Inventory and Production Management in Supply Chains, Fourth Edition

1. 1Fix the definition before measuring: forecast error is the gap between forecast and observed demand for the same item and the same period. Track two things separately: the size of the errors (magnitude) and whether they lean one way (bias). The source's contents place these under 'Measures of Forecast Accuracy' (3.5.1) and 'Monitoring Bias' (3.5.3), as distinct topics.
2. 2Rank SKUs by annual dollar usage (the DBV analysis) and set the accuracy-review effort by class. Class A is the first 5%–10% of SKUs (up to 20% in some settings), usually 50% or more of annual dollar movement. The source says model parameters and demand estimates are reviewed more often for A items, so A items get the most frequent accuracy review.
3. 3Match the forecast level to the decision. The source separates aggregate medium-range methods (3.3) from individual-item short-term methods (3.4). Score each forecast at the level where the decision is made, and do not borrow an aggregate accuracy figure for an item-level decision.
4. 4Measure the standard deviation of forecast errors over the replenishment lead time, not just over one period (3.5.2). The lead-time error is the quantity the inventory decision depends on, so compute it on the same horizon.
5. 5Monitor bias on its own track (3.5.3). A forecast can have small average error magnitude and still run persistently high or low, so keep the signed and unsigned views of the error separate.
6. 6Screen out anomalous demand (3.6) and record human judgment inputs (3.7) so you can see their effect. The source lists factors where judgment input is needed and guidelines for inputting and monitoring it. Log each override with its reason, then compare overridden and un-overridden errors.
7. 7Segment items that standard accuracy measures misrepresent: items with limited history, intermittent and erratic demand, service parts, and terminal demand (3.8). Evaluate each class on its own terms, not against the smooth-demand baseline.
8. 8When accuracy degrades, apply the corrective-action step (3.5.4), such as revisiting smoothing constants (3.4.5). Then judge the process on two levels: statistical accuracy (3.9.1) and strategic issues (3.9.2).

Table 2.1 in the source ranks SKUs by descending annual dollar usage. SKU 1 (the top 0.5% of SKUs) carries $3,000, or 13.3% of total usage. SKU 2 brings the cumulative figure to $5,600 (24.9%), and SKU 3 brings it to $7,900 (about 35%). Three SKUs together drive roughly a third of dollar movement. A planner with limited review time puts these three first, because a given percentage error costs the most there. The planner computes their error magnitude and bias each period and reviews the demand estimates behind them more often than for B or C items, as the source prescribes for class A. The long tail of class C items is handled by grouping, with the source noting that a single order rule and monitoring system can serve each control group, so one group-level check replaces many item-level reviews. The passages supply the prioritisation logic and the structure of the accuracy section, not the error formulas, so the planner adopts the specific error statistics from sections 3.5.1–3.5.3.

• If an SKU falls in class A (top 5%–10% of SKUs, 50%+ of dollar usage), then review its forecast error and demand parameters more often than B or C items.
• If errors are small on average but consistently one-signed, then treat it as a bias problem (3.5.3) and apply the corrective-action step (3.5.4) rather than accepting the average.
• If a safety-stock or reorder decision depends on the forecast, then measure error spread over the lead time (3.5.2), not the single-period error.
• If an item has limited history, intermittent or erratic demand, is a service part, or faces terminal demand, then do not judge it against the smooth-demand benchmark. Evaluate it as its own class (3.8).
• If a C item's forecast is poor, then fix it at the control-group level (shared usage rate, supplier, seasonality, lead time) rather than spending item-by-item effort.

• Reporting one blended accuracy number across all SKUs. The tell: a healthy headline figure while class A items still stock out or overstock.
• Watching error size but never bias. The tell: forecasts stay 'close' yet inventory drifts persistently high or low.
• Measuring error over one period when the decision exposure is the lead time. The tell: safety stock is repeatedly too thin or too fat despite acceptable period accuracy.
• Letting judgment overrides go unlogged. The tell: nobody can say whether manual adjustments improved or worsened the forecast.
• Scoring intermittent, service-part, or end-of-life items with the same yardstick as steady sellers. The tell: those items always look 'inaccurate' and are either over-reviewed or ignored.
• Ignoring record accuracy. The source warns that inaccurate stock records cripple decision systems. The tell: forecast error looks large because the 'actual' demand or on-hand data is wrong.

• ▸SKU ID and ABC class: record the class from the dollar-usage ranking (A = top 5%–10% of SKUs).
• ▸Cumulative % of annual dollar usage: copy from the ranking table to justify review frequency.
• ▸Forecast level and horizon: aggregate or item-level, and the lead time the forecast must cover.
• ▸Error magnitude (period): the size of forecast-minus-actual gaps, using the chosen measure from 3.5.1.
• ▸Lead-time error spread: standard deviation of forecast errors over the replenishment lead time.
• ▸Bias reading: signed error trend; note whether it leans high or low and for how many periods.
• ▸Anomalous demand flagged: periods excluded or adjusted, with the reason.
• ▸Judgment overrides: date, who, direction, reason, and the resulting error versus the statistical forecast.
• ▸Demand class: standard, limited history, intermittent/erratic, service part, or terminal.
• ▸Corrective action and next review date: smoothing-constant change or other fix, plus when A/B/C cadence says to recheck.

1. 1Split total inventory into the four forms in the construct definition: cycle stock, safety stock, anticipation inventory, and pipeline/work-in-process. Record units and dollars for each. A single aggregate number hides which form is driving carrying cost.
2. 2Express each form as an inventory-to-sales ratio, not an absolute level. The source tracks inventory-to-sales for retailers, manufacturers, and wholesalers, and absolute dollars rise with sales even when performance improves.
3. 3Map where the stock sits in the chain: raw material, WIP, finished goods at the manufacturer, and wholesale or retail stock downstream. The source shows inventory migrating toward the customer. Manufacturers held 46% of U.S. inventory in 1992 and 37% in 2014, so a plant-level view can miss where the carrying cost now lands.
4. 4Rank items with the A–B–C classification (Chapter 2) before choosing a control method. Reserve detailed item-level models for the A items and use simpler rules for the rest, because stock-keeping units are too diverse to manage one by one.
5. 5For each A item, trace the driver of its level to one of the givens the source names: replenishment lead time (global sourcing lengthens it), demand uncertainty, product variety or short life cycle, and gross margin. The source reports that global sourcing, higher gross margin, and higher demand uncertainty are each associated with higher inventory.
6. 6Attack the givens before tuning the policy. The source argues that optimizing against fixed lead times and demand variability yields efficiency that is 'insignificant compared to the efficiencies available by changing the givens'. Shorten lead times and reduce observed demand variability, for example by sharing demand information with supply chain partners, before refining order quantities and safety stock.
7. 7Only after the givens are addressed, set order quantities (EOQ-type, Chapters 4–5) and safety stock under probabilistic demand (Chapter 6) to size cycle and safety stock for the remaining lead time and variability.
8. 8Review quarterly: recompute the ratios, the form-by-form split, and the location split. Investigate any form that rises while sales are flat, because the source ties abnormal inventory levels to poor stock returns and excess-inventory announcements to negative market reactions.

The source puts U.S. inventories at over $2 trillion in 2007 and close to $2.5 trillion by 2014, with manufacturers holding 37% of the 2014 total. Applying the share to the approximate total gives about $0.93 trillion at manufacturers and about $1.57 trillion at wholesalers and retailers. This is rough arithmetic on approximate figures, not a reported number. The reading is that most stock now sits downstream, so a retailer's carrying-cost problem is not the manufacturer's. The source also explains why: global sourcing lengthens lead times, and the Jain et al. (2013) study finds firms with more global sourcing hold more inventory. Chen et al. (2005) found a 2% average annual decline in inventory levels from 1981 to 2000, mostly from reduced WIP, while finished goods barely moved. Applied to your own firm, a high finished-goods or downstream level points to lead time and demand uncertainty as the givens to attack. A high WIP level points to the production-flow side, where the historical gains came from.

