Supply chain optimization through segmentation and analytics

1. 1List every data element the segmentation needs for each object you segment (product, customer, supplier), taken from the segmentation criteria you chose in the earlier steps. Do not audit all master data. Audit only what drives a segment decision.
2. 2Check that each element exists in the Master Data. Start with lead time: confirm it separates production lead time from logistics lead time (and any other lead-time components) instead of holding one lump number.
3. 3Test whether the stored values are real. Lead times often get plugged in only to satisfy the ERP system and have nothing to do with reality. Compare a sample against actual receipts or deliveries, and flag fields where every record carries the same value.
4. 4Review how each data element is calculated. Write down what the calculation includes and what it leaves out. Look specifically for cancelled orders and order postponements that were never recorded, because incomplete transaction sets understate or obscure true demand.
5. 5Plot the data and run statistical checks to find anomalies. Look for promotion and conference spikes that overstate normal sales activity. Distorted history shows up as unnecessarily high variability, which corrupts forecasts and safety stock calculations.
6. 6Check classification and completeness. Look for customer or part classifications that put items into inappropriate groups, and for incomplete customer records that push a customer into the wrong tier and give it lower service than intended.
7. 7Write a correction method for each error type (fill, reclassify, adjust or remove event-driven spikes, add missing transactions), then execute the cleansing.
8. 8Re-run the anomaly review on the cleansed data and repeat the analyze, correct, and execute loop until the data set is valid enough to use for segmentation. Treat this as a loop, not a one-time pass.

The source describes the typical failures rather than one named case, so this example applies them in sequence. A planner builds a product segmentation on demand variability and lead time. Step one lists the needed elements: order history and lead time. Step two shows the Master Data holds a single lead-time figure per part. Step three shows the figure does not split production lead time from logistics lead time, and many values were entered only to satisfy the ERP system, so delivery scheduling is hard. Reviewing how order history is calculated reveals that cancelled orders and postponements were never recorded. Graphing the history shows spikes in promotion and conference months that overstate normal sales. The planner flags that this distortion produces unnecessarily high variability, which would inflate the forecast and safety stock inputs and make them meaningless. The planner adjusts the spikes, reconstructs lead time by component, reclassifies mislabelled parts, and re-plots the data. The loop repeats until no unexplained anomalies remain, and only then does segmentation begin.

• If a lead-time field is a single number with no production versus logistics split, then treat it as missing data and rebuild it by component before using it for scheduling or segmentation.
• If many records share an identical lead-time value, then suspect an ERP plug and verify against actual receipts.
• If order history shows spikes aligned with promotions or conferences, then separate those events from normal demand before computing variability, forecasts, or safety stock.
• If cancelled orders and postponements are not captured, then the history is incomplete; add them before calculating demand patterns.
• If a customer's record is incomplete and the tier looks too low, then fix the data and re-tier before concluding the customer deserves less service.
• If the re-review after cleansing still shows anomalies, then return to the analysis step and repeat rather than proceeding to segmentation.

• Treating cleansing as a one-pass task. The tell: no re-review of the cleansed data, and anomalies reappear in the segmentation output.
• Accepting ERP-plugged lead times as real. The tell: identical lead times across dissimilar parts and repeated delivery scheduling misses.
• Leaving promotion and conference spikes in the history. The tell: unusually high variability, inflated safety stock, and forecasts that cannot be trusted.
• Recording only completed orders. The tell: cancelled and postponed orders are absent, so recorded demand looks cleaner and different from true demand.
• Segmenting on incomplete or misclassified master records. The tell: customers or parts landing in tiers that conflict with what the business knows about them, and service lower than intended.

• ▸Data element: name the field used in segmentation (e.g., lead time, order history, customer classification).
• ▸Segment object: product, customer, or supplier the element applies to.
• ▸Exists in Master Data (Y/N): confirm the field is present and populated.
• ▸Calculation review: list what the calculation includes and what is missing (e.g., cancellations, postponements).
• ▸Anomaly found: describe the spike, gap, plugged value, or misclassification, and how you detected it.
• ▸Cause: missing data, inaccurate data, or incomplete data.
• ▸Correction method: state the fix (rebuild by component, adjust event spike, reclassify, complete record).
• ▸Re-review result: record the outcome after re-plotting the cleansed data.
• ▸Iteration count and sign-off: number of loops run and who approved the data as valid for segmentation.

