Safety stock and replenishment parameter tuning keeps safety stock, reorder points, lot sizes, coverage targets and planned lead times current against the service target and the measured variability of demand and supply, and writes the approved values into the ERP and the planning suite each cycle.
First Monday of the quarter, 10:00. L. Tan opens the safety stock review for plant 1210, a role built from the blueprint. Juno lists 312 item-locations where the current safety stock sits more than 25% away from the value the policy gives, each with the reason: lead time drift, a change of class, a service miss, a variability that moved.
For material 7741-X the measured lead time from Acme Components GmbH has drifted from 21 to 29 days over six months, and the policy, 95% service on a class Y item, gives 640 pieces where the material master holds 420. For 188 C-class items where the stock change is under $5,000, version 2 of the policy lets Juno write the value to SAP S/4HANA on its own decision, and the change documents are back by 10:20.
The 124 larger changes wait for L. Tan, who approves 117 and edits 7. P. Sharma approves the planned delivery time change for Acme as a master-data change. J. Harrison sees the quarter’s inventory envelope move by $310,000 inside the agreed band, and the audit trail holds each change with its rule version and its approver.
Gartner names automated replenishment parameter changes, with touchless forecasting, as the place to focus early agentic efforts, because the actions are many and the cost of one wrong change is low (Gartner, 20 May 2026) 1. 59% of supply chain organisations have inventory optimisation initiatives for 2026, and 72% report AI in advanced planning and scheduling (The Hackett Group, 14 May 2026) 8.
Deloitte describes inventory optimisation agents as continuous service-level and safety stock adjustment and reports more than half of surveyed supply chain executives deploying agents, with the sample size unstated (Deloitte, 31 March 2026) 10. The named results are the most in the chain, most of them vendor-published: GAINS claims automation of up to 90% of buying decisions, 19% less inventory and 30% lower handling cost at Bell Flight, and Border States automated across 250,000 SKUs (GAINS, page modified 11 September 2026) 43.
ToolsGroup’s planning agent executes within human-controlled guardrails for 400 customers (ToolsGroup, 27 May 2026) 42. Oracle moved replenishment from periodic manual review to exception-based optimisation inside guardrails, SAP explains safety stock calculations in plain language for maintenance inventory, and the RELEX inventory control agent runs safety stock scenarios inside its rules engine (Oracle, 29 June 2026; SAP, 14 April 2026; RELEX, 2 July 2026) 31 27 44.
The money at stake is the largest in the chain. Inventory distortion, out-of-stocks plus overstocks, costs retailers $1.73 trillion a year, 6.5% of sales (IHL Group, 10 September 2025) 61; autonomous planning cuts inventory by up to 20% (McKinsey, 2 March 2022) 12; and Sainsbury’s reports 98% availability on about 30,000 SKUs and 150,000 fewer disposals in the 2025 peak (diginomica, 18 May 2026; Forbes, 5 June 2026) 38 37.
Quarterly, or when service slips, the inventory planner exports the planning parameters for a plant to a spreadsheet: MRP type (how the planning run treats the item), safety stock, reorder point, lot size, coverage profile, planned delivery time. Demand history and receipt history follow, and the planner computes variability and actual lead time per item and applies a service-level formula, or a weeks-of-cover rule of thumb, by ABC and XYZ class.
The proposed values are reviewed with the inventory manager and with finance for the working-capital effect, then keyed into the material master one item at a time or through a mass-change upload, with the master-data steward checking the fields. Two to four weeks per plant review is a fair estimate.
Most item-locations are reviewed once or twice a year. Between reviews the parameters drift as lead times and variability change, and planned delivery times are often years old. The result is the distortion IHL measures: too much of some items and too little of others, at the same time, in the same plant.
The manual path from the note, and the designed path from the scene. Showcase estate; the designed role is yet to be built.
The material master and the planning run at plant 1210 were registered in the first week for the PO cancellation agent, and the ontology already holds material at plant with its planning parameters. This use case would add the receipt history that gives measured lead times, the service targets by segment and the planning suite’s inventory module, and would give the agent a policy and an envelope to run inside.
| Row | Kind | What it tests or does | Data it reads | Then |
|---|---|---|---|---|
| 01 | Check | The item’s service target, from its ABC class | material master, segment | carry on |
| 02 | Check | Variability and lead time measured over six months | order history, receipt history | carry on |
| 03 | Action | Compute safety stock and reorder point; compare with the master | current parameters, stock, forecast error | row 04 decides the route |
| 04 | Limit | C class and stock change under $5,000 | stock change at standard cost | under: write · otherwise: L. Tan |
| 05 | Route | A planned delivery time change | planned delivery time, measured lead time | to P. Sharma as master data |
A plausible policy sentence for the role, and the planning table it would compile to. Illustrative values; the mechanics are the product’s.
