Autonomous Resource Planning

Autonomous Resource Planning is an architecture in which the system that holds the state of an enterprise's resources is also the system that decides and acts on them.

Definition · Maturity levels · Governance · The ARP test

Fig. 1 · The bounded loopThe definition, drawn
GRANTED AUTHORITYObserve: Take in transactions, events, resource states, constraints and outside signals as they happen.ObserveInterpret: Work out what those signals mean for this business: its policies, its history, its commitments.InterpretPlan: Decide what should happen next, given dependencies, policies, objectives and constraints.PlanAct: Execute the permitted actions, across every resource the plan touches.ActReconcile: Confirm the intended outcome actually occurred, record the consequence, and prove the record against the world outside the system.ReconcileAdapt: Re-plan as conditions move, learn from what the reconciliation showed, and escalate what needs judgement.AdaptTURNS UNASKEDWRITESREADSLedgerHOLDSMoney: cash, receivables, payables, the ledger, the close, treasuryMoneyMaterials: inventory, procurement, suppliers, logisticsMaterialsPeople: labour, roles, assignment, scheduling, workforce planningPeopleCapacity: equipment, facilities, production capacity, throughputCapacityTime: lead times, commitments, calendars, sequencingTimeInformation: master data, documents, contracts, the enterprise record itselfInformationAgents: compute, licences, permissions, authority budgets, escalation capacityAgentsNEWESCALATES WHAT IT CANNOT HOLD

Six stages turning without being asked, inside an authority someone granted, writing to and reading from the one ledger that holds the seven classes of resource. The only line that leaves is the exception.

Autonomous Resource Planning is an architecture in which the system that holds the state of an enterprise's resources is also the system that decides and acts on them. It observes resource state and external signals continuously, forms plans against stated objectives and constraints, executes actions within explicitly granted authority, records the consequence in the same ledger it read from, confirms the intended outcome occurred, and escalates what falls outside its authority or its confidence.

What it excludes

Each clause is there to rule something out. A system that proposes and waits is ruled out by decides and acts. A system that acts into a queue somebody rekeys is ruled out by the same ledger it read from. A system that acts and never checks is ruled out by confirms the intended outcome occurred. A system that acts past what it was given is ruled out by explicitly granted authority.

The definition is written to be failed. If every clause held of every product on the market, it would be describing the market, not a category inside it.

The definition, clause by clause

Three things at once

The three hold together or not at all. A system with any two of them is a familiar thing behind a new interface, and each of the three pairs has been sold for decades.

  • It plans without being asked

    The work begins because something in the world moved: a receipt, a bank line, a revised forecast, a delivery that did not arrive. A system that waits to be opened leaves the hardest judgement, which is when the decision was needed, with whoever remembered to open it.

  • It commits rather than recommends

    The decision lands in the record the system already holds, under an authority somebody granted in writing, with its reversal recorded beside it. A recommendation handed to a person to key in has moved the typing and nothing else.

  • It answers for the outcome

    Acting is not the end of the sequence. The system confirms that the result it intended actually occurred, posts the consequence, and raises what it could not absorb, instead of reporting success from its own intent.

Every era automated the record and left the judgement

Each of these was the right answer to the problem in front of it, and none of them was a mistake. Read down the last column instead: what was left to people barely moves in sixty years, which is the whole argument for a new name.

  1. 1960s to 1970sMaterial requirements planningWhether the master schedule was achievable, what to do when it was not, and every decision about capacity, labour and money that the material plan implied.
  2. 1980sManufacturing resource planningClosing the loop. People decided whether the plan was feasible, negotiated the trade-offs between the plant, the sales promise and the cash position, and re-entered the answer.
  3. 1990sEnterprise resource planningEvery decision. The suite held an accurate, shared account of the enterprise, and people read it, judged it and keyed the consequence back in.
  4. Early to mid 1990sClient/server deliveryThe same decisions as before, now in front of more people in more companies, most of whom had never had an integrated system to argue with.
  5. 2000s onwardCloud ERPEvery workflow inside it. The system was reachable from anywhere and updated without a project, and a person still decided what to order, what to pay and what to promise.
  6. 2010sComposable and best of breedThe joins. People carried context between systems, reconciled the versions of a record that had drifted apart, and owned the integrations when they broke.
  7. 2023 onwardAssistive AIAuthority. The software proposes and a person accepts, which means the decision, the timing and the consequence all remain where they were before.
  8. NowAutonomous resource planningObjectives, constraints, the grants themselves, and everything the system stops on: ambiguity, low confidence and anything outside the bounds it was given.

What each era automated, and why it was right at the time

The loop that has to close

No stage below is new on its own. Reporting observed, analytics interpreted, planning engines planned and workflow acted, decades before any of this was called autonomous. What is new is one system running all six without a person carrying the result between them.