• If total inventory dollars rise but the inventory-to-sales ratio is flat or falling, treat it as growth, not a level problem.
• If an item is sourced globally and its stock is high, then shorten or de-risk the lead time before touching safety stock parameters, because longer lead times drive higher levels.
• If an item has high gross margin, short life cycle, and uncertain demand, then expect higher inventory by nature. Judge it against similar items, not against stable low-margin items.
• If an item is class A, then give it a dedicated item-level model. If it is class C, then use a simple rule so management attention goes to the items that carry the cost.
• If lead time and demand variability are treated as fixed, then stop optimizing and first test whether information sharing or lead-time reduction can change them.

• Reading the aggregate only: the total looks fine while one form (for example finished goods) is bloated. The tell is that the total ratio improves but finished goods barely move, which is the pattern Chen et al. (2005) found.
• Assuming information sharing or 'strategic alliances' alone cuts inventory. The tell is data flowing in with no change to ordering or safety stock rules, and levels staying the same.
• Optimizing policy on given lead time and variability while ignoring the givens. The tell is that sophisticated reorder-point or EOQ models are in place and inventory still exceeds peers.
• Looking only at your own site's inventory while stock migrates downstream. The tell is that the plant's inventory falls while total chain stock and expedite costs rise.
• Ignoring the finance signal: excess inventory is treated as an operations matter only. The tell is that no one monitors abnormal inventory levels, even though the source links them to negative market reaction and poor returns.

• ▸Item or family and ABC class: record the class (A, B, or C) to decide how much modeling effort it gets.
• ▸Cycle stock (units / $): enter the quantity held in the normal replenishment cycle.
• ▸Safety stock (units / $): enter the buffer held against demand and lead-time uncertainty.
• ▸Anticipation inventory (units / $): enter stock built ahead of expected demand or events.
• ▸Pipeline / WIP (units / $): enter stock in transit or in process.
• ▸Inventory-to-sales ratio and prior-period ratio: compute both and note the direction.
• ▸Location split: list the share held at manufacturer versus wholesaler or retailer.
• ▸Driver check: mark lead time (global sourcing), demand uncertainty, variety or short life cycle, and gross margin.
• ▸Given to change first: name the one lead-time or variability change to pursue before any policy retuning, with an owner and date.

1. 1Fix the exposure window first. In an (s,Q) system, stockout risk exists only over the replenishment lead time, because the order is triggered at s and nothing arrives until the lead time ends. In an (R,S) system, the window is the review interval plus the lead time. Write the window down for each SKU before touching any numbers.
2. 2Name the two uncertainties that create the risk: demand and supply. Safety stock exists only because the demand rate and the delivery time are not known with certainty. If both are truly known, the stockout risk is zero and no safety stock is needed. Record which source dominates for each SKU. The treatment of variability in the lead time itself is a separate section (6.10).
3. 3Estimate the spread of lead-time demand from forecast errors, not from raw demand history. The source treats the standard deviation of forecast errors over a lead time as the input (3.5.2). Check forecast bias (3.5.3) before you trust that spread. A biased forecast shifts the whole distribution, and no safety factor repairs that.
4. 4Choose the service measure before you choose s. The source offers these as distinct decision rules: a specified safety factor k; a cost B1 per stockout occasion; a fractional charge B2 per unit short; a fractional charge B3 per unit short per unit time; a charge B4 per customer line item short; a probability P1 of no stockout per cycle; a fraction P2 of demand met from the shelf; and an average time between stockout occasions (TBS). These measures answer different questions. P1 counts occasions. P2 counts units. B1 prices occasions. B2 and B3 price units. B4 prices order lines. Match the measure to how a stockout actually hurts the customer.
5. 5Convert the chosen measure into the reorder point using the matching decision rule. Rules 6.7.4 to 6.7.11 each give a different route to s. Then read back the implied costs and performance measures (6.8) for the s you got. Confirm that the implied probability, fill fraction and cost are tolerable for that SKU.
6. 6Price the cost of disservice from the customer's perspective. List the measurable costs: expediting, rescheduling, split lots, changeovers for emergency orders, emergency shipments, and substitution of a less profitable item. Then state your judgment on the fuzzy ones: whether the customer waits (backorder) or the sale is lost, goodwill, return likelihood, and word of mouth. The source says most SKUs can only be handled by considered opinion, so record the judgment and who made it.
7. 7Allocate a limited safety-stock budget across SKUs. When total safety stock (TSS) is capped, use the allocation rules that minimize expected total stockout occasions per year (ETSOPY) or expected total value of shortages per year (ETVSPY). Do not apply one uniform k to every item.
8. 8Test the distribution assumption only where it matters. The source says results are often insensitive to the distribution for a given mean and variance, so a normal distribution is usually adequate. Use the nonnormal treatment (6.7.14) only where lead-time demand is clearly not normal. For lumpy items, a discrete distribution is workable (6.6.2). Apply the A–B–C logic when you decide how much effort to spend: A items get frequent parameter review, and C items get simple rules.

The source describes a manufacturer whose leadership saw inventory turnover below the industry average and ordered it raised to that average. The company made in-house some parts that competitors bought, and its work-in-process (WIP) inventory was needed to keep production flowing. The controller recalculated the ordering rules so lot sizes were cut and raw material inventory fell. In stockout-risk terms, the cycle and safety stock behind the exposure window were shrunk without regard to demand or lead-time uncertainty. Shipments continued for a few months while warehouse inventory was depleted. Then finished goods and component inventories bottomed out, and customer service degenerated. The stockouts were rampant and caused massive, expensive expediting. The smaller lot sizes also raised product cost, and profits plummeted. The lesson is operational. Any inventory-reduction target must be tested against the stockout measure (P1, P2 or TBS) it implies. Otherwise the stockout risk only shows up later, as expediting cost and lost service. The delay of a few months is the tell: a stockout-risk increase from lower inventory is invisible until the buffer is gone.

• If demand and lead time are both known with certainty, then carry no safety stock. If either is uncertain, then size safety stock to the chosen service measure and not to a rule of thumb.
• If the damage from a stockout does not depend on how many units are short (for example, a line stops or the customer cancels the order), then use an occasion-based rule: B1, P1 or TBS. If damage grows with the quantity or duration short, then use a unit-based rule: B2, B3 or P2.
• If the item is a class C SKU, then keep a relatively large number of units on hand and use a simple system such as two-bin, grouped into control groups. Spend analysis effort on A items and review their parameters more often.
• If a turnover or inventory-reduction target is imposed, then recompute the implied stockout probability or fill fraction (6.8) before cutting lot sizes or buffers. Reject the cut if the implied service is below what the SKU needs.
• If the cost of disservice cannot be measured, then still set an explicit written estimate (backorder or lost sale, goodwill, return likelihood). Do not leave the cost at zero by default.

• Targeting an aggregate turnover number and cutting stock across the board. The tell is that service holds for a few months, then finished goods run out and expediting spikes, as in the manufacturer example.
• Treating stockout risk as one number. The tell is a single k or one fill-rate target applied to every SKU, with no distinction between occasion-based and unit-based harm.
• Computing safety stock from the average demand history instead of forecast-error spread over the lead time. The tell is stockouts that cluster after demand shifts, and a persistent forecast bias that nobody monitors.
• Ignoring supply-side variability. The tell is stockouts that occur while demand sits at its forecast, because the lead time itself varied.
• Putting zero on the cost of lost goodwill because it is hard to measure. The tell is a reorder point that looks cost-optimal on paper while customers defect, and no written assumption about whether they wait or leave.

