Governance and security
Governance of the budget, and governance of what the AI agent does in it.
Two layers, written before the AI agent touches your cycle. That is what your risk team needs to read.
We assess both governance layers of your cycle in 30 minutes, with no commitment.
The AI agent executes
within the user's permissions
The platform records
author, date and the originating conversation
You review and approve
only the approved version counts
The paradox
AI chat is already in your cycle, and nobody recorded anything.
Today your team pastes assumptions into the chat, the data leaves the company and the number comes back with no trace. Bringing AI inside solves this, on one condition: approving cell by cell wipes out the time savings, so the AI agent works within a scope, records everything it does, and no version counts until someone approves it.
Layer 1
Governance of the budgeting process
It applies to everyone who touches the budget, with or without the AI agent.
Assumptions with an owner
Every assumption has a name behind it, and any change goes through the person accountable for that number.
One version counts
Reports and KPIs read only the approved version. The others exist as working copies and do not muddle the decision.
Named sign-off
Each approval step is tied to a person, and whoever signed off shows up in the audit trail.
Allocation out of reach
Changing the allocation rule requires permission and leaves a trace, so nobody changes the distribution on their own.
Least-privilege access
Each person reaches only what their role requires, down to the assumption and version level.
Layer 2
Governance of the AI agent's actions
The AI agent writes to the budget, because approving entry by entry makes the cycle unworkable. Control lies in where it can write, in what gets recorded and in who approves the version that counts.
Scope
The AI agent inherits the permissions of the user who triggered it, and reaches nothing beyond them.
Logged execution
It writes, and every change goes into the full audit trail, with author, date and the conversation that originated it.
Official version
Nothing it does becomes an official number until a person approves the version.
Deterministic calculation
Once the assumption is recorded, the value comes from a formula. The same assumption always produces the same result.
Data and usage
Your budget stays in the contracted environment, does not train third-party models, and usage is capped.
The audit trail, in practice
What each line of the audit trail holds
Four types of record: the AI agent's write, the formula propagation, the human review and the approval of the official version. Each one with author and time, and none overwrites the previous one.
| What changed | Who | When |
|---|---|---|
| WriteCollective wage adjustment, from 4.2% to 4.5% conversation: 2027 payroll review | AI agent scope: payroll assumptions | Mar 14, 09:12 |
| PropagationPayroll, Sales and Operations, recalculated by formula, with no AI agent in the way | Platform deterministic | Mar 14, 09:15 |
| ReviewWage adjustment checked against the collective agreement comment attached to the number | Ana Prado CFO | Mar 14, 11:12 |
| OfficialVersion 2027.v3 approved as official from here on, the number counts | Ana Prado CFO | Mar 14, 11:40 |
Export as PDF or CSV
Requirements map
Which control answers which requirement
| Requirement | What BudgetXpert delivers |
|---|---|
| LGPD: data purpose and access | Per-object permissions and a written data use policy |
| Joint Resolution 18/2025: data traceability | Audit trail with author and date, exportable to PDF and CSV |
| CMN Resolution 4,557: operational and technology risk | Declared scope, execution log and a named official version |
| AI governance framework | Two separate layers: the process and the AI agent's actions |
The platform is designed to meet these requirements. Formal compliance belongs to the institution, and BudgetXpert delivers the controls and the evidence that support accountability.
Free assessment
Find out where your budget is exposed
A governance diagnostic of your budgeting cycle, prepared by the BudgetXpert team. Free of charge and with no purchase commitment.
You answer the assessment
Eight questions, two minutes, on the two control layers of your cycle.
You get the diagnostic right away
One page with the gaps found, the risk of each one and where to start, ready to download as a PDF.
If you like, book 30 minutes
With someone who knows the budgeting process, not a salesperson, to dig into the result.