Level 0
Record
Stores decisions people have already made, and reports them back.
A clerk decides what to order and enters the purchase order. The system holds it, prints it and reports on it.
The ladder grades one thing: how much of a resource decision a system carries on its own, from holding a record of what people already decided to trading one resource against another inside a governed risk envelope.
A level says how much of the decision, the action and the consequence a system holds without a person in the path. It says nothing about how the system was built, how recently, or what it is called. Two systems a decade apart can sit on the same rung, and one product can sit on different rungs in different parts of a business.
| Level | Name | What the system does | Who carries the consequence |
|---|---|---|---|
| 0 | Record | Stores decisions people have already made, and reports them back. | A person |
| 1 | Advise | Computes a recommendation and presents it. A person decides and enters the result. | A person |
| 2 | Execute on instruction | Performs the steps once a person has decided. Deterministic, scripted, no judgement of its own. | A person |
| 3 | Act within bounds | Decides routine cases inside explicit thresholds, executes them, records what it did, and escalates everything else. | Shared, inside written bounds |
| 4 | Own the outcome | Plans continuously, acts, reconciles the result against the world, detects its own failures and corrects them. | The system |
| 5 | Allocate across resources | Trades one resource against another toward stated objectives, inside a governed risk envelope. | The system |
The break sits between level 2 and level 3. Up to level 2 a person has decided and the system has carried the decision out, however elaborate the carrying out has become. From level 3 the system decides some cases itself, which is the first point at which it can be wrong about something no person chose. Bounds, audit and escalation all begin at that line, with act within bounds.
Level 0
Stores decisions people have already made, and reports them back.
A clerk decides what to order and enters the purchase order. The system holds it, prints it and reports on it.
Level 1
Computes a recommendation and presents it. A person decides and enters the result.
A reorder report proposes quantities every Monday. Somebody reads it, changes two lines, and keys the orders.
Level 2
Performs the steps once a person has decided. Deterministic, scripted, no judgement of its own.
An approved requisition becomes a purchase order, a notification and a commitment without anyone retyping it. Nothing was decided by the system.
Level 3
Decides routine cases inside explicit thresholds, executes them, records what it did, and escalates everything else.
Shortages under a value limit, with an approved supplier and a lead time inside the window, are ordered by the system. Anything outside those bounds is raised to a buyer.
Level 4
Plans continuously, acts, reconciles the result against the world, detects its own failures and corrects them.
The system holds the service level for a category: it orders, watches fulfilment, notices a late shipment, re-plans around it, and reports the exception it could not absorb.
Level 5
Trades one resource against another toward stated objectives, inside a governed risk envelope.
Cash, inventory cover and expedited freight are traded against each other to hold a delivery promise, because a human set the objective and the tolerance rather than the action.
The levels are easy to read and easy to misuse. Two rules do more work than the six rungs, and a reading that ignores either one produces a number that feels precise and describes nothing.
A level is a property of a system and one resource class together. Cash, inventory, capacity, labour, suppliers and the rest are planned by different mechanisms, on different clocks, under different tolerances, and they reach different levels in the same building. A business whose cash application runs at level 4 while its capacity planning sits at level 0 is the ordinary case, not a pathology.
That gap is the useful output of the exercise. It names, in one line, the place where the next increment of autonomy is worth paying for and the places where it is not. A single figure averaged across a company hides exactly the thing worth knowing, which is why a company-level score is the one number this model refuses to produce.
The reasoning is short enough to check. A system reaches level 3 only if it decides routine cases itself, inside thresholds somebody wrote down in advance, records what it did under its own authority, and raises the rest. Most of what is offered today does something adjacent to that: it drafts a document a person approves, it answers questions about data already recorded, or it assembles an action and queues it for somebody to release.
Each of those leaves both the decision and the consequence with the person, which is level 1 or level 2 by the definitions above. The interface is new and the boundary is where it has been for thirty years. This is a claim about where the line falls, not a verdict on any particular system, and it is meant to be checked rather than believed. The ARP test is the check: eight questions, three of them load bearing, answerable about a system you can actually see.
A ladder invites climbing, and this one is not built for it. The right level for a function is set by what a wrong decision costs, not by how far up the rungs go. Three limits follow.
They are an ordering of how much a system carries, not a sequence anyone is obliged to walk. Nothing here says a function at level 1 should next be at level 2, or that a business at level 3 is behind. Most of the value in the model is in reading a business accurately at one point in time, which is a diagnostic act, not a plan.
The right level for a function is set by what happens when the system is wrong and how long the error stays invisible. Where a mistake is expensive, slow to surface, or hard to reverse, keeping the decision with a person is a sound engineering choice rather than a shortfall of ambition. Level 5 over a resource whose misallocation cannot be undone is a worse system than level 2 over the same resource, and calling it more mature does not make it safer.
Judgement that turns on facts the system cannot observe belongs to people, and no amount of instrumentation moves it. A decision that rests on a conversation with a supplier, on a commitment made outside the record, or on a position the business has to take in its own name is one where the best available design computes the options well and then stops. For that work, advising is the end state and not a stage on the way to somewhere else.
A level claimed about a system nobody can demonstrate is a level somebody assumed.
Take the ARP test