• ▸SKU / control group: identify the item and the group it is controlled with (same supplier, usage rate, lead time).
• ▸A–B–C class: record the class and the review frequency it implies.
• ▸Control system and exposure window: (s,Q) lead time, or (R,S) review interval plus lead time.
• ▸Uncertainty sources: mark demand, supply (lead-time variability) or both, and the evidence for each.
• ▸Forecast-error spread over the lead time and bias check: record the value and the date you last checked bias.
• ▸Chosen service measure: pick one (k, B1, B2, B3, B4, P1, P2 or TBS) and give the reason it matches how the customer is hurt.
• ▸Target value of the measure: the numeric target, and who approved it.
• ▸Resulting reorder point s and implied performance (6.8): record the implied stockout probability, fill fraction and cost at that s.
• ▸Disservice cost assumptions: measured costs (expediting, emergency shipment, substitution) and judged costs (backorder vs. lost sale, goodwill, return), with the estimator named.

1. 1Segment items before choosing a method. Use the A–B–C classification by annual dollar usage so that control effort goes to the few SKUs that carry most of the dollar volume. Give A items individual attention and C items simple rules. The source treats the A–B–C classification as the basis for designing individual-item decision models.
2. 2Build one demand forecast per item that everyone plans against. Pick the statistical procedure to match the pattern: level, trend, or seasonal (Winters). Handle anomalous demand explicitly instead of letting it distort the history. Flag special classes for separate treatment: limited history, intermittent or erratic demand, service parts, and terminal demand.
3. 3Measure the forecast, because alignment cannot be better than the forecast. Track forecast accuracy and monitor bias separately. Estimate the standard deviation of forecast errors over the replenishment lead time, since that figure sizes the protection against shortage. Define corrective actions in advance for when bias or error leaves its limits.
4. 4Run judgment through a controlled channel. Record who adjusted a forecast, why, and on what assumptions. Galbraith and Merrill found that executives sometimes change forecasts to influence the stock price or internal resource allocations, not to improve accuracy. An unrecorded override that serves another motive pushes production away from demand.
5. 5Set up a consensus-forecasting process, a core part of Sales and Operations Planning (S&OP). Hold a series of meetings with operations planning, marketing, and finance. Keep sales, forecasts, and assumptions in one shared IT system. Oliva and Watson document this as the fix for poor cross-functional alignment.
6. 6Match the quantity rule to the demand pattern. For approximately level demand, use the EOQ with sensitivity analysis. For time-varying demand, choose among lot-for-lot, EOQ expressed as a time supply (POQ), the Silver–Meal heuristic, and Wagner-Whitin, so lots follow the demand timeline instead of a fixed size.
7. 7Under probabilistic demand, pick the control system and state the stockout treatment. Choose among (s,Q), (s,S), (R,S), and (R,s,S), and choose continuous versus periodic review. Decide whether unmet demand is backordered or lost, and set explicit cost or service objectives for the safety stock.
8. 8Close the loop each cycle. Compare actual demand to forecast and to what was produced or bought. Record shortages and excess stock by item, and feed the pattern back into forecast parameters, safety stock, and the consensus meeting agenda.

The source's clearest real case is Oliva and Watson's study of a high-tech manufacturer. There, the organization structure and the forecasting process produced poor forecasts. Resource allocation decisions built on those forecasts were poor too, so production and procurement drifted away from what customers wanted. Their remedy had two parts. First, a common IT system held all the relevant information: sales, forecasts, and assumptions. Second, a consensus-forecasting process held a series of meetings with stakeholders, typically operations planning, marketing, and finance, until they agreed on one forecast. This is the core of S&OP. Galbraith and Merrill explain why the fix is needed. Some executives revise forecasts to influence the stock price or internal resource allocation, not to improve accuracy. Putting assumptions in a shared system makes those revisions visible. Applying this to your own firm, you would pull each function's separate numbers into one record, discuss the gaps, and have operations plan from the agreed figure only.

• If demand for an item is approximately level, use the EOQ and run a sensitivity check. If demand is significantly variable over time, use a time-phased lot-sizing rule such as Silver–Meal, POQ, or lot-for-lot instead of a fixed EOQ.
• If an item is class A, give it individual forecasting, safety stock, and review attention. If it is class C, use simple rules and keep review effort low.
• If forecast bias persists on an item, correct the forecast process or parameters before raising safety stock. Buffering a biased forecast hides the misalignment without removing it.
• If the forecast is revised by judgment, require a stated reason and assumptions in the shared system. If no reason can be given, keep the statistical forecast.
• If an item has limited history, intermittent demand, is a service part, or faces terminal demand, do not use the standard smoothing procedure unchanged. Use the special-class treatment.

• Each function plans from its own numbers. The tell: sales, finance, and operations quote different forecasts for the same period, and production is sized to none of them.
• Forecasts are padded or inflated for motives such as share price or resource allocation. The tell: forecasts are consistently biased in one direction and nobody can trace the adjustments.
• Fixed order quantities are applied to items with lumpy, time-varying demand. The tell: stock builds in slow periods and runs out in peak periods, even though the average lot looks reasonable.
• Accuracy is tracked but bias is not. The tell: average error looks acceptable while the same item is repeatedly over- or under-forecast in the same direction.
• Intermittent, service-part, or end-of-life items are run through the standard exponential smoothing. The tell: erratic forecasts, shortages on sporadic demand, and leftover stock after demand ends.

• ▸Item / family and A-B-C class: record the SKU group and its class by annual dollar usage.
• ▸Demand pattern: mark level, trend, seasonal, intermittent, service part, or terminal.
• ▸Forecast owner and consensus attendees: name who set the agreed forecast and which functions (operations, marketing, finance) signed off.
• ▸Forecast accuracy and bias: enter the current accuracy measure and bias reading, and say whether either breached its limits.
• ▸Std. deviation of forecast errors over lead time: record the figure that sizes safety stock.
• ▸Judgment overrides: list each adjustment to the statistical forecast, who made it, and the stated reason and assumptions.
• ▸Quantity rule in use: record EOQ, lot-for-lot, POQ, Silver–Meal, or Wagner-Whitin, and why it fits the demand pattern.
• ▸Control system and unmet-demand treatment: record (s,Q), (s,S), (R,S), or (R,s,S), review type, and backorder versus lost sale.
• ▸Shortages and excess observed: note stockouts and surplus this cycle, with the apparent cause.
• ▸Corrective action and owner: state the change to forecast, parameters, or process, and who completes it by when.

1. 1Name the decision first: order quantity, reorder point or safety stock, review interval, or order-up-to level. Total system cost is only the sum of costs that change when that decision changes. Write the decision variable at the top of the page.
2. 2Sort every cost into the three families the source builds its models on: carrying cost (cost of holding stock), replenishment cost (ordering or setup, incurred per replenishment), and shortage cost (incurred when demand exceeds stock). Put each cost in one family only, so nothing is counted twice.
3. 3Apply the relevance test to each line: does this cost change if the decision variable changes? Drop fixed overheads that do not move. Keep costs that do move, even if accounting does not report them under an inventory label.
4. 4Settle the shortage treatment before costing it. The source flags backorders versus lost sales as a basic issue under probabilistic demand. Under backorders, shortage cost is the cost of delay and expediting. Under lost sales, it is the lost margin plus any goodwill loss. Record which assumption you use.
5. 5Measure explicit costs directly (Chapter 2.9 approach). If shortage cost cannot be measured credibly, use implicit measurement: set a service target, then use an exchange curve to see the inventory needed to reach it. The curve shows the trade-off without a guessed dollar figure.
6. 6Build the total cost as carrying + replenishment + shortage, expressed as a function of the decision variable. Find the minimum with the item-level rule for that system (the EOQ derivation for level demand, the (s,Q) or (R,S) rules for probabilistic demand).
7. 7Run sensitivity analysis on the inputs (Chapter 4.3 approach). Vary each cost estimate and check whether the chosen decision moves. If it barely moves, stop refining that estimate.
8. 8Apply the decision by item importance (A, B, C classification). Spend measurement effort on Class A items and use simple rules for Class C. Then check that stock counts and transient effects do not distort the baseline you compare against.

