process to apply.

Output / checkpoint

No pipeline checkpoint (business layer, see business/README.md). What it owes: a stated separation between who requests/benefits from a spend and who approves it (or a named compensating control where headcount doesn't allow real separation), an aging view of receivables reviewed on a schedule, and a named owner per recurring spend with budget tracked against actuals on a cadence.

Guardrails

  • Never let the same person both request/benefit from and approve the same transaction, without an explicit compensating control replacing that check.
  • Never let an overdue invoice go unchased until cashflow makes it urgent — track on a schedule, not on pressure.
  • Never leave a recurring subscription/contract with no named owner.
  • This block has no dedicated in-house finance/accounting expertise behind it yet: a solid set of structural defaults, not proven doctrine, and not accounting or tax advice — a real bookkeeper or controller's judgement outranks it without discussion.

Origin

Sourcing pass 2026-10-20 (status stays 🟡, nothing was run). Read for facts, own words: the US federal internal-control standard (a US government publication; section 10 on segregation of duties; the PDF text was extracted locally), a French business-service page on invoice mentions, a small-business agency page on budgeting and receivables, and a COSO page. Confirmed: the separation-of-duties principle (step 1.1 to 1.3) including the alternative-control case for small teams. Added: the limits of separation (override, collusion) and the French invoice-mention example. Not found: any source for the faster-detection claim (1.4, now marked), for aging buckets or for a budget review cadence (the agency page gave only generic tracking advice and the COSO page did not name the framework components); those stay own guidance. The French payment delay page returned the wrong content and the commercial-code article could not be opened.

Assembled from public sources, written without internal finance/accounting expertise. Separation of duties (requester ≠ approver ≠ reviewer, compensating controls for small teams, faster fraud detection with the control in place): published internal-controls guidance aimed at small businesses and nonprofits, where the headcount-constrained case is explicitly addressed rather than assumed away. Invoicing/receivables aging and budget-as-tracked-commitment: general financial-operations practice, generalised rather than tied to one accounting platform. Stamped 2026-08-11.

Independent verification 2026-10-21: 9 checked, 8 matched, 0 corrected, 1 unverified