1. 1Tag every SKU with a lifecycle stage (NPI/launch, growth, maturity, decline, EOL) as a field in the segmentation data. Stage is a planning attribute, not a label in a product deck. The source's premise is that no one-size-fits-all planning approach exists, so stage becomes one of the segmentation criteria.
2. 2Run the tagging separately for supply-side SKUs (incoming materials) and demand-side SKUs (finished goods). The source says these are two segmentation exercises with different goals: supply-side cost and consolidation, demand-side customer responsiveness and finished-goods minimization. A component can sit at a different stage than the product it feeds.
3. 3For NPI items, expect no demand history for similar products. Do not forecast from isolated functional opinions. The source names siloed stakeholders and missing history as what makes NPI forecasts so hard. Build one cross-functional 'single number forecast' (game plan) that every supply chain partner subscribes to, and review it through S&OP.
4. 4For NPI, set the supply-chain variables deliberately instead of inheriting defaults: batch sizes, EOQs, MOQs, and lead times. The source says optimizing inventory requires reassessing exactly these variables.
5. 5For maturity, shift the objective to price and delivery competitiveness, which the source calls extremely important for mature-stage products. Use the integrated-supply-chain levers it lists for this: lower safety stock, shorter cycle times, fewer intermediaries, and reduced inventories.
6. 6For decline and EOL, plan the run-down explicitly. Demand erodes as products approach EOL, and slow-moving inventory strands in the supply chain. Re-check EOQs, MOQs, and batch sizes against the falling demand, and apply an EOL planning method (the source treats EOL planning as its own key concept).
7. 7Re-tag stages on a recurring S&OP cadence. In high-tech, lifecycles and technology cycles keep shortening, so a stage assignment goes stale quickly. Treat a stage change as the trigger to switch the planning method and parameters for that SKU.
8. 8Validate the segmentation after building it: run and test the logic, confirm that each stage maps to a defined planning and scheduling strategy, and check that no SKU lacks a stage.

The source's case is F.A. Lambay's Wipro article on new product introductions in high-tech. Lifecycles and technology cycles keep shortening. NPI forecasts are hard because stakeholders work in silos and no history exists for similar products. At the other end, products approaching EOL see demand erode, and slow-moving inventory becomes stranded in the supply chain. Lambay's remedy has three parts. First, reassess batch sizes, EOQs, MOQs, and lead times. Second, strengthen the S&OP process so it responds to rapidly changing demand. Third, use analytics together with a single-number forecast that all supply chain partners subscribe to. Applied to your own SKUs: a launch item gets a collaboratively built single forecast and deliberately set lot sizes and MOQs. The same SKU years later gets run-down lot sizing so it does not strand stock. The product is the same; the stage changes the method. (The passage excerpt is partial, so check the full article for its specific results.)

• If a SKU is at NPI and has no history from similar products, then build a single-number, cross-functional forecast through S&OP and do not rely on any one stakeholder's view.
• If a SKU is mature, then optimize for price and delivery competitiveness, and cut safety stock, cycle time, and inventory through supply chain integration.
• If a SKU is approaching EOL, then expect eroding demand, re-examine EOQs, MOQs, and batch sizes, and use EOL planning to avoid stranded inventory.
• If the product is in a high-tech or fast-turnover category, then shorten the stage-review interval, because lifecycles are shrinking.
• If a SKU appears on both the supply and demand sides, then segment it separately on each side and let the two planner groups use different tools.

• Applying one planning method to every stage. The tell: launch items and late-life items use the same lot sizes, safety stock logic, and forecast approach.
• Forecasting NPIs from siloed stakeholders. The tell: sales, operations, and supply partners each hold a different demand number for the same new item.
• Leaving EOQs, MOQs, and batch sizes unchanged as demand erodes. The tell: slow-moving or stranded inventory on products near EOL.
• Running one segmentation for both procurement and delivery SKUs. The tell: supply-side planners are given customer-responsiveness targets, or demand-side planners are given consolidation targets.
• Setting stage once and never revisiting it. The tell: a 'growth' SKU whose plan parameters still reflect launch, in a category with shortening lifecycles.

• ▸SKU / family and side (supply or demand): record which segmentation structure the item belongs to.
• ▸Current stage: NPI, growth, maturity, decline, or EOL.
• ▸Date of last stage review and next review: set the cadence for re-tagging in S&OP.
• ▸Forecast source and history available: state whether similar-product history exists and who contributes to the single-number forecast.
• ▸Primary objective for this stage: for example, collaborative forecast accuracy at NPI, price and delivery competitiveness at maturity, or run-down at EOL.
• ▸Lot-sizing parameters: current batch size, EOQ, MOQ, and lead time, with the date each was last reassessed.
• ▸Safety stock and inventory stance: the target level and its rationale for this stage.
• ▸Stranded-inventory exposure: identify slow-moving stock and the demand-erosion signal.
• ▸Planning method assigned: the segment's planning/scheduling strategy, including whether an EOL plan applies.