The source's only hard numbers are macro-level, but they show why you sum components instead of watching one. Table 1.1 reports U.S. inventory value above $2 trillion in 2007, rising to close to $2.5 trillion by 2014, which is roughly a 25% increase in carrying-cost base. Over the same period, total logistics cost as a share of GDP fell from 9.9% in 2000 to 8.2% in 2014, a 1.7-point drop. If you tracked only inventory value, the system looks more expensive. If you tracked total logistics cost relative to output, it looks cheaper. Neither alone is a decision measure. The two series do not isolate an item-level trade-off, but they illustrate it: a larger inventory can coexist with a lower total cost if replenishment and shortage costs fall faster than carrying cost rises. The source also notes that inventory has shifted downstream toward the customer. At item level, that means re-running the three-family sum at each stocking point rather than carrying one cost set across the network.

• If a cost does not change when the decision variable changes, then exclude it from total system cost, however large it is.
• If shortage cost cannot be estimated with confidence, then use an exchange curve (inventory versus service) and choose the service level explicitly instead of burying a guess in the cost function.
• If sensitivity analysis shows the decision is stable across the plausible range of a cost input, then stop refining that input and move to the next item.
• If an item is Class A, then measure all three cost families explicitly. If it is Class C, then use a simple rule and accept an approximate total.
• If demand is probabilistic, then confirm the backorder-versus-lost-sales assumption before comparing any two policies, because the shortage term differs between them.

• Optimizing one component, usually carrying cost, because it is the easiest to see. The tell: inventory targets fall while expediting, stockouts or setup counts rise.
• Counting the same cost in two families, for example expediting in both replenishment and shortage. The tell: the total exceeds what the finance records support.
• Including non-relevant costs, such as allocated overhead, in the cost function. The tell: the optimal decision changes when overhead allocation rules change, though nothing physical changed.
• Applying a backorder shortage cost to a lost-sales environment, or the reverse. The tell: modeled service levels diverge sharply from actual customer behavior during stockouts.
• Comparing policies against a baseline distorted by transient effects or inaccurate stock records. The tell: the apparent savings appear in the first period after the change and vanish later, or physical counts disagree with system records.

• ▸Decision variable: state exactly what is being chosen (order quantity, reorder point, review interval, order-up-to level).
• ▸Item and class: record the item and its A/B/C class to set how much measurement effort it deserves.
• ▸Carrying cost items: list each cost that varies with stock held, with its estimate and source.
• ▸Replenishment cost items: list each cost incurred per order or setup, with its estimate and source.
• ▸Shortage assumption: mark backorders or lost sales, and list the shortage cost items that follow from that choice.
• ▸Shortage cost basis: record explicit dollar estimate or implicit service target (exchange curve), and say why.
• ▸Relevance check: for every line above, write yes or no on whether it changes with the decision variable, and remove the no lines.
• ▸Total cost at candidate decisions: record carrying + replenishment + shortage for the current and proposed values.
• ▸Sensitivity result: note which input, varied over its plausible range, changes the chosen decision, and which does not.

1. 1Pick the service measure first and name it in writing: P1 (probability of no stockout per replenishment cycle) or P2 (fraction of demand satisfied directly from the shelf). The two are not interchangeable. P1 counts cycles with no stockout. P2 counts units or demand filled.
2. 2Fix the control system before setting the target. In an (s,Q) system, protection is needed only over the replenishment lead time. In a periodic-review (R,S) system, it is needed over the review interval plus the lead time. Service-level math applies to that protection window.
3. 3Describe demand over that window. The source's decision rules start from a normal distribution with a safety factor k. Treat that as the default, and switch to the nonnormal rules if demand is lumpy or skewed. The source also gives a discrete-distribution route for small, countable demand.
4. 4Convert the target into a reorder point s (or order-up-to level S) with the matching rule. For P1, choose s so the chance that lead-time demand exceeds s equals 1 − P1. For P2, choose s so expected units short per cycle, divided by order quantity Q, equals 1 − P2. Q enters the P2 rule and does not enter the P1 rule.
5. 5Include lead-time variability if it exists. Use either the total demand over the full lead time or the demand-rate distribution combined with the lead-time distribution. A fixed-lead-time assumption understates the protection needed.
6. 6Declare how unmet demand is treated, backordered or lost, before computing shortages. The assumption changes the shortage arithmetic.
7. 7Set targets differently by item importance (A, B, C), not one rate for everything. Use the single-item exchange curve (inventory versus service) to see the safety stock cost of each extra point of service. Then use composite exchange curves to allocate a total safety-stock budget across items.
8. 8Check the result against reality. Compute the implied cost and performance measures, such as the implied P2 when you set P1. Then compare against actual fill and stockout frequency, and adjust.

This is an illustrative construction using the definitions of P1 and P2, not an example from the source. Assume Q = 100 units and lead-time demand takes four values: 60 (probability 0.2), 80 (0.3), 100 (0.3), 120 (0.2). Cumulative probabilities are 0.2, 0.5, 0.8 and 1.0. At s = 80, P1 = 0.5. Shortage occurs when demand is 100 (20 short × 0.3 = 6) or 120 (40 short × 0.2 = 8). Expected shortage is 14 units per cycle, so P2 = 1 − 14/100 = 0.86. At s = 100, P1 = 0.8. Shortage occurs only at demand 120 (20 × 0.2 = 4 units), so P2 = 1 − 4/100 = 0.96. At s = 120, both measures equal 1.0. A target of P1 = 0.8 therefore sets s = 100, which happens to deliver 96% fill. The same s = 80 that stocks out in half the cycles still fills 86% of demand. The two measures can look very different for the same stock level, so the target must be written as P1 or P2.

• If the contract or customer promise is about units filled (fill rate), use P2. If it is about the chance of being out of stock when an order cycle ends, use P1.
• If Q is large relative to lead-time demand variability, expect P2 to be high even at a modest P1. Check the implied P2 before paying for extra safety stock to reach P1.
• If lead time itself varies, use one of the two lead-time-variability approaches instead of a fixed-lead-time reorder point.
• If lead-time demand is visibly nonnormal (lumpy, intermittent, skewed), use the nonnormal or discrete-distribution decision rule instead of the safety factor k.
• If safety stock is budget-limited across many items, equalize the marginal service gained per dollar using the composite exchange curve, not equal time supplies.

• Quoting 'service level' without saying whether it is P1 or P2. The tell: two teams report different numbers for the same item and policy.
• Setting reorder points as equal time supplies across items. The tell: fast and slow items show very different service for the same inventory investment, which the source's exchange-curve comparison is designed to expose.
• Using a normal-distribution safety factor on lumpy demand. The tell: actual stockout frequency departs persistently from the target.
• Ignoring lead-time variability. The tell: stockouts cluster after late deliveries even though demand was on forecast.
• Assuming backorders when sales are actually lost, or the reverse. The tell: reported fill rate disagrees with computed shortage per cycle.

