ClariFi · How the AI Works
AI that does the work. Numbers it can't make up.
ClariFi's AI drafts the close, flags anomalies and explains what moved. It never computes a figure and never writes to your books: code calculates every number from your ledger, every number is cited to its source, and your finance team reviews and approves before anything reaches you.
Six mechanisms, not promises
Anyone can say their AI is careful. These hold whether or not anyone remembers to be careful.
The AI never computes a number
Every amount and date comes from deterministic functions, not from a language model. The AI's only job is proposing inputs from your documents, as a draft, for a person to confirm.
Read-only by construction, enforced in CI
A static test in our build pipeline fails if any agent could write to your books. Not a policy someone has to remember: a check that blocks the release. Your books stay in QuickBooks, Xero or Zoho, and a bookkeeper makes every entry.
Nothing counts until a person approves it
Agent output lands as a draft and stays there until your finance team confirms it. Every run is wrapped in an audit record, so there is a trail of what was proposed, by what, and who approved it.
Every figure is checked against accounting identities
Twenty-two tie-out checks run whenever your numbers update, before anything reaches a report: debits equal credits, cash rolls forward, margins and the MRR bridge reconcile, AR aging sums. A trial balance that does not balance cannot be imported, so it can never reach your runway, board pack or investor update.
Every number carries its source
Figures in AI-generated answers come with clickable citations back to the exact place in your own data, and extraction that cannot be verified against a source is rejected rather than guessed.
Benchmarks are approved by a person, with sources
Industry benchmarks researched by AI are saved with their cited sources and used only after a person approves them. Until then, ClariFi uses its published reference tables.
Why a person signs off on every number
We believe the right way to use AI in finance is human-in-the-loop. AI drafts the close package and flags the anomalies to speed the work up; your finance team reviews and approves. You get AI's speed with a person accountable for every figure.
Four rules ClariFi keeps
- Deterministic. The same inputs always give the same answer, and every figure traces to its source.
- People decide. Nothing ClariFi or its AI proposes takes effect until a named person accepts it, and every acceptance is logged.
- AI explains. It puts the working into plain language and drafts the routine steps, and it never computes a figure or makes a call.
- Show the working. Every figure traces to its formula and its sources, with a citation you can click.
See the rules every agent follows on the AI agents page.
Frequently asked questions
Can ClariFi's AI hallucinate a number?
No figure in ClariFi comes from a language model. Deterministic code calculates every amount from your own books, and the AI only puts the results into words. In Ask ClariFi, the AI must cite the source of every figure it uses, each citation is checked against your actual data, and a citation that cannot be matched is dropped rather than shown, so a citation chip is proof the number came from your books.
How do I know ClariFi's AI answers are accurate?
Every number in an AI-generated answer carries a citation that links back to the exact source in your own data, so you can verify it in one click. Ask ClariFi answers only from your financials, never from general knowledge, and is read-only by design. ClariFi follows five trust principles (no hallucination, deterministic, transparent, traceable and auditable), and our experts review every deliverable before it reaches you.
Can ClariFi's AI write to my books?
No. ClariFi reads from QuickBooks Online, Xero and Zoho Books and never posts entries. A test in the build pipeline fails the release if any agent could write to your books, and a bookkeeper makes every entry. Agents you build over MCP can only propose changes inside ClariFi, which wait for a person to approve them.
Does ClariFi use AI agents, and is it safe?
Yes. Agentic AI aids the repetitive work: Close Copilot drafts the monthly close package, Cash Sentinel flags burn spikes, and the account mapping agent proposes how trial-balance accounts map to your KPIs. Each one stages a draft and a person approves it. Every run is logged, and the agents have read-only access to your financials.
Is ClariFi secure? Where is my financial data stored?
ClariFi is built on enterprise-grade security controls, with data encrypted in transit (TLS 1.3) and at rest (AES-256). Financial data is stored in US-based cloud regions with role-based access controls and full audit logs. We never share customer data with third parties or train public AI models on it.
Check our work on your own numbers
30 minutes with our team. Ask anything about how ClariFi's AI handles your books.