1. 1Pull order history for the SKU set you are segmenting. Before computing anything, run the source's data-cleansing checks: confirm the needed data elements exist in Master Data, review how each one is calculated, and chart the data to spot anomalies.
2. 2Find the distortions that inflate variability. The source names three: cancelled orders not recorded, order postponements not recorded, and promotions or conferences that overstate normal sales activity. Correct or flag each one. Distorted data reports unnecessarily high variability, which makes forecasts and safety stock meaningless.
3. 3Repeat the cleansing loop (analyze, reduce errors, execute, re-review) until the data is valid. Compute variability only on the cleansed set.
4. 4Calculate the coefficient of variation (CoV) per SKU. Record it next to the product's lifecycle stage. The source ties variability to stage: high and inconsistent at introduction, minimal at growth/maturity, intermittent at end of life.
5. 5Treat CoV as one segmentation factor alongside service level, forecast reliability, inventory stocking rules, and shelf life. Do not segment on CoV alone. The source says the best structure comes from analytics across these factors together.
6. 6Set the CoV cut points from your own distribution. The source gives no thresholds, so do not borrow any. Document the breakpoints so segment membership can be audited.
7. 7Map a planning and buffering strategy to each segment. High-variability segments get high safety stock, a high service level, and analytics-heavy forecasting. Stable segments get minimal inventory and easy service. Declining or intermittent segments get tight inventory control to avoid obsolete stock.
8. 8Re-run the CoV after promotions, lifecycle transitions, and data fixes. A SKU's segment moves as its stage moves.

The source walks a high-tech product through its lifecycle, and variability drives the planning choices at each stage. At introduction, demand is variable and inconsistent, so forecasting is hard. The planner estimates sales with tools like S&OP, overstocks inventory in hope of a surge, sets a high service level, and holds high safety stock to buffer variability. Very high early margins pay for the excess. In growth and maturity, volume variability is at its minimum and reasonably predictable. High service is attainable at minimal inventory, and the product is a cash cow. In decline, demand is falling and often intermittent. Margins are at a minimum, service levels fall, and the risk shifts to obsolete inventory. Same SKU, three CoV profiles, three different buffering policies. A separate warning applies to the inputs: unrecorded cancellations and promotion spikes inflate measured variability, so a mature SKU can look volatile and be over-buffered.

• If a SKU is in introduction and its demand is variable and inconsistent, then carry high safety stock and a high service level, and put it under analytics-based forecasting. Early margins absorb the cost.
• If a SKU is in growth or maturity and variability is at its minimum, then hold minimal inventory. Service is easy to hit, so extra buffer is waste.
• If demand is declining or intermittent at end of life, then cut buffering and plan inventory carefully. Maintaining stock and service often costs more than the margins earned.
• If a stable-looking SKU shows high CoV, then check for unrecorded cancellations, postponements, and promotion or conference spikes before changing safety stock.
• If two SKUs share a CoV but differ in service level, forecast reliability, or shelf life, then do not put them in the same segment automatically. Segmentation weighs all of these factors together.

• Computing CoV on raw order history. The tell: high variability on SKUs with a known promotion or conference, and safety stock that looks inflated.
• Omitting cancelled and postponed orders from the data. The tell: the order record is cleaner than what customers actually did, and the forecast misses.
• Segmenting on variability alone. The tell: SKUs with very different service targets or shelf life sit in one segment and get the same policy.
• Leaving a SKU in its introduction-stage buffer after it matures. The tell: high safety stock persists while demand has stabilized, and inventory stays overstocked.
• Missing the end-of-life shift. The tell: intermittent demand, shrinking margins, and growing obsolete stock because the buffer was never reduced.

• ▸SKU / product family: identify the item being assessed.
• ▸Lifecycle stage: introduction, growth/maturity, or decline/EOL.
• ▸Data-cleansing status: note cancelled orders, postponements, and promotion/conference spikes found and corrected.
• ▸Cleansed history window: state the period and number of order periods used.
• ▸CoV (calculated on cleansed data): enter the value.
• ▸Segment assigned and cut-point used: record the breakpoint, set from your own distribution.
• ▸Other segmentation factors: service level, forecast reliability, stocking rules, shelf life.
• ▸Buffer policy: safety stock level and target service level for this segment.
• ▸Re-review trigger: stage change, new promotion, or data correction that forces recalculation.

