Risk AI puts the right checks and approvals around your finance work. Payments and entries go to the right person by amount and nobody approves their own request, each person and the AI see only the companies they should, and every action is written to an audit log that cannot be edited. It also finds conflicting permissions, checks expense claims against policy and runs regular access reviews.
What does Risk AI do?
Risk AI is the control layer of the finance department. It decides who may do what, makes sure the right person says yes, catches mistakes before they reach your reports, and keeps the record that proves all of it. The same rules apply to people, to API keys and to the AI.
| Area | How it works in Naarvin |
|---|---|
| Approvals | Amount tiers per action decide who approves what, and the route is fixed when the request is raised. Nobody approves their own request, and each approver decides once. |
| Controls | Data quality checks on the loaded books, a daily review of recent entries for anomalies, and the automatic checks behind each month-end sign-off |
| Access | Roles are granted per company. Each person, and the AI answering them, sees only the companies they should. |
| Audit trail | Every action is written to a log that only ever grows, where each entry carries the fingerprint of the one before it |
| Segregation of duties | Finds people who hold two permissions one person should not hold together, and people who did both halves of a sensitive pair |
| Expenses and custody | Checks claims against your policy and has them approved by someone other than the claimant; tracks custody advances and petty cash |
| Access reviews | Regular reviews of who can see and approve what, where a reviewer keeps, suspends or removes each person |
The actions in the authority matrix include payment release, budget change, receivable write-off, journal posting, capital expenditure, asset disposal and access requests. Sensitive actions ask for a fresh verification, even inside an active session. Our guide to approval matrices and segregation of duties explains how to design both, including in a small team.
The segregation of duties report has two views. By right, it lists people who hold two permissions one person should not hold together for the same companies. By action, it reads the audit log for cases where one person actually did both halves of a sensitive pair: changing a supplier's bank details and then generating a payment file, drafting a journal and then confirming it posted, or importing the numbers and then signing off the close. An access review starts from a snapshot of every member's roles, companies and last sign-in, and it completes only when every person has been decided.
How does a month run with Risk AI?
Some controls run every day, some every month, and some on the cycle your auditors expect. A typical month looks like this.
- Every request is routedA payment, journal or budget change is raised, and its approval route is fixed by amount at that moment. The approver is notified, in Slack or Teams too if you use them, with no amount in the notice, and decides in Naarvin.
- Every morning, the anomaly reviewRecent journal entries are reviewed for duplicate payments, round-amount and weekend manual journals, amounts unusual for the account and postings into a period already reported. Every finding lists the records behind it.
- Every sync, the data checksData quality checks show whether each company's books loaded on time, balance month by month and have their profit and loss accounts mapped, before anyone reports on them.
- Through the month, expensesClaims are checked against your policy when they are submitted. A missing receipt the policy requires stops the claim; other flags go to the approver, who cannot be the claimant.
- At month end, the sign-offThe close is signed off only when its automatic checks pass. Postings made into a signed-off month afterwards are flagged.
- Periodically, duties and accessThe segregation of duties report and an access review are run on the cycle you choose, for example each quarter. Every decision is written to the audit log.
Who is Risk AI for?
Risk AI is for the CFO who answers to the board for financial control, the controllers and finance managers who run the approval routes, and internal audit or risk teams who test that controls work. It also serves the administrators who grant access, and your external auditors, who can read the segregation of duties report and the audit log when they test controls.
It matters most where a few people do a lot: a small finance team where the same person might raise and approve, or a group where a manager should see one country and not the others. Every company in a group follows the same rules, and each company's managers see only their own.
What stays with your team?
Risk AI enforces the rules; people set them and make the decisions inside them.
- Setting the authority matrix: which actions need approval, the amount tiers and who sits in each.
- Every approval. An API key can never approve, sign off or complete a four-eyes review: those always need a person.
- Accepting a conflict. When a role would give one person two duties that should be separated and the pair is high or critical risk, the change is refused until an administrator records why it is acceptable and how it is mitigated. The pair still shows in the report.
- Deciding each person in an access review. Nobody reviews themselves.
- Controls outside Naarvin. These checks cover approvals and access inside Naarvin; duties inside your ERP and your bank portal still need their own review.
How does Risk AI keep data and the AI in scope?
Each organization's records are separated by rules inside the database itself, not only in application code. The CFO Copilot answers inside the scope of the person asking, so it cannot reach a company that person cannot open. The AI never receives ERP passwords, API secrets or access tokens, and it has no open database access: it can only call finance tools that check permissions first.
The audit log can be exported with every fingerprint, so a changed or missing entry shows as a broken chain. Sign-in can require a passkey or an authenticator code, and people can sign in through your identity provider. Details are on the security page.
Which systems does Risk AI work with?
Single sign-on works over OpenID Connect (OIDC) or SAML 2.0 with signed assertions, and SCIM 2.0 can add, update and deactivate people from your identity provider. Approval notices go to Slack and Microsoft Teams without amounts, and your own systems can receive signed events through webhooks. The controls read the books Naarvin syncs from each company's ERP, such as ERPNext or Odoo. Plans are on request; see pricing or book a demo.
Questions people ask
Can an API key or the AI approve a payment?
No. An API key acts as a service account with one role and chosen companies, and it can never approve, sign off or complete a four-eyes review. The AI prepares work; a person with the authority decides.
Can the Copilot show a manager figures for a company they cannot see?
No. The Copilot answers inside the scope of the person asking, so it cannot reach a company that person cannot open.
Our team is small. What if one person needs two conflicting roles?
You can allow it. For a high or critical risk pair, an administrator records why it is acceptable and how it is mitigated; that is kept with the assignment and in the audit trail, and the pair still shows in the segregation of duties report.
Can the audit log be changed?
Entries are only ever added. Each one carries the fingerprint of the entry before it, so a changed or missing entry breaks the chain, and the log can be exported with every fingerprint.
Does Naarvin hold a security certification such as SOC 2 or ISO 27001?
Not today. An independent certification is on the roadmap, and hosting inside the Kingdom is not available yet. If your security team has a questionnaire, we answer it item by item.