The market runs parameter tuning at draft-and-approve, with execute within rules for low-value items: the vendors cited above run changes inside guardrails with review, or by exception, and the Gartner note of 20 May 2026 names replenishment parameter changes as a low-risk place to start. The blueprint designs safety-stock policy optimisation and replenishment policy setting at A2, coverage and buffer placement at A3, and parameter self-tuning at A4 once the record supports it. The showcase runs the low-value band at A3 inside an A2 agent, the split the PO cancellation agent reached after six weeks.
| Element | Setting | Set by | Changes |
|---|---|---|---|
| Objective | Hold 95% service on A and B items and 90% on C at plant 1210, at the lowest inventory value that achieves it | L. Tan, planner | New version |
| Data scope | Material master parameters, planning run, demand and receipt history, forecast error, segments and targets; plants 1210 and 1220 | A. Novak, data lead | New version |
| Rules | The safety stock policy and its planning table, version 2, with the formula by class | L. Tan with the Planner Agent | New version, simulated first |
| Trigger | The monthly parameter review, a lead time or forecast error beyond its band, a service miss, a class change, a confirmed supplier lead-time change | L. Tan, planner | At once |
| Actions permitted | Update safety stock, reorder point and lot size · update planned delivery time · assign the coverage profile · write the safety stock key figure · raise a review task | L. Tan, owner | New version |
| Limits | Direct write only on C items under $5,000 of stock change · inventory envelope ±$500,000 a quarter · 400 changes a day · confidence floor 0.80 | J. Ridley and J. Harrison, finance | New version |
| Escalation | Changes above the band to L. Tan · planned delivery time and lot size to P. Sharma · the envelope beyond its band to J. Harrison | L. Tan, owner | At once |
| Level | A2 Supervised with L. Tan as the named approver; the low-value band at A3 | J. Ridley, director | Up on evidence, down in one step |
| Feedback | Every approve, edit and reject is recorded and used in the next cycle | Always on | Fixed |
Every row is a setting or a versioned rule; the level and the limits are the ones a manager signs.
Lima’s agent blueprint is a design of 960 agent roles across five end-to-end chains. It is the map of what Juno can be configured to do, and the first agents are built from it one at a time, the way the PO cancellation agent was.
Additional capability in Plan-to-Produce lists the rest of the map for this chain.
| Measure |
|---|
| Service level and fill rate against target by segment |
| Days of inventory and inventory value (case baselines: inventory down 19% at Bell Flight, GAINS; up to 20% with autonomous planning, McKinsey) |
| Stock-outs and excess value |
| Share of item-locations with parameters reviewed in the last 90 days |
| Share of parameter changes executed within rules, approved, edited or rejected |
| Working capital released |
Baseline service level and fill rate against target by segment first, because the policy is written in those terms and a change of parameters shows there within a cycle or two. Days of inventory and inventory value are the finance measures and the envelope the agent runs inside, so J. Harrison sees them move from the first cycle. The share of item-locations with parameters reviewed in the last 90 days is the measure that moves fastest: from once or twice a year to every cycle.
Because it is a master-data field, and a wrong lead time moves every order date the planning run creates. P. Sharma validates it the way she validates any master-data change, and the change document carries her name.
The forecast is in scope: a launch, a design win or a moved customer forecast changes the segment before the parameter is recomputed, and a conflict goes to L. Tan as a review task. A service miss is a trigger.
On evidence: a passing test on the recorded changes, J. Ridley’s approval and a live route to L. Tan for the exceptions, band by band. The envelope and the steward’s approval stay in place at every level.
The plan is right and the parameters are current. Then a placement machine stops on a Tuesday afternoon. Who re-plans the week, and how fast?
The agent would read the event, re-check material and capacity, propose the re-sequence that protects every committed date and release only the orders that are ready.
Bring two systems and one decision your planners make every week. In the first week we register the data, confirm the mapping with your team, answer your questions in ChatLima and run the first agent at the level you choose.