• ▸Item / class (A, B, C): record the item and its importance class.
• ▸Control system: (s,Q), (s,S), (R,S) or (R,s,S); this sets the protection window.
• ▸Service measure chosen: P1 or P2, with the written target value.
• ▸Unmet demand treatment: backordered or lost sales.
• ▸Protection window length: lead time (or R + lead time), with the lead-time mean and variability noted.
• ▸Demand distribution over the window: normal, nonnormal or discrete, with parameters.
• ▸Order quantity Q: needed for the P2 calculation.
• ▸Computed s (or S) and safety factor k: record the result of the matching decision rule.
• ▸Implied other measure: P2 if P1 was set, or P1 if P2 was set; compare to actual performance.

1. 1Trace profit to its drivers before changing anything. The source's financial tree runs Revenue (product sales, service, royalties, interest/dividend income) minus operating expenses to net profit, and net profit to shareholder value. Working capital (inventory, cash, receivables, prepaid expenses) and fixed capital sit on the investment side. Write down which branch your proposed operational change touches.
2. 2Pair every cost move with its revenue effect. Revenue is driven by customer service and costs by operational efficiency, so no inventory or production change is judged on cost alone. For each proposal, list the expected change in inventory investment and the expected change in service level (stockouts, delivery delays).
3. 3Anchor decisions in the firm's strategic position. Functional strategy (maximizing resource productivity) is constrained by business strategy (competing on price, service, or another factor). Before you set a target such as inventory turnover, ask what the business strategy requires. The source says competitive advantage affects the appropriate turnover number.
4. 4Set policy at the aggregate level, then derive item-level rules. Choose a target operating point on the aggregate exchange curve (total average inventory dollars vs. replenishments per year). That choice implies a value of A/r, which then feeds the EOQ formula for each item. Use cost estimates only as ballpark baselines and adjust them until they reproduce the aggregate policy you chose.
5. 5Locate your process type and read its profit lever. Job shop: labor and machine utilization and shifting bottlenecks. Batch flow: balancing stages around a single bottleneck. Assembly line: line balancing and staffing, where small allocation changes materially move profitability, plus worker morale. Process industry: minimal equipment downtime. Put your improvement effort on that lever first.
6. 6Segment inventory by value before tuning. Roughly 20% of SKUs account for 80% of annual dollar usage. Concentrate cost-reduction and service attention on that group, since it holds most of the inventory dollars and most of the profit exposure.
7. 7Test for service degradation after an inventory cut. Inventory reductions can show up as benefits for a few months while warehouse stocks deplete, and the damage arrives later. Track finished-goods and component stock, stockouts, and expediting cost for several months after any lot-size change, not just the first quarter.
8. 8Watch the external signal on inventory. The source cites studies where excess-inventory announcements are followed by negative stock-market reactions and abnormal inventory levels relate to poor stock returns. Review inventory-to-sales against its trend and against the business cycle, so you can separate a market-wide buildup from a firm-specific one.

The source describes a firm whose management wanted higher inventory turnover. The firm had a special competence in a type of manufacturing that needed WIP inventory to keep production flowing. The edict was effectively 'buy more and make less.' The controller recalculated the ordering rules, cutting lot sizes and reducing raw material inventory. For a few months shipments continued because warehouse inventories were being depleted, so the turnover metric looked better. Then finished-goods and component inventories bottomed out and customer service degenerated. Rampant stockouts forced massive, expensive expediting. The smaller lot sizes also drove up product cost, and profits plummeted. Both profit drivers moved the wrong way at once: revenue fell with service and costs rose with expediting and unit cost. The lesson is that the correct turnover target depends on the firm's competitive advantage, and a cost-side ratio cannot be set in isolation from service.

• If a proposal lowers inventory investment, then require an explicit estimate of its effect on service level and on expediting cost before approving it.
• If the source of advantage is a special manufacturing competence that needs WIP to flow smoothly, then do not apply a generic turnover target; set the target from the competitive strategy.
• If your process is a job shop, then attack utilization and shifting bottlenecks first; if it is an assembly line, then re-balance station allocation, because small shifts there carry large profit effects.
• If cost measurements are uncertain, then proceed with ballpark estimates and calibrate to the aggregate policy, because inventory and production models are relatively insensitive to cost errors.
• If you must prioritize attention, then start with the roughly 20% of SKUs that carry about 80% of annual dollar usage.

• Optimizing a single ratio such as inventory turnover. Tell: turnover improves for a few months while stockouts and expediting costs climb later.
• Treating profitability as a cost-only problem. Tell: lot sizes are cut and unit cost rises, yet no one has measured the revenue lost to poorer customer service.
• Making production and inventory decisions in a vacuum. Tell: operating targets conflict with the business strategy (price vs. service) that top management has set.
• Applying the wrong process lever. Tell: a line-balanced assembly operation is managed for machine utilization, or a process plant ignores downtime, while morale or bottlenecks go unaddressed.
• Spending precision on item-level cost estimates instead of the aggregate policy. Tell: long cost-measurement efforts that barely change the resulting decision rules.

• ▸Proposed change: describe it in one line (e.g., cut lot sizes, reduce raw material inventory).
• ▸Business strategy it must fit: state whether the firm competes on price, service, or another factor.
• ▸Process type and key lever: job shop, batch flow, assembly line, or process industry, and its main lever (utilization, balance, line balancing, downtime).
• ▸Expected inventory investment change: estimate the dollar change in total average inventory.
• ▸Expected cost effects: list ordering, unit cost, and expediting effects.
• ▸Expected service/revenue effects: list stockout risk and delivery-delay risk and which revenue lines they touch.
• ▸Target operating point: record the chosen aggregate point on the exchange curve and the implied A/r.
• ▸Review checkpoints: set dates for tracking finished-goods stock, stockouts, and expediting after the change (past the first few months).
• ▸Net effect on operating profit and shareholder value: state the expected direction and the reasoning.

1. 1Classify demand first, using the five time-series components: level, trend, seasonal, cyclical, irregular. Plot the item's history and mark which components are present. A constant series is level only. A positive or negative slope adds trend. A repeating annual pattern adds seasonality. Whatever remains after you remove the other components is irregular.
2. 2For any seasonal component, record its source. Natural forces (weather) and human decisions or customs (the school-year start that lifts department store sales) both produce predictable peaks. Predictable peaks and troughs are what justify anticipation stock rather than overtime.
3. 3Check whether the seasonality is on the supply side. Tomatoes ripen in a short window and are processed into ketchup sold all year. Arctic shipping is restricted to certain months. Strikes or war crises also make supply lower than demand. Each of these calls for anticipation inventory built ahead of the shortfall.
4. 4Characterize supply: lead time, transit time between adjacent locations, and reliability. Pipeline (WIP) inventory is proportional to usage rate times transit time, so compute it directly: usage rate × transit time. Direct-to-consumer firms such as Land's End and Dell have negligible pipeline inventory because they deliver directly to the consumer.
5. 5Place the item on the product-process matrix: job shop, batch flow, assembly, or process (continuous). Record the position-dependent facts: number of customers and products, degree of differentiation, flexibility in rate and output, equipment cost, maintenance style, and capacity-change lead time.
6. 6Read the position for operating consequences. Process industries run at full capacity (3 shifts, 7 days) and need highly reliable equipment. They do maintenance during extended shutdowns, and capacity changes take a long time. Job shops and batch flow use cheaper equipment and maintain as needed. Assembly needs tight coordination of many components and labor. Process industries face more natural variability in raw material characteristics, such as agricultural or mineral inputs.
7. 7Separate demand uncertainty from other drivers of stock. In the producer examples, 45% of finished goods inventory was held due to variability, and most of that was demand variability. Assume demand uncertainty is significant unless data shows otherwise, then size the forecasting effort accordingly.
8. 8Write the contextual conditions down as fixed 'givens' for the model. Then ask separately which givens (lead time, demand variability) could be changed, because changing the givens yields larger gains than optimizing within them.