The loop: Observe, Interpret, Plan, Act, Reconcile, AdaptSix stages drawn as a closed cycle, each one feeding the next and the last feeding the first, so the output of reconciling is what adapting works from.Fig. 2 · The loopSix stages, one cycleObserve: Take in transactions, events, resource states, constraints and outside signals as they happen.01ObserveInterpret: Work out what those signals mean for this business: its policies, its history, its commitments.02InterpretPlan: Decide what should happen next, given dependencies, policies, objectives and constraints.03PlanAct: Execute the permitted actions, across every resource the plan touches.04ActReconcile: Confirm the intended outcome actually occurred, record the consequence, and prove the record against the world outside the system.05ReconcileAdapt: Re-plan as conditions move, learn from what the reconciliation showed, and escalate what needs judgement.06AdaptSix stages run continuously, each feeding the next, and adapt feeding observe.
  1. 01 ObserveTake in transactions, events, resource states, constraints and outside signals as they happen.
  2. 02 InterpretWork out what those signals mean for this business: its policies, its history, its commitments.
  3. 03 PlanDecide what should happen next, given dependencies, policies, objectives and constraints.
  4. 04 ActExecute the permitted actions, across every resource the plan touches.
  5. 05 ReconcileConfirm the intended outcome actually occurred, record the consequence, and prove the record against the world outside the system.
  6. 06 AdaptRe-plan as conditions move, learn from what the reconciliation showed, and escalate what needs judgement.

The break is nearly always in the same two places: between acting and reconciling, where a system asserts an outcome it never proved, and between reconciling and adapting, where it records what went wrong and changes nothing. The loop, stage by stage

Six levels, and the one where consequence moves

The ladder grades one thing: how much of a resource decision a system carries on its own. The break sits between level 2 and level 3, the first point at which the system can be wrong about something no person chose.

LevelNameWho carries the consequence
0RecordA person
1AdviseA person
2Execute on instructionA person
3Act within boundsShared, inside written bounds
4Own the outcomeThe system
5Allocate across resourcesThe system

One rule makes the ladder useful, and its absence makes it decorative: grade per resource class, never per company. Cash, inventory, capacity and labour are planned by different mechanisms on different clocks, and they reach different levels in the same building. A single company figure hides the one thing worth knowing, which is where the next increment of autonomy is worth paying for. The maturity model

Seven resource classes

A resource planning system is read class by class, because that is how it is built and how it fails. Six of these have been planned by software since the 1960s. The seventh arrives with the autonomy itself.

  • Money

    cash, receivables, payables, the ledger, the close, treasury

  • Materials

    inventory, procurement, suppliers, logistics

  • People

    labour, roles, assignment, scheduling, workforce planning

  • Capacity

    equipment, facilities, production capacity, throughput

  • Time

    lead times, commitments, calendars, sequencing

  • Information

    master data, documents, contracts, the enterprise record itself

  • Agents

    New

    compute, licences, permissions, authority budgets, escalation capacity

No previous era planned the seventh, because no previous era gave software authority of its own. Once software holds authority, that authority is finite, costed and allocable, with its own utilisation and its own segregation of duties. An autonomous system therefore has to plan the thing doing the planning, which is the one structural addition this category makes to a model that has otherwise held since manufacturing resource planning.

Eight questions, three of which decide it

A definition is worth as much as the test that applies it. Each of these eight is answerable about a system in production, and each has an answer that sounds like a pass and is not.

  1. 01Does the system start the work itself, or does a person start it every time?
  2. 02Can you print the list of actions the system may take on its own, with the limits on each?Load bearing
  3. 03Does the action land in the system of record, or in a queue somebody rekeys?Load bearing
  4. 04Does the system notice its own failures before a person does?
  5. 05For one action taken last quarter, can you recover the inputs, the policy, the alternatives and the evidence?Load bearing
  6. 06When the system is wrong, is there a recorded reversal, or a person cleaning up by hand?
  7. 07When conditions move, does the plan move with them without being asked again?
  8. 08Does the system stop when it is outside its bounds or below its confidence, and hand over with its evidence?

A system can miss any of the other five and still be doing this, less completely. Miss one of the three marked and it is not: without enumerated authority nobody can say what it was permitted to do, without closure nothing it decided actually happened, and without reconstruction no single decision it took can be reviewed afterwards. The test in full, with the passing and the weak answers

Where this argument is weak

The term is already in public use by other people, and the claim made here is depth rather than priority. What is offered here is a definition specific enough to fail a product against, not a claim to have coined anything. If the name settles somewhere else, the six levels, the seven classes and the eight questions survive the renaming, because none of them depends on the label.

The levels are easy to self-assess generously. Nothing in a ladder stops a team reading its own approval queue as bounded authority, or its nightly rerun as continuous planning. The test exists because the model on its own is too comfortable, and even the test is usually answered by the people with an interest in the answer.

Almost nothing in production today reaches level 4 in any resource class. That makes this a description of a direction and not of a market, and a reader is entitled to discount it accordingly. The honest claim is narrower: the distance between what is sold and what is defined here is the interesting quantity, and naming it precisely is worth more than predicting when it closes.

Autonomy raises accountability questions that are not settled anywhere. Who answers for a decision no person made, what an auditor will accept as evidence of control, and how segregation of duties survives when one system holds both sides of it stay open, and are marked as open, not answered. The objections, including the unresolved ones