1. 1Fix the goal before you assign any touch. The source says the first and most important step in segmentation is identifying the goal (customer service, shipping cost, inventory reduction, being competitive). Write the primary goal and the secondary goal in one line each. Touch allocation follows from them.
2. 2Split the SKU universe by side before you segment. In manufacturing, supply-side SKUs (incoming materials) and demand-side SKUs (finished goods) are separate populations with different goals. Run two segmentation exercises and assign supply planners and demand planners separately. Retail and distribution use one SKU set, so one exercise is enough.
3. 3Place each SKU in a lifecycle stage: Launch/NPI, Active/Growth, Stable/Maturity, or Declining/EOL. Lifecycle stage is the main driver of touch. Check it against the other drivers the source names: service level, demand variability, forecast reliability, inventory stocking rules, and shelf life.
4. 4Assign a touch level to each stage. NPI gets the most intensive monitoring, because demand trend, pricing, competition, and service level are all uncertain, and analytics is used heavily. Growth and Stable get low touch: ERP forecasting and inventory optimization with minimal planner intervention. Declining/EOL brings planners back to repurpose and redeploy capacity and inventory, with the aim of avoiding a buildup of obsolete inventory.
5. 5Map a planning and scheduling tool to each segment on a characteristics chart. Mark each characteristic (Min/Max, OIL, 2-Bin, EOQ, S&OP, ATP/CTP, BTO/MTO, manual vs. automated processes) as applicable, somewhat applicable, or not applicable. A segment that lands on automated tools such as Min/Max, OIL, or 2-Bin is a low-touch segment. A segment with poor data and unstable history, supported by S&OP, is a high-touch (Custom) segment.
6. 6Set a target touch distribution and publish it. Write down the share of planner effort you expect per segment, next to the expected revenue and inventory share. Include 'Number of Touch Points' as a metric.
7. 7Run the first design against last year's data and iterate. The source says tool selection may not be right the first time and that analytics is what resolves it. Test whether the segments you marked as automated actually behave predictably, and move any that do not into a touched segment.

The source's aerospace case has six segments (Charts 6.13 and 6.14). Different supply chain tools are assigned to each segment, based on the characteristics identified for it. The goal-mapping chart shows the target distribution by touch level of planners. The company expects most planner effort in Segments 1 and 4 and expects automation to be the primary driver for Segments 2, 3, 5, and 6. Segment 1 is new products and Segment 2 is stable products. Most revenue is expected from Segments 1 and 2, and most inventory is expected there as well. Segment 1 is therefore both the highest-revenue and a high-touch segment, which is where planner attention is worth spending. Segment 2 carries similar revenue but runs on automation. The design principle is that revenue and touch are decoupled: stability, not revenue size, decides how much a planner touches a segment.

• If a product is in Launch/NPI and demand, pricing, or competition are uncertain, then assign product planners to monitor it intensively and use analytics. Do not put it on an automated stable-demand tool.
• If a product's cycles and demand are repetitive and predictable (Growth or Maturity), then automate it with ERP forecasting and inventory optimization, and design a 'low touch' process with minimum planner intervention.
• If a product enters decline or EOL, then bring planners back to repurpose capacity and inventory and avoid obsolete stock.
• If a segment has poor data, unstable transaction history, and weak predictability, then treat it as a Custom supply chain and support it with S&OP and planner involvement, not full automation.
• If the target distribution puts high revenue in a segment, then do not assume it needs high touch. Check stability first, as in the aerospace case, where stable Segment 2 carries revenue but runs on automation.

• Spreading touch evenly across all SKUs. Tell: planners spend time on stable replenishment items while NPI and EOL items are reviewed late, and one tool or one process applies to everything (the 'one size fits all' trap).
• Treating supply-side and demand-side SKUs as one population. Tell: a single segmentation with a single touch policy for both, even though the goals differ (cost and consolidation vs. responsiveness and finished-goods minimization).
• Automating NPI because the tools exist. Tell: forecasts for a new product run unreviewed through ERP while demand, pricing, and competition are still uncertain.
• Forgetting that EOL needs planners again after the automated middle stages. Tell: declining items remain on automated replenishment and obsolete inventory builds up.
• Setting touch levels once and never testing them. Tell: no validation against historical data, and segments marked 'automated' keep generating manual expedites and corrections.

• ▸Segment name and side (supply or demand): name the segment and say which side it belongs to.
• ▸Segment goal: state the primary and secondary goal (e.g., customer service, cost, inventory, competitiveness).
• ▸Lifecycle stage: NPI, Growth, Maturity, or Decline/EOL.
• ▸Uncertainty drivers: rate demand variability, forecast reliability, and data quality, plus shelf life if it applies.
• ▸Planning and scheduling tool: list the tools from the characteristics map (e.g., Min/Max, OIL, 2-Bin, S&OP, BTO/MTO).
• ▸Touch level: high, medium, or low, with the named planner role responsible.
• ▸Target share of planner effort, revenue, and inventory: enter an expected percentage for each.
• ▸Planner trigger for intervention: define the event that moves an automated segment to manual review.
• ▸Validation result: record how the touch design performed against last year's data and what you changed.

