Multi-Approver Routing & Escalation Engine

The problem this kills

You sent the approval three days ago. It's sitting unread in someone's inbox, two layers down from a meeting invite, and the project is stalled. Nobody knows whose desk it's on. Nobody escalated it because nobody was tracking the clock. When it finally gets approved, there's no record of who looked at it, when, or why the rejection happened.

Approvals that live in email and chat threads are invisible, un-trackable, and slow. When an approver is on vacation, the whole chain freezes. When someone forgets, nothing happens - there's no SLA, no nudge, no backup. And when an auditor or your boss asks "who approved this and when," you're scrolling through inboxes trying to reconstruct the story.

This plan kills all of that. You define the approval flow once - the steps, who approves each step, how long they have, and who picks it up if they go silent - and the engine does the routing, the reminding, and the escalating for you.

What you'll build

A private, reusable approval engine for your team:

• Flow definitions you create once and reuse: ordered or parallel steps, an approver (or set of approvers) per step, an SLA clock per step, and an escalation target for each.
• Item submission - drop in the things that need approving (from a CSV, a Sheet, or a simple form) and pick which flow they run through.
• Automatic routing - the engine moves each item from step to step, only advancing when the current step is satisfied.
• Reminders and escalation on the SLA clock - if an approver doesn't respond in time, the engine nudges them, then reassigns to the backup/manager. It escalates, it never auto-approves.
• A human decision at every step - each approver clicks Approve or Reject with an optional comment. Nothing is committed to "done" until the whole chain completes.
• A complete audit trail and a one-click CSV export of outcomes in the exact columns your system of record expects.

What's inside the Implementation Plan

The plan is a single markdown file you paste into Claude Code. It walks the AI through building the whole tool, step by step, with a ready-to-copy prompt at the end of each step.

Crucially, it opens by interviewing you about your business - your real approval steps, who approves what, your SLA windows, how you handle vacations and delegations, your naming conventions, and your messy exceptions. It reads a short tailored spec back to you and waits for your thumbs-up before building anything. You get an engine shaped to your approvals, not a generic template you have to bend to fit.

Inside you'll find: the discovery interview, the data model for flows / steps / items / decisions / audit events, the routing and SLA-escalation logic, the approver screens, the reminder and escalation emails, login and per-organization data isolation, duplicate guards, and the CSV import/export fallback so it's fully buildable today even if you have no API into your current system.

The governance it includes (this is the point)

This isn't a toy. Governance is built in from the first prompt:

• Login so only your team can use the tool.
• Row-level security so each organization only ever sees its own flows and items.
• A complete audit trail - every submission, decision, reminder, escalation, and reassignment is logged with who and when.
• A hard human-in-the-loop gate - approvers decide; the engine routes. Escalation reassigns the decision to a backup; it never approves on anyone's behalf.
• Duplicate guards - the same item can't be processed twice through the same step (deduped on item + approval step).

Who it's for

PMO leads, operations managers, finance and procurement approvers, compliance teams - anyone whose sign-offs currently die in an inbox. If you can describe your approval rules in plain English, you can build this. No prior coding required.

You've got this - paste the first prompt and let the agent interview you.