Works the exceptions.
An exception is an item that does not match, such as a cash difference between your books and the bank. The agent finds the likely cause and proposes the fix, with the records it used.
For fund administrators and finance operations teams
Stop fixing the same exception twice.
The agent works your exceptions. Your reviewer corrects it once, and the correction becomes a tested rule that a supervisor approves. Each month you see what the agent handled, what your team fixed, and the time saved.
What Lumiveo does
01
02
Learns from your corrections.
When a reviewer corrects the agent, Lumiveo writes the correction as a rule in plain language, tests it on past cases, and applies it after a supervisor approves.
03
Records every decision.
For each item: what the agent proposed, what data it used, which rule applied, who approved it and when. Your auditors can export it.
04
Shows what it saved.
A monthly report per client fund: items handled, share accepted with no edit, repeat mistakes, and review time per item.
What you see each month →One correction, from start to finish
Example with made-up data
| Step | What happened |
|---|---|
| Exception | Fund A, 30 Sep. Cash at the broker is $41,250 lower than in the fund’s books. |
| Agent proposed | Possible missing payment. Open a query with the broker. |
| Reviewer corrected | Not an error. This broker posts dividends one business day after the pay date. The amount equals the dividend due on 30 Sep. |
| Rule written | Fund A, this broker: when a cash shortfall equals a dividend due on the last business day, treat it as a timing difference. Close it when the cash arrives. Raise it if the cash has not arrived after two business days. |
| Tested on past cases | Checked against 212 past exceptions for Fund A. The rule covers 9 of them and agrees with the reviewer’s decision in all 9. |
| Approved | Approved by the supervisor on 1 Oct. |
| Applied since | Used on 14 later cases. None reversed by a reviewer. |
Why AI tools in operations stop getting better
Every client has different rules.
Fees, valuation, expenses and report formats differ from one fund to the next. No AI model knows them in advance.
The rules are not written down.
They are in your accountants’ heads and personal notes. When someone leaves, the knowledge leaves too.
Corrections are thrown away.
A reviewer fixes the agent’s answer, and the next day the agent makes the same mistake again.
Your firm’s intelligence, compounding
Each correction your team makes becomes a rule the agent uses on every later case. Every month there are more approved rules and fewer repeat mistakes. That growing body of rules is your firm’s operating intelligence.
Written down.
Every approved rule is a short instruction in plain language, for one client fund, that your team can read and edit.
Yours.
The rules belong to you. One client’s rules are never used for another client.
It stays when people leave.
The knowledge in your accountants’ heads becomes rules the agent follows, with a record of who approved each one and when.
You stay in control
- No rule that changes an answer is used until a person approves it.
- You decide which items the agent may close alone and which need a person.
- Any rule can be edited or switched off at any time, and the record shows every case where it was used.
- One client’s rules are never used for another client.
What exists today
Live today
The method, in our research agent.
We first built this for quant research. Our agent, Edge, reads a research paper, writes the code, runs the backtest, and records every assumption as a decision a researcher can correct.
Open the Edge demoIn progress
The version for operations exceptions.
We are taking a small number of design partners. A pilot covers one task and one team for four to six weeks, at a fixed price.
Ask about a pilotDoes your agent repeat mistakes your team has already corrected?
Tell us the task and how your team reviews it today. We will tell you whether a pilot makes sense.