Reported operating use · Accounting and financial operations
AI month-end accruals: Campfire example
Campfire, an accounting platform, reports using Anthropic Claude Sonnet within its Ember assistant for month-end accrual analysis, financial queries and close-checklist work. A March 2026 product update says Ember 1.4 added sandboxed Python execution after earlier arithmetic errors. The source does not establish accounting accuracy.
Month-end accrual analysis, financial queries, code-executed calculations and close-checklist integration
Campfire
Customer-facing accounting capabilities shipped; Ember 1.4 actions corroborated by March 1, 2026 product update.
A published third-party example, not a Nuvepro customer case or evidence of a partnership.
The workflow at a glance
Campfire reported work
Illustrative summary of the published account
Reported work components. Their arrangement does not imply a fixed execution order or simultaneous work.
- Accounting platform
Recurring accrual analysis
Accounting work includes reconciliation and close integration.
- Calculation tool
Sandboxed Python
Shipped execution addresses earlier plain-text arithmetic mistakes.
What the source reports
The March 2026 update reports Ember 1.4 shipped with sandboxed Python execution after prior plain-text arithmetic errors. It also describes recurring accrual analysis. The earlier Anthropic case describes reconciliation and external-data integration. Code execution improves numerical reliability but does not establish whether accounting assumptions or business classifications are correct.
Reported technology: Anthropic Claude Sonnet within Ember. Scope and outcomes are source-reported.
What the enterprise environment would need
The following is a proposed project design for an enterprise doing similar work. Published details are marked; the remaining systems, data and scenarios would be agreed for your project.
Proposed practice environment
Reconcile accruals and month-end balances
Illustrative project design to scope with your enterprise.
- 1 · Set up the environment
Frozen close ledger
Fictional periods, currencies, balances, payments and supporting invoices stay linked.
- 2 · Preparation and agent contributions
Analyze accruals and balances
Run calculations and prepare reconciliation with accounting assumptions visible.
- 3 · Independent checks
Recompute and inspect policy
Independent reconciliation checks period, currency and source inputs beyond successful execution.
- 4 · People decide acceptance
Controller accepts close evidence
Accountant resolves differences; controller reviews judgments for the exact practice version.
A reviewable evidence package
A reconciliation pack, accrual assumptions and unresolved close items.
If a check fails → withhold acceptance.
Correct arithmetic uses the wrong period or includes a duplicate invoice. Withhold acceptance, repair, independently recheck and return to owner review.
Failure or owner rejection returns to review
- Hold
- → 2 · Revise work
- → 3 · Rerun independent checks
- → 4 · Owner review
Operating boundary. Test ledger and sandbox only; no production posting or close authorization.
- Systems and tools
- Published: accounting platform and Python sandbox. Proposed: reuse test ledgers and real calculation runtime.
- Data and state
- Proposed: preserve ledger balances, periods, currencies and linked transaction evidence.
- People and handoffs
- Proposed: accountant investigates differences; controller approves judgments and close.
- Rules and authority
- Proposed: posting rights, period locks and accrual policy must match actual authority.
- Normal and exception cases
- Proposed: partial payment, duplicate invoice, rounding discrepancy and missing supporting record.
- Reset and replay
- Proposed: frozen ledger and checklist snapshots; no published reset protocol.
A proposed simulation
An illustrative practice design derived from this work, not an executed assessment.
Analyze accruals and reconcile a close with partial payments, duplicate invoices and period boundaries; submit unresolved judgments to a controller.
Review gate: Independently reconcile balances and source inputs; check period, currency, policy and approval, not just whether Python ran.
Challenge the result: Provide correctly executable code using the wrong accounting period. The readiness check must fail the result despite valid arithmetic.
What needs checking
- Published: code-executed numerical results address prior calculation errors.
- Proposed: independent ledger reconciliation and controller-reviewed assumptions; executing code alone is not proof.
