Skip to tool
MM/FINIMO
MM/FINIMOAI in FP&A, the controlled way

Your finance agent

A double with unlimited time, and a leash

It is 06:00 on WD1 of Helix's September 2026 close. Watch what a well-designed finance agent does before anyone is in, and what it is not allowed to do.

Helix Consumer Goods is a fictitious company. Every figure is synthetic.

WD1 · 1 Oct 2026 · September close

Helix finance agent, morning run

06:00Scheduled
Lines checked
Material
Data issues
Drafts waiting
Approvals done
Blocked by permissions

About five seconds. Nothing is sent, posted or stored.

The run

Eight steps before anyone is in

The same checklist every WD1. Checks come before words, and the run stops at the gate.

    The gate

    Three drafts and one data issue

    Nothing here has left finance. You are the reviewer: approve or reject each draft. The status and the control log update.

    The queue fills when the run reaches step 5.

    The leash

    Six permissions, set before the first run

    Each one is locked. Try to switch one. These six are a practitioner's starting point, not a standard.

    Permission set v1 · Helix WD1 agentOwner: Controller

      Control log

      Every output, its source and its approver

      Written by the agent at 06:40. Each decision is added with the approver's role and the time.

      Output idSourceRuleClassStatusApprover roleTime

      Replicate it

      Write the specification before the first run

      One copy gives you the template your team fills in, the two rules exactly as this run applies them, and a controlled system prompt. Plain text, no tool required.

      1 · The specificationInputs, checks in order, what it may do, what it may never do, stop points, what every output carries, escalation.
      2 · The rulesData issue first, then 5% or 100,000 EUR. As a rule list and as one spreadsheet formula.
      3 · The system promptReads, checks and drafts. Cites the source row. Stops at the gate for a named role.
      
          

      The lesson

      A good agent asks. A person decides.

      1. Give it read access, a checklist and a human gate. Read-only keeps it out of the ledger. The checklist makes it run the same way every morning. The gate makes a person own what leaves.
      2. Its best output is a question, not an answer. At 06:38 nobody knows why Digital Commerce is 258,767 EUR short. The agent cannot find out. It can draft the question to the person who knows, before they are in.
      3. Every output carries its source and its approver. A line that cannot point to a row and a person does not go in the pack. That is what makes the work auditable, and the review short.

      Where to start

      Write your agent's job description in 10 lines: what it reads, what it drafts, what it may never do. If a line is hard to write, the agent is not ready to run.

      Free series

      AI in FP&A, the controlled way

      Short, practical tools that show finance teams where AI really helps, without losing lineage, approvals or the controller's trust. Follow me on LinkedIn for the next one.

      Not sure where to start? Take the 10-question AI Reality Check