Take ketchup. Supply is seasonal: tomatoes ripen over a short period. Demand is spread across the year. Supply is therefore lower than demand for much of the year, which is the textbook condition for anticipation stock. The processor builds inventory during the harvest window and sells it down through the year. The product-process position points the same way. Ketchup is a standardized, commodity-like product made by a flow process, so it runs near full capacity when operating, and capacity can't be adjusted quickly. Raw material (agricultural) also varies naturally in its characteristics. Together these conditions favor stock built ahead of need over chasing demand with production. A direct-mail firm such as Dell or Land's End sits at the opposite pole on pipeline stock: it delivers directly to the consumer, so its pipeline inventory is negligible. The conditions, not a generic formula, determine which stock types matter.

• If demand is regularly lower than average in part of the year and higher in another, then build anticipation stock to stabilize the production rate rather than working overtime in the plant.
• If supply is seasonal or subject to expected disruptions (harvest windows, restricted shipping seasons, strikes, war crises), then plan anticipation inventory sized to the period when supply rate is below demand rate.
• If the item moves between echelons or stations, then size pipeline stock as usage rate × transit time. If you deliver directly to the consumer, expect it to be negligible.
• If the item is in a process (continuous) setting, then expect full-capacity operation, shutdown-based maintenance, long capacity-change lead times, and natural raw material variability, and plan around them. If it is in a job shop or batch flow, expect flexible rates and as-needed maintenance.
• If you have accurately kept demand history for an item or group, then use time-series methods, with human judgment layered on top. If you have firm orders, planned shipments, or promotions, fold those in as informed judgment.

• Treating all demand as level. The tell: forecast errors that are persistently one-signed in the same months each year, which signals an unmodeled seasonal or trend component.
• Using overtime or capacity chasing against a predictable seasonal peak. The tell: recurring peak-season overtime with no pre-built stock.
• Ignoring supply-side seasonality or expected disruptions. The tell: stockouts or idle plants whenever harvest, shipping season, or a strike changes the supply rate.
• Applying job-shop assumptions (flexible rate, as-needed maintenance) to a process-industry item. The tell: plans that assume quick capacity changes or ad hoc downtime where the process runs 3 shifts, 7 days and needs scheduled shutdowns.
• Optimizing within the givens and never questioning them. The tell: sophisticated models tuned to existing lead times and demand variability while no one has asked whether those lead times or variability could be reduced.

• ▸Item / item group: name the item and its echelon or work station.
• ▸Demand components present: mark level, trend, seasonal, cyclical, irregular, with one line of evidence each.
• ▸Seasonal source: natural forces, human decision or custom, or none; give the timing.
• ▸Supply pattern: lead time, transit time between locations, reliability, any seasonal or expected disruptions (harvest, shipping window, strike).
• ▸Pipeline stock estimate: usage rate × transit time.
• ▸Process type: job shop, batch flow, assembly, or process; note equipment cost and maintenance style.
• ▸Capacity flexibility: rate and output flexibility, and the lead time to change capacity.
• ▸Number of customers / products / differentiation: record the market characteristics at this process position.
• ▸Historical data quality and judgmental inputs: note whether demand history is accurate and what firm orders, promotions, or planned shipments apply.
• ▸Givens that could be changed: list lead time and demand variability items to challenge before optimizing.

1. 1Classify the forecast by scope and horizon first. The source separates aggregate, medium-range forecasts (regression procedures, section 3.3) from individual-item, short-term forecasts (moving average and exponential smoothing, section 3.4). Pick the family before you pick a technique.
2. 2Decompose the series using the components of time-series analysis (section 3.1). Look for level, trend and seasonality, and state which are present. This decides whether a simple model is enough or you need the trend or seasonal variant.
3. 3Match the model to the components. Use the simple moving average (3.4.1) or simple exponential smoothing (3.4.2) when there is no trend or seasonality. Use exponential smoothing for a trend model (3.4.3) when trend is present. Use the Winters procedure (3.4.4) when the series is seasonal.
4. 4Follow the chapter's three-step sequence for statistically forecasting a time series (section 3.2). Set the smoothing constants deliberately (3.4.5) rather than accepting a default.
5. 5Screen anomalous demand (3.6) before it enters the model, and decide in advance how an outlier is handled. Otherwise one unusual period contaminates every later forecast.
6. 6Measure performance with forecast-accuracy measures (3.5.1). Estimate the standard deviation of forecast errors over the lead time (3.5.2), not just one period ahead, because inventory decisions depend on lead-time error. Monitor bias (3.5.3) separately from accuracy.
7. 7Define corrective actions (3.5.4) and an error distribution (3.5.5) up front, so a bias signal triggers a known response instead of a debate.
8. 8Layer in human judgment only through the guidelines in 3.7. Name the factors where judgment is needed (3.7.1), then record and monitor each judgmental input (3.7.2).
9. 9Route special items out of the standard path: limited history (3.8.1), intermittent or erratic demand (3.8.2), replacement or service parts (3.8.3), and terminal demand (3.8.4). Review the whole procedure on both statistical accuracy and strategic issues (3.9).

The passages give a table of contents, not a numerical case, so this example shows the routing the chapter implies and uses no invented data. Take a fast-moving item with a steady upward drift and a yearly selling peak. Section 3.1 says to separate level, trend and seasonality. All three are present, so the simple moving average and simple exponential smoothing are out, and the Winters seasonal procedure (3.4.4) is the fit. You set its smoothing constants under 3.4.5, then track accuracy (3.5.1) and the lead-time error standard deviation (3.5.2). When the bias monitor (3.5.3) shows forecasts running consistently one-sided, you apply a corrective action from 3.5.4 rather than waiting. A one-time promotion spike is handled as anomalous demand (3.6) so it does not enter the seasonal estimate. A product-launch item with a few weeks of sales never enters this path. It goes to 'items with limited history' (3.8.1), and a spare part for a discontinued product goes to terminal demand (3.8.4).

• If the forecast covers an aggregate group over the medium range, then use aggregate methods such as regression procedures. If it covers one item over the short term, then use moving average or exponential smoothing.
• If the series shows trend, then use the trend model. If it also shows seasonality, then use the Winters procedure. If neither, then use simple exponential smoothing or a simple moving average.
• If the item has limited history, intermittent or erratic demand, is a service part, or faces terminal demand, then do not run the standard smoothing routine. Use the special-class treatment in 3.8.
• If a demand observation is anomalous, then handle it explicitly (3.6) before it updates the forecast.
• If a judgmental adjustment is proposed, then confirm it falls under a factor where judgment is needed (3.7.1) and log it for monitoring (3.7.2). Do not accept it just because someone senior asked.

• Using one technique for every item. The tell: seasonal or trending items show persistent one-sided error while flat items look fine.
• Letting anomalous demand flow straight into the smoothing model. The tell: forecasts jump after a one-off event and the distortion persists for several periods.
• Judging quality only by one-period accuracy and ignoring the lead-time error standard deviation (3.5.2). The tell: the forecast looks accurate on paper but safety stock or service falls short.
• Tracking accuracy but not bias (3.5.3). The tell: errors are small yet consistently over or under, and nobody has a trigger for corrective action.
• Forcing intermittent, new, service-part or end-of-life items through standard statistical models. The tell: forecasts for sporadic or declining demand are meaningless or wildly unstable.
• Adding unmonitored judgment. The tell: manual overrides accumulate and nobody can say whether they improved the forecast.

