Egisedge AI Agents for Finance (in development) on Oracle ERP

AI agents are software assistants that can carry out defined steps in a business process, check their work, and hand over to a person when a decision is needed. Egisedge is developing AI Agents for Finance, designed to work with Oracle ERP and the finance team's existing controls.

This video is a preview of work in development. We show the design principles and walk through example scenarios we are building for: supplier invoice exceptions, bank reconciliation follow-up, and month-end close checklists. We explain how each agent reads data, proposes actions, and waits for approval before anything changes in the ledger.

Features shown may change before release. The demo uses sample data only. If you want to take part in early conversations about your use cases, get in touch.
Learn more: https://egisedge.com

#AIAgents #FinanceTransformation #OracleERP

Transcript

[Scene: Presenter on camera. On-screen label, top right, for the whole video: "Preview – in development".]

PRESENTER: Egisedge is developing AI Agents for Finance. This video is a preview. What you see is work in progress, using sample data. Features may change before release. We want to show you how we are thinking about it, and the kind of finance work we are building for.

[Scene: Chapter card, "What we mean by an AI agent".]

PRESENTER: First, what do we mean by an agent? A chatbot answers questions. An agent carries out defined steps in a process. It reads data, decides what the next step should be, proposes an action, and checks the result. When a decision needs a person, it stops and asks.

[Scene: Simple loop graphic: Read, Propose, Approve, Act, Check. The "Approve" step is highlighted with a person icon.]

[Scene: Chapter card, "Design principles".]

PRESENTER: We are building to four principles. One. The agent works inside your existing controls. It uses a role with limited access, like any other user. Two. It proposes; a person approves anything that changes the ledger or leaves the organisation. Three. Every step is logged, so reviewers can see what the agent looked at and why it suggested what it did. Four. It works with your ERP. Our first focus is Oracle ERP.

[Scene: Four-tile graphic: "Your controls", "Human approval", "Full log", "Works with Oracle ERP".]

[Scene: Chapter card, "Scenario: invoice exceptions".]

PRESENTER: Let's look at the first scenario we are building for. Supplier invoices that are on hold. Every payables team has a list of them. Price differences, quantity differences, missing receipts.

[Scene: Screen recording of the preview console. A list of sample invoices on hold, grouped by hold reason.]

PRESENTER: The agent reads the hold list. For each invoice, it gathers the related purchase order and receipt. It then sorts them. This one has a receipt that simply has not been entered yet. This one has a price difference within the tolerance your policy allows. This one needs the buyer to confirm a new price.

[Scene: Click one invoice. Side panel shows the agent's summary, the documents it checked, and a proposed action: "Email buyer for price confirmation". Buttons: Approve, Edit, Reject.]

PRESENTER: For each, the agent proposes the next action. The user approves, edits or rejects. Nothing is released without that approval.

[Scene: Chapter card, "Scenario: bank reconciliation follow-up".]

PRESENTER: Second scenario. Unreconciled bank lines. After automatic matching, there are always lines left over.

[Scene: Preview console. List of unreconciled sample bank lines.]

PRESENTER: The agent looks at each leftover line, searches for likely matches by amount, date and reference, and explains its suggestion in plain language. Where it cannot find a match, it drafts a query to the right person, such as the treasury or receivables team.

[Scene: Show a suggestion card: "Likely match: customer receipt, same amount, reference differs by one character". Then a drafted internal query.]

[Scene: Chapter card, "Scenario: close checklist".]

PRESENTER: Third scenario. The month-end close checklist. The agent tracks the checklist, checks which tasks are done in the system, and reminds owners of tasks that are late. It prepares a short status note for the controller each morning of close.

[Scene: Close checklist view with task status and a drafted status note.]

PRESENTER: The controller still owns the close. The agent saves the chasing.

[Scene: Chapter card, "What is next". Presenter on camera.]

PRESENTER: What is next? We are refining these scenarios and testing them with sample data. We are also developing AI Agents for HCM and Consol360, an AI consolidation product. If your finance team has a process you would like an agent to help with, we would like to hear about it.

[Scene: Chapter card, "Recap".]

PRESENTER: To recap. An agent carries out defined steps and asks a person when a decision is needed. Ours are being built to work inside your controls, with human approval and a full log. Early scenarios are invoice exceptions, bank reconciliation follow-up and the close checklist. This is a preview, and features may change. To talk to us about your use cases, visit egisedge dot com.

[Scene: End card with egisedge.com. Label: "Preview – in development".]