1. 1Name the single primary goal before touching any numbers. The source asks: reduce shipping cost, improve service on urgent items, or reduce inventory through better planning? The goal sets the segmentation criteria, so write it down first.
2. 2Name the secondary constraint that stops the primary goal from running away. In the mail-order pharmacy, the primary goal (right product, right customer, on time) is already met by overnighting everything. The secondary goal, being competitive, is what exposes the cost problem.
3. 3Segment the items on the factors the source lists: service level, demand variability, forecast reliability, inventory stocking rules, and shelf life. Group items that share a profile, and do not give every SKU its own rule or every SKU the same rule.
4. 4In manufacturing, run two segmentation exercises: one for the procurement (supply) SKUs and one for the delivery (demand) SKUs. Retail and distribution use one SKU set on both sides, so one exercise is enough.
5. 5Convert inventory time into money with a carrying-cost rate. Cost per unit = (days tied up ÷ 365) × holding-cost rate × unit value. Compute it for your current cash-to-cash days and for your target.
6. 6Assign a planning method to each segment from the SCS toolset: statistical safety stock, optimal service level, coefficient of variation, reorder point cycle, EOQ, optimal inventory level, min/max, build-to-order, or end-of-life planning. Match the method to the segment's variability and shelf life.
7. 7Treat stock-outs as a service-level decision per segment. Distinguish urgent from non-urgent items (for example, a drug running low at a customer location) so only the urgent ones get premium handling.
8. 8Run, test, and validate the segmentation, then track inventory holding cost, stock write-offs or discounts, and service reliability together. Judge success on the combination, not any one of them.

The source's computer-industry case shows what days of inventory and cash tied up are worth. The industry-average cash-to-cash cycle is 106 days. Best-of-class is 21 days, a gap of 85 days. For a $1,000 computer with a 7 percent holding cost, 85 ÷ 365 × 0.07 × $1,000 = $16.30 of extra cost per unit. Dell runs at minus 7 days because customers pay over the Internet before Dell buys materials. Against the industry average that is a 113-day advantage, or 113 ÷ 365 × 0.07 × $1,000 = $21.67 per unit. Dell also holds no finished-goods stock waiting for orders. The source flags obsolescence as the added risk in a volatile industry like computers. Use the same arithmetic on your own segments. Convert days held into dollars per unit, then compare segments to decide where tighter stock pays most.

• If the goal is reducing inventory, then set segment criteria around planning accuracy (demand variability, forecast reliability). If the goal is service, then weight urgency and service level instead.
• If every order is being treated as urgent because nothing distinguishes them, then segment by urgency and apply premium handling only to the urgent segment.
• If you are a manufacturer, then segment supply-side and demand-side SKUs separately. If you are a retailer or distributor with one SKU set, then segment once, focused on the customer.
• If the product is volatile or quickly obsolete (the computer example), then prioritize shortening days held and cash-to-cash time over buffer stock.
• If a segment has a short shelf life or is approaching end of life, then use end-of-life or build-to-order planning rather than standing safety stock.

• Hitting the service goal with no cost counterweight. The tell: the pharmacy overnights everything, service is perfect, and cost threatens solvency.
• Standardizing one rigid policy across hundreds or thousands of customers and items for the sake of consistency. The tell: individual customers and SKUs collapse into one list, and urgent and routine items get the same treatment.
• Reading inventory in isolation from cash tied up. The tell: you quote turns or stock levels but never the days of financing or the per-unit carrying cost (such as $16.30 per $1,000 unit).
• Using one segmentation for both procurement and delivery in a manufacturing setting. The tell: the segments fit finished goods but make no sense for purchased materials, or the reverse.
• Ignoring obsolescence and write-offs when judging efficiency. The tell: holding cost falls but stock write-offs or discounts rise.

• ▸Primary goal: state in one line (reduce shipping cost, improve service, or reduce inventory).
• ▸Secondary constraint: state the limit on the primary goal (for example, being competitive on cost).
• ▸Segment name and SKU/item scope: say whether it is supply-side or demand-side.
• ▸Segment profile: service level, demand variability, forecast reliability, stocking rule, shelf life.
• ▸Current days tied up (cash-to-cash or days held): enter the measured figure.
• ▸Carrying-cost rate and unit value: enter the annual holding-cost % and the value per