• ▸Item or group and scope: state whether it is an individual item (short-term) or an aggregate group (medium-range).
• ▸Components observed: tick level, trend, seasonality and note the evidence for each (section 3.1).
• ▸Special class: tick limited history, intermittent/erratic, service part, terminal demand, or none (section 3.8).
• ▸Chosen method: moving average, simple exponential smoothing, trend model, Winters, or regression.
• ▸Smoothing constants and rationale: record the values chosen and how they were selected (3.4.5).
• ▸Anomaly handling rule: state how outlier periods are identified and treated (3.6).
• ▸Accuracy and lead-time error: record the accuracy measure used and the lead-time error standard deviation (3.5.1, 3.5.2).
• ▸Bias check and corrective action: state the bias monitoring method and the action it triggers (3.5.3, 3.5.4).
• ▸Judgment log: record each judgmental adjustment, its factor, owner and later accuracy effect (3.7).

1. 1Rank the SKUs first. For each SKU, record the unit value v and annual usage D, compute the product Dv, and sort all SKUs by Dv. This is the A–B–C classification, and it decides how much analytical effort each item gets. Spend the most on A items, a simple rule on C items.
2. 2Fix the definition of stock level before you set any trigger. Passage Section 6.1.1 flags that stock level has several definitions, so state which one the reorder point is compared against. Then record whether unmet demand is backordered or lost, because that changes the cost of a stockout.
3. 3Choose the review type per item class: continuous review (stock status known at every transaction) or periodic review (checked every R time units). The more important the item, the more the tighter monitoring of continuous review is worth paying for.
4. 4Choose one of the four control forms. (s,Q): when stock reaches the order point s, order a fixed quantity Q. (s,S): at the order point, order up to level S. (R,S): every R periods, order up to S. (R,s,S): every R periods, order up to S only if stock is at or below s.
5. 5Resolve the three issues a probabilistic-demand control system must answer: how often to review, when to order (the reorder point), and how much to order. Write all three down for each item so none is left implicit.
6. 6Set the reorder point so it protects over the replenishment lead time. Demand is uncertain during that interval, so the safety stock must cover it. Treat the safety stock as the policy's main tuning variable.
7. 7Pick one safety-stock rationale and state it. The options are a simple-minded approach, minimizing cost, a customer-service target, or an aggregate budget across items. Within cost or service, choose the specific rule: a safety factor k, a cost per stockout occasion (B1), a fractional charge per unit short (B2), per unit short per unit time (B3), or per customer line item short (B4). The service alternatives are a probability of no stockout per cycle (P1) or a fraction of demand filled from the shelf (P2).
8. 8Check the policy against the rest of operations strategy. Inventory rules should be consistent with the operations objectives and with the other management levers. Do not spend heavily on cost reduction where inventory is not the binding objective.
9. 9Keep using the policy as data improves. Real-time information reduces the demand variability you observe but does not remove it, so keep safety stock and review rules in place.

The source passages give no numeric inventory example, so this is a structural walk-through rather than a calculation. Start with the full SKU list. Compute Dv for each SKU, sort descending, and mark the top group A. Take one A item and state that stock level means the quantity compared with the order point, and that unmet demand is backordered. Assign continuous review with an (s,Q) rule, and set s to cover lead-time demand plus safety stock. Justify the safety stock with a P2 service target (fraction of demand filled from the shelf) or a B1 stockout cost. Take a low-Dv C item and give it a simple periodic (R,S) rule with a simple-minded safety stock. The passages also supply a real warning about consistency. Firms pushed quality programs while still rewarding workers for output volume, so the objective and the policy conflicted. In the same way, a tight service target on an item contradicts a policy that gives it little review attention or little safety stock.

• If an item ranks high on Dv (A class), then use continuous review and a carefully derived reorder point. If it ranks low (C class), then use a simple periodic rule with a simple-minded safety stock.
• If stockouts are backordered, then cost them through the B1–B4 charges. If they are lost sales, then treat the penalty as lost margin and set the service targets (P1, P2) more tightly.
• If you can state the cost of a stockout (per occasion, per unit, per unit-time, or per line item), then use the matching B-rule. If you can only state a customer promise, then use P1 (no stockout per cycle) or P2 (fraction filled from the shelf).
• If management sets a total inventory budget across items, then allocate safety stock on aggregate considerations rather than item by item.
• If the firm's operations objectives do not prioritize inventory cost, then do not spend large effort optimizing the policy. Choose a simple, consistent rule instead.

• Classifying by usage alone. The tell is that high-volume cheap items get the same scrutiny as costly ones. Use Dv (value times annual usage), not D.
• Setting the reorder point on average demand and ignoring lead-time protection. The tell is frequent stockouts just before replenishments arrive.
• Leaving the stock-level definition and the backorder-versus-lost-sales assumption unstated. The tell is that two planners compute different reorder points for the same item.
• Applying one policy type to every SKU. The tell is analyst time wasted on C items while A items run on crude rules.
• Treating real-time data as a cure for variability. The tell is that safety stock is cut after a data rollout and stockouts rise, because significant variability remains.
• Setting service or cost targets that conflict with other levers, as the quality-and-incentive case did. The tell is a policy that people are rewarded for ignoring.

• ▸SKU and unit value v: identify the item and its unit value.
• ▸Annual usage D and Dv: record annual demand and compute Dv to assign the A, B, or C class.
• ▸Stock-level definition: state which stock measure the order point is compared against.
• ▸Unmet demand treatment: backorder or lost sale, and why.
• ▸Review type and interval: continuous, or periodic with its review interval R.
• ▸Control form: (s,Q), (s,S), (R,S) or (R,s,S), with the values of s, Q or S.
• ▸Lead time and protection: record the replenishment lead time the reorder point must cover.
• ▸Safety-stock rationale: simple-minded, cost-based (k, B1–B4), service-based (P1, P2) or aggregate, with its target value.
• ▸Consistency check: note any conflict with operations objectives or other levers, and who resolves it.

1. 1Start with the context, not the technique. Write down the operations strategy as the source frames it: mission, objectives, and management levers (Ch. 1.4). Pick the planning system only after you know which objectives it must serve and which levers you can pull.
2. 2Fix the measures of effectiveness before choosing a system (Ch. 1.5). Agree on how aggregate inventory investment, service, and cost will be judged, so MRP, JIT, or OPT can be compared against the same yardstick.
3. 3Classify the production situation using the three frameworks in Ch. 2.5: where the product sits on its life cycle, which type of production process you run, and where you land on the product-process matrix. Treat this as a screen on which system can fit.
4. 4Account for the limits of human decision-making (bounded rationality, Ch. 2.2). Because SKUs are diverse, use conceptual and physical decision aids. Apply the A–B–C classification so tight, detailed control goes only to the items that justify it.
5. 5Choose a modeling strategy deliberately from the three in Ch. 2.7: (1) detailed modeling with analytic selection of a limited number of decision variables, (2) broader-scope modeling with less optimization, or (3) minimizing inventories with little modeling. State which one you are using and why.
6. 6Fix the forecasting input (Ch. 3). Managerial expectations about future demand drive every planning decision, so decide how demand is forecast and how forecast accuracy will be assessed before any schedule is generated.
7. 7Quantify costs. Use explicit cost measurement where costs can be found (Ch. 2.9). Where they cannot, use implicit cost measurement and exchange curves (Ch. 2.10) to show the trade-off between, for example, inventory and service.
8. 8Run the change as a structured study (Ch. 2.11): consideration, analysis, synthesis, choosing among alternatives, control, evaluation. Plan for transient effects during the switch and verify records with physical stock counts before going live.

