Spare parts planning AI agent: a blueprint for materials management
How a governed AI agent workforce handles spare parts planning: it reviews demand, lead times, criticality and installed base for each item in SAP S/4HANA, IBM Maximo or Oracle Fusion, proposes stock policy with its reasoning, and executes only what materials management approves.
- Problem
- Planners set stock for tens of thousands of MRO items from memory, so critical parts run out while obsolete stock piles up.
- Agent design
- Installed-base, planning, execution and inventory health agents under materials management guardrails.
- Systems
- ERP materials and maintenance modules or the CMMS, storerooms, supplier and OEM data.
- Hard limits
- Agents never set criticality, approve supersessions, write off stock or create purchase orders.
- Value levers
- Fewer critical stockouts, less spares capital, fewer expedites, long-tail coverage.
- First release
- Proposal-only at one site, then automatic updates for non-critical items in a narrow band.
Where spare parts planning breaks today
A large maintenance store carries tens of thousands of active items, and most move only a few times a year. Reorder points and maximums are set when the item is created and rarely revisited. Planners spend their week on fast movers and late deliveries, so the long tail runs on stale settings. Forecasting built for steady demand handles intermittent spares badly: a part with three issues in five years looks like zero demand until a pump seal fails.
The data that should drive the decision is scattered. Criticality sits in a reliability spreadsheet or is set to high on half the catalog. Equipment BOMs are incomplete, so the planner cannot see which assets a part protects. Purchasing lead times are quotes, not what actually happens. Duplicate item numbers split demand, and parts for retired equipment stay on the shelf because nobody owns their removal. The result is expedites and downtime on critical items alongside a growing balance of obsolete stock.
| Planning decision | What triggers it | What the agent does | Can agents act alone? |
|---|---|---|---|
| Reorder point and max update | Demand, lead time or installed base has moved on a non-critical item | Recalculates from consumption, actual lead times and planned work; applies inside the band | Non-critical, inside band |
| Replenishment requisition | Stock on hand plus open orders falls below an approved reorder point | Creates the purchase requisition with the approved quantity, source and need date | Approved policy only |
| Inter-site rebalancing | Surplus at one storeroom, shortage or shutdown demand at another | Proposes a transfer with freight cost and residual risk at the sending site | Planner approves |
| Late delivery or expedite | Open order past its confirmed date while a work order reservation waits | Ranks late orders by the work they hold up; drafts the expedite for the buyer | Buyer acts |
| Supersession or interchangeability | OEM bulletin, supplier notice, or two item numbers for one part | Assembles evidence and affected BOMs for the reliability engineer | Never |
| Criticality change or insurance spare | New asset, changed failure history, or a zero-demand item protecting a critical asset | Flags the gap with downtime exposure and lead time; takes no position | Never |
| Obsolescence or disposal candidate | No issues in the review window, or no remaining installed base | Lists value, last issue and original assets, then routes it | Never |
The spare parts planning AI agent workforce: four roles, one approval line
The live design names a planning agent and an execution agent. This blueprint keeps that split and adds an agent that repairs the installed-base link and one that reports on inventory health. Proposing and executing stay with different agents under separate system identities, so no agent both sets a stock level and spends against it.
| Agent | Job | Reads | Can write | Never does |
|---|---|---|---|---|
| Installed-base agent | Links each item to the equipment it protects and finds BOM and criticality gaps | Equipment, equipment BOMs, work order history, parts issued to orders | Proposed BOM links and data-gap tasks for the master data team | Edit BOMs, item master or criticality fields |
| Planning agent | Proposes stock policy per item with its reasoning | Consumption, reservations from planned work, actual lead times, unit cost, guardrail policy | Policy proposal with method, inputs, service and capital effect | Apply a policy or create any transaction |
| Execution agent | Applies approved policy, raises requisitions and transfers, tracks late deliveries | Approved proposals, stock and open orders, confirmed delivery dates | Plant-level planning parameters, purchase and stock transport requisitions through standard functions | Convert requisitions to purchase orders, change suppliers or prices, or move stock held for a critical asset |
| Inventory health agent | Reports stockouts, excess and obsolescence by site, asset class and planner | Audit log, stock values, issues, proposals and overrides | Monthly inventory health report and the obsolescence review list | Act on any item, policy or order |
Architecture: spares agents beside the ERP and CMMS, writing only approved policy
The agents read item master, stock, maintenance and purchasing records through published interfaces and write through the functions a planner uses. Every write carries its proposal and guardrail version. A control plane outside the agents holds the materials-owned guardrail policy, approval workflow, audit log and a pause switch that freezes every policy while replenishment keeps running.