- Proposed: validate chart of accounts, period rules and approval sequence against the customer reference tenant.
Review environment fidelity, human capability and agent performance separately. The enterprise’s accountable owners decide acceptance of the shared workflow and any operating use.
How an enterprise could build similar work
A proposed build order for an enterprise doing similar work, written by Nuvepro from the published account and our interpretation of the environment and checks above. It is not how Campfire built or operates this work.
- Frame a proposed test projectThis is a proposed design for a similar project, not a description of how Campfire built Ember. Define the mission: analyze accruals and reconcile a close with partial payments, duplicate invoices and period boundaries, sending unresolved judgments to a controller.
- Preserve ledger data and stateKeep ledger balances, periods, currencies and linked transaction evidence in the test environment. Reuse test ledgers and a real calculation runtime. Freeze ledger and checklist snapshots so every run starts from the same state.
- Match rules to real authorityMake posting rights, period locks and accrual policy match actual authority. An accountant investigates differences, and a controller approves judgments and the close. Unresolved judgments go to the controller instead of being settled automatically.
- Add exception casesInclude partial payment, duplicate invoice, rounding discrepancy and missing supporting record. These cases show whether the work holds up when the books are not clean, rather than only on the routine path.
- Reconcile balances independentlyIndependently reconcile balances and source inputs. Check period, currency, policy and approval, not just whether Python ran. Validate the chart of accounts, period rules and approval sequence against a customer reference tenant.
- Run the failure testProvide correctly executable code that uses the wrong accounting period. The readiness check must fail the result despite valid arithmetic. No reset protocol is published, so replay from frozen snapshots is also proposed.
Your systems, data, rules and accountable owners decide the real project. Agree acceptance criteria, escalation owners and operating responsibility with the people accountable for the work before anything runs on live records. Treat the first run as a rehearsal on test records: compare the draft output with an independently prepared answer, and record what the people reviewing it change.
Questions about this example
- What does Campfire use AI for in month-end accounting?
- Campfire uses Anthropic Claude Sonnet within Ember for recurring accrual analysis, financial queries, calculations run through code and integration with the close checklist. The March 2026 update says Ember 1.4 shipped with sandboxed Python execution after earlier plain-text arithmetic errors.
- Who checks the AI's accounting work at Campfire?
- The source material does not say who reviews Ember's output or who holds posting authority. Nuvepro proposes that an accountant investigates differences and a controller approves judgments and the close, but that is a suggested design, not Campfire's reported practice.
- How strong is the evidence for this example?
- The evidence class is reported use: Campfire's March 2026 product update corroborates that Ember 1.4 actions shipped. Independent accounting error rates, test data representativeness and production posting authority are not established. Code execution improves arithmetic reliability but does not show that accounting assumptions are correct.
- What would an enterprise need to build something similar?
- Nuvepro's proposed design, not Campfire's, uses test ledgers with balances, periods and currencies, a real calculation runtime, and cases such as partial payments and duplicate invoices. Independent ledger reconciliation and controller-reviewed assumptions check results, because running code alone is not proof.
Shape this around your enterprise
Bring a similar workflow and the people responsible for it. Together, define the records and systems to use, the result your team will build or configure, and how its owners will check it. Nuvepro scopes the simulation, environment and preparation. Agree any existing-agent evaluation, implementation assistance and production responsibilities separately.
Sources and limitations
Curation review: . This is the date this source example was reviewed, separate from source publication dates.
- Independent accounting error rates, fixture representativeness and production posting authority are not established by the reported calculation capability.
More finance and accounting examples
- AI purchase request review: Anthropic exampleAnthropic
- AI startup finance model dashboards: Alex Lieberman exampleUnnamed startup (reported by Alex Lieberman)
- AI AP exception resolution: Varick pattern by Eyad KhraisVarick AP pattern (customer unnamed)
- AI billing discrepancy review: Telstra exampleTelstra