The supplied passages give frameworks and chapter structure rather than a numerical MRP, JIT, or OPT case, so this example uses only the one concrete point the text makes. The authors warn that some analysts optimize given the lead times and demand variability the firm observes. They reach high efficiency within those givens but miss that the gains from changing the givens are far larger. Apply this to a planning-system choice. A planner builds a detailed schedule model (strategy 1 in Ch. 2.7) and tunes lot sizes carefully. First check whether the lead times and demand variability feeding the model can themselves be reduced. If they can, a broader-scope or inventory-minimizing approach (strategies 2 or 3) may serve the stated objectives better than refining the model. Place the plant on the product-process matrix first, then decide how much modeling effort is justified. The test is whether the system is optimizing the givens or questioning them.

• If you cannot state the operations strategy (mission, objectives, levers), then do not select a planning system yet. Define the strategy first.
• If the lead times and demand variability you plan around are treated as fixed, then ask whether changing them would yield more than optimizing within them.
• If the item is a low-value C item, then use simple aids and light control. If it is an A item, then justify detailed modeling and close attention.
• If costs cannot be measured explicitly, then use implicit cost measurement and exchange curves rather than inventing precise figures.
• If the product is at a different life-cycle stage or process type than the system assumes, then revisit the choice using the product-process matrix before implementing.

• Optimizing inside inflexible givens: a refined schedule is built on lead times and demand variability nobody has challenged. The tell is high model efficiency alongside persistent long lead times.
• Picking a system by name or fashion rather than fit: no one can explain where the plant sits on the product-process matrix or its life-cycle stage.
• Over-modeling every item: the same detailed treatment is applied to all SKUs, ignoring bounded rationality and A–B–C priorities. The tell is planners overwhelmed by exceptions on trivial items.
• Ignoring forecast quality: the plan is treated as exact while the demand forecast is never assessed. The tell is frequent rescheduling after demand misses.
• Skipping the study discipline: no control or evaluation phase, no allowance for transient effects, and no stock-count reconciliation. The tell is a system that looks correct on paper while records disagree with the shelves.

• ▸Mission and objectives: state the operations strategy in one or two lines.
• ▸Measures of effectiveness: list the inventory, service, and cost measures that will judge the system.
• ▸Product life-cycle stage: record where each major product sits.
• ▸Process type and product-process matrix position: record the process type and its place on the matrix.
• ▸Modeling strategy chosen: pick one of the three in Ch. 2.7 and give the reason.
• ▸Givens to challenge: list the lead times and demand variability you accept, and whether they could change.
• ▸Item classification (A–B–C): note which items get detailed control and which get simple rules.
• ▸Cost measurement approach: mark each cost as explicit or implicit and name its source.
• ▸Study phase and control plan: record the current phase, how transient effects and stock counts will be handled, and the evaluation date.

1. 1Map the network before choosing a tool. List every stocking point and mark which of the three coordination contexts each sits in: (1) a family of items at one location sharing a supplier, transport mode, or production equipment; (2) several locations in one firm (multiechelon); (3) locations run by different firms. Each context has its own mechanism, and applying the wrong one is the commonest error.
2. 2Single stocking point: write down the shared resource that links the items (common supplier, common truck, common production equipment). Then list the advantages and disadvantages of coordinating (Ch. 10.1) before committing. Coordination is a trade-off, not a default.
3. 3Select the family model by demand type. Deterministic demand: use a family decision rule for replenishment quantities, with or without group discounts. Probabilistic demand: use a can-order (S,c,s) system or a periodic review system. A full-truckload or shipping-consolidation constraint is handled by the dedicated procedures in 10.5.1 and 10.7.
4. 4Multiechelon inside one firm: treat each downstream replenishment request as demand at the upstream stocking point. Do not forecast the upstream node independently. Choose the structure (serial or arborescent) and the demand type: level, time-varying, or probabilistic. For probabilistic demand, start from a base stock control system.
5. 5Decide where stock sits using the economic incentives to centralize stocks (11.6.1), the question of where to deploy stock (11.6.2), and whether lateral transshipments between locations are worth allowing (11.6.3).
6. 6Inside the firm, run Sales and Operations Planning as the cross-functional process that sets one shared plan. Add collaborative forecasting where an outside partner holds relevant demand information.
7. 7Between firms, first check for information distortion (12.1): do orders placed upstream vary more or differently than the demand downstream? If so, set up information sharing and consider Vendor-Managed Inventory, where the supplier manages the buyer's inventory.
8. 8Where incentives conflict, choose a contract type from the four in Ch. 12.4: wholesale price, buyback, revenue-sharing, or service-level agreement. Then test it against the implementation challenges (12.4.5) before signing.

The supplied passages give no numerical case, so this is a structural walk-through, not a calculation. A firm buys a product family from one supplier, runs a central warehouse feeding regional stocking points, and sells through an independent retailer. Step 1 classifies the situation into all three contexts. The family sharing a supplier goes to the Chapter 10 tools: with probabilistic demand, a can-order (S,c,s) or periodic review system is the candidate, and the 10.1 advantages and disadvantages are listed first. The warehouse-to-region link goes to Chapter 11: regional replenishment requests are the warehouse's demand, so the warehouse is set up with a base stock logic driven by those requests, and centralizing stock versus deploying it regionally is tested using 11.6. Internally, S&OP aligns the plan across functions. The retailer link is where Chapter 12 applies: you check whether the retailer's orders distort true demand, share data or use VMI if so, and choose among wholesale-price, buyback, revenue-sharing, and service-level contracts. The exact formulas for each step sit in the referenced sections; the point is that every node is assigned to one context and one tool.

• If items share a supplier, transport mode, or production equipment, then evaluate joint replenishment (Ch. 10) against its listed advantages and disadvantages; if they share nothing, manage them individually.
• If demand is probabilistic and items are coordinated, then use a can-order (S,c,s) or periodic review system; if demand is deterministic, use the family decision rule, adding the group-discount version when discounts apply.
• If one firm holds stock at several echelons, then treat downstream replenishment requests as upstream demand and use multiechelon methods (base stock for probabilistic demand) rather than independent item-by-item control.
• If the problem is cross-functional misalignment inside the firm, then use S&OP; if the problem is distorted or hidden demand information between firms, then use information sharing, collaborative forecasting, or VMI.
• If trading partners' incentives diverge, then choose a contract type from Ch. 12.4 and check the implementation challenges before adopting it; a service-level agreement is the option when the issue is performance level rather than price or revenue split.

• Forecasting upstream stocking points independently of downstream orders. The tell: upstream stock-outs and excess coexist while downstream locations report normal service.
• Coordinating a family of items without weighing the disadvantages of coordination. The tell: joint orders are larger or later than some items need, and individual items show shortages or excess.
• Using a bilateral contract or VMI to fix a problem caused by information distortion. The tell: order patterns upstream remain more erratic than end demand despite the agreement.
• Signing an incentive-aligning contract without testing implementation challenges. The tell: the contract exists on paper but partners do not change ordering behavior or dispute terms.
• Running S&OP as a reporting meeting instead of a coordinating process. The tell: functions keep separate plans and the meeting only reconciles numbers after the fact.

• ▸Stocking points involved: list every location and the firm that controls each.
• ▸Context (1/2/3): single-location item family, multiechelon within one firm, or multi-firm network.
• ▸Linking resource or relationship: the shared supplier, truck, equipment, upstream node, or trading partner that connects the nodes.
• ▸Demand type: deterministic (level or time-varying) or probabilistic.
• ▸Candidate mechanism: joint replenishment rule, can-order/periodic review, base stock multiechelon, S&OP, collaborative forecasting, or VMI.
• ▸Advantages vs. disadvantages of coordinating: one line each, written before adopting the mechanism.
• ▸Information distortion check: compare upstream order pattern to end demand and note the gap.
• ▸Contract type under consideration: wholesale price, buyback, revenue-sharing, or service-level agreement, with the incentive conflict it addresses.
• ▸Implementation challenges identified and the owner assigned to each.