AI agents for finance are software agents that each own a defined area of finance work, such as reconciling bank lines to their documents or preparing accruals, and carry it out on the company's books within limits the finance team sets. In Rever, each agent has a published scope: what starts its work, what it checks, who it hands off to, what it files and what stays with your team.
Rever runs fifteen of them in five work groups. Eve, Rever's agentic finance manager, coordinates them and is not one of the fifteen; What Rever is covers the platform.
A CFO reading a roster asks what each agent does and what stops it from doing more. The tables answer the first, and the controls the second.
The fifteen AI agents for finance, by work group
Recover value and protect profit
| Agent | What it owns | What it may do |
|---|---|---|
| Entitlement | Finds rebates, credits, discounts and commissions a contract says are due | Assembles the supported claim; submission and settlement go to your Inbox |
| Deduction | Tests short payments and deductions against invoices, agreements and delivery evidence | Prepares the next recovery step; disputes and write-offs go to your Inbox |
| Spend Guard | Checks for duplicate invoices and payments and price drift | Recommends corrections; every payment hold needs Inbox approval before it is placed |
| Revenue Assurance | Checks that delivered work and contract price changes are billed as agreed | Prepares correction requests and verifies the invoices Journal creates after approval |
Verify the transaction, understand the difference
| Agent | What it owns | What it may do |
|---|---|---|
| Matcher | Reconciles records across contracts, orders, invoices, payments and ledgers | Commits supported matches; does not amend source documents, pay or post corrections |
| Investigator | Traces an issue to its root cause and establishes the financial effect | Recommends a correction; cannot approve its own proposed resolution |
Keep the books and close moving
| Agent | What it owns | What it may do |
|---|---|---|
| Journal | Prepares supported journals, accruals, provisions, reversals and invoice drafts | Hands approved entries back as an import file your team loads; cannot approve its own entry |
| Close Orchestrator | Keeps close tasks, reconciliations, schedules and sign-offs moving | Routes sign-off to your Inbox; does not certify an incomplete close |
| Flux Analyst | Explains movements in revenue, costs, margins and balances | Proposes corrections; you accept any material residual in the Inbox |
Improve cash visibility and evidence
| Agent | What it owns | What it may do |
|---|---|---|
| Cash Pilot | Builds a 13-week cash forecast and prioritises collections | Recommends; a forecast is not payment authority |
| Inventory | Reviews stock existence, ownership, movements, cut-off and valuation | Raises differences; counts and adjustments go to your Inbox |
| Evidence | Keeps original records, calculations and source references connected to the work | Assembles evidence packs; external requests follow your communication policy |
Keep the agent team accountable
| Agent | What it owns | What it may do |
|---|---|---|
| Sentinel | Detects unusual changes and potential control abuse, such as vendor-bank changes | Proposes safeguards; no setting lets an agent place a payment hold on its own |
| Steward | Monitors stale data, failed syncs and stalled tasks, and their financial impact | Runs routine permitted retries; repair ownership and access go to your Inbox |
| Auditor | Tests whether agents followed approved procedures and whether controls worked | Reports gaps and untested areas; the authorised owner remediates, and Auditor retests |
Read down the last column and you can see where each agent stops: the agents prepare and recommend, a person approves, and only approved work leaves Rever, for your team to apply.
AI agents compared with RPA and chatbots
Many finance teams already run RPA bots, which replay recorded clicks, and chatbots, which answer what someone types. AI agents in finance each own an area of work and follow an exception to its evidence.
| RPA bot | Chatbot | Rever agent | |
|---|---|---|---|
| Works from | A recorded script of steps | The prompt a person types | A published scope and the SOP your team approved |
| Meets an exception | Stops when an input changes shape | Answers if someone asks about it | Investigator traces it to its root cause and raises a finding |
| Shows its evidence | A log of the steps it ran | Sources, where the product cites them | Each number drills down to its finding, workpaper and record |
| Can post | Yes, if its login allows it | Only through a connection someone builds | No: approved entries come back as an import file your team loads |
How an agent earns its role
Each agent follows SOPs your team approved. Eve drafts each one, the process owner confirms it before use, and a change publishes only after approval. You set the thresholds each agent works within, such as Matcher's value and date tolerances or Journal's review bands; shared materiality stays in Settings.
A new agent starts as an observer. Eve reviews its performance against the approved SOP and promotes or demotes it, and you can read the record.
The controls under every agent
Rever's floors apply to every agent, in every plan, including Free:
- Rever never writes to your systems, and nothing leaves Rever without a recorded approval from an authorised approver.
- The identity that prepares an action cannot approve it.
- Material amounts need dual approval, at thresholds you set.
- Every decision is recorded with its inputs, source records and approver.
Connections are read-only, through direct, deterministic integrations rather than an MCP server or an agent holding credentials; Rever's trust page lists the rest of the controls.
Eve holds the same line. She cannot approve her own proposals, create invoices, post journals, move money or bypass the agent responsible for an action.
A sixteenth agent
Your team carries routines no product ships. Agent Studio, in early access on Growth Pro and Enterprise, turns one into an agent: you describe the need, Eve interviews your finance users, records the SOP and names the lead and supporting agents, and the new agent starts as an observer under the same controls.
See the work on your own records before you decide: the free Proof of Value runs the zero-day scan on a limited period of your data and walks you through the supported findings. The specialist agents come with a Growth plan.