How one spare parts recommendation moves through the design
Two questions route every proposal. Does it touch criticality, a supersession, an insurance spare or a write-off? If so, a named person decides and the agents only assemble evidence. If not, does the change stay inside the band set for that item class, in both stock value and service level? Only a change that passes both is applied automatically, and a sample is reviewed weekly.
ERP, CMMS and supplier data integration for spare parts agents
Spares planning usually spans the system that holds stock and requisitions and the one that holds equipment and work orders. The installed-base link between them decides integration order, because a planning agent that cannot see what a part protects can only repeat history.
| Platform | Where spares planning lives | How agents read | How agents write | Design note |
|---|---|---|---|---|
| SAP S/4HANA (MM and Plant Maintenance) | Plant-level reorder point and maximum stock, equipment BOMs, reservations | OData APIs for product master, stock, reservations and maintenance orders | Planning parameters through the standard material change function; requisitions | If changes need a master data request, the agent raises the request |
| IBM Maximo | Storeroom inventory with reorder point and lead time; asset spare parts | REST API for items, inventory, assets and work order materials | Inventory planning fields through the integration framework | Existing reorder runs raise requisitions from agent values, so demand is never raised twice |
| Oracle Fusion Cloud (Inventory and Maintenance) | Min-max planning, maintenance work orders, assets | Inventory, maintenance and purchasing REST resources | Min-max values and requisitions through standard interfaces | Hold guardrails at the same level as the min-max plan |
| Supplier, OEM and reliability data | Confirmed dates, supersession notices, failure history, criticality | Order confirmations, OEM bulletins, criticality register exports | None; changes route to the buyer or reliability engineer | Plan on actual receipt lead times; reliability owns criticality |
The agents can be built with ERP-native agent tooling, Microsoft Copilot Studio, or a custom orchestration layer, chosen by where planners work and where identity and audit controls are strongest. An existing inventory optimization tool stays the calculation engine.
Human oversight in spare parts and materials management
Spares risk runs both ways: too little stock stops a critical asset, too much ties up capital for years. The agents keep routine policy current for non-critical items; every decision about what an asset needs to keep running stays with a named engineer or manager.
| Decision | Who decides | Agent's role | Control |
|---|---|---|---|
| Policy change inside band, non-critical item | Agent | Calculates, applies, logs reasoning | 5% sampled weekly; reversals above 2% suspend that item class |
| Requisition under an approved policy | Agent | Creates the requisition | Existing purchasing release strategy applies; agents never create the purchase order |
| Policy change outside band or above value threshold | MRO planner | Proposes with service and capital effect | Approval task with reasoning; large increases go to the materials manager |
| Inter-site transfer | Planner at both sites | Proposes the move and the residual risk | Sending site confirms; critical-asset stock excluded |
| Expedite of a late order | Buyer | Ranks by downtime exposure, drafts the request | Buyer owns supplier contact and premium freight |
| Supersession or interchangeability | Reliability engineer | Assembles notices and affected BOMs | Named approval; agents cannot link item numbers |
| Criticality classification or override | Reliability engineering | Shows exposure and flags gaps | Named approval on the item; agents read only |
| Obsolescence write-down or disposal | Materials manager and finance | Lists candidates with installed-base evidence | Existing disposal approval; out of agent scope |
Illustrative economics of spare parts planning agents
Example inputs describe a multi-site operation holding $60 million of spares across 45,000 items; change them to your own. Carrying cost and expedite savings are cash, avoided downtime is real but uneven, and planner hours become cash only if redeployed. The one-time working capital release is excluded from annual value.
Fixed assumptions: the inventory reduction comes as excess is consumed, without disposals; the build excludes internal team time and master data cleanup. Example inputs only. They describe no client and are not a forecast.
The cash view pays back slowly on purpose: carrying cost on a modest reduction is the only lever finance banks without argument. The case usually turns on avoided downtime, so the baseline needs a defensible count of stockouts that delayed work orders and an agreed cost per downtime hour. If those cannot be agreed, scope the first release around long-tail coverage and expedites, and let the stockout record build during proposal-only mode.
Rollout roadmap for spare parts agents, one site at a time
The live page sets the order: confirm item master and criticality readiness, run the planning agent proposal-only at one site, then enable execution inside tight guardrails and widen them as the record builds. The materials lead signs off each gate. Durations depend most on the state of BOMs and criticality data.
- Weeks 0 to 6
Data readiness
Profile one site's item master: duplicates, missing lead times, blank criticality, items with no equipment link. Baseline stockouts, expedites, excess and planner hours. Write guardrail policy v1.Gate: criticality set for critical-asset items, guardrail policy v1 signed - Weeks 6 to 14
Proposal-only
The installed-base and planning agents propose policy for every item due for review; planners decide as normal. Compare proposals with decisions and with what consumption did next.Gate: 90% or better acceptance on non-critical items - Weeks 14 to 22
Controlled execution
Enable execution for non-critical items inside a narrow value and service band, with 5% weekly sampling and a live pause drill.Gate: no critical stockout traced to an agent change, reversal under 2% - Week 22 onward
Scale
Widen bands by item class on the evidence, add cross-site rebalancing, then roll out site by site with site guardrails.Gate: per item class, per site
How spare parts planning performance is measured
The live page names critical stockouts, spares capital, obsolete share and long-tail coverage. This blueprint adds a control metric and two supporting measures, baselined from your data, with no targets claimed.
| Metric | Definition | Why it matters |
|---|---|---|
| Critical part stockouts | Stockouts on critical-asset items that delayed a work order | The failure the whole design exists to prevent |
| Working capital in spares | Spares value by site and item class at month-end | Shows whether capital is freed or just moved between sites |
| Obsolete and slow-moving share | Spares value with no issues in 24 months or no installed base | Stock that service-level settings never clear |
| Long-tail planner coverage | Active items whose policy was reviewed in 12 months | The capacity lever: most items are never reviewed |
| Proposal override rate | Proposals planners changed or rejected, and sampled updates reversed | The control metric that decides whether bands can widen |
| Reservation fill rate | Work order reservations issued complete on the required date | Ties spares service to maintenance schedule compliance |
| Expedite rate | Expedites as a share of spares orders, with premium cost | Direct cost of policy set too low |
Questions materials managers ask about spare parts agents
Can the agent decide which parts are critical?
No. Criticality stays with reliability engineering under named approval. The agents show the downtime exposure and lead time behind it and flag critical-asset items with no classification.
How does it plan parts that almost never move?
With methods built for intermittent demand, such as Croston-type forecasts or bootstrapping issue history, plus planned work and installed base. Insurance spares with no usage go to a person and are decided on risk.
Does this replace our MRP or inventory optimization tool?
No. The ERP or CMMS keeps running reorder point or min-max planning on the values it holds. An existing optimization tool stays the calculation engine; the agents handle data gaps, proposals, approvals and execution around it.
What happens to duplicate item numbers in our catalog?
The installed-base agent flags likely duplicates with evidence such as matching manufacturer part numbers. The master data team decides and merges; agents never merge item numbers.
How long until the agents change stock levels on their own?
Typically about fourteen weeks for one site: six of data readiness and eight proposal-only. Automatic updates start for non-critical items once that gate is met.
Working with WGA on spare parts and MRO inventory
Engagements follow WGA's AI Workforce Solution method and start with a readiness assessment of the item master. WGA profiles one site's catalog, measures how many items have a usable equipment link and criticality, counts stockouts that delayed work orders, and shows which item classes are ready for proposal-only mode. Guardrails are written with materials leadership and reliability engineering.
Senior WGA consultants lead the work with fees fixed against the outcome tied to the first release. Your materials, maintenance and IT teams inherit the guardrail policy, runbooks, training and pause procedure.
Discuss spare parts planning in your operation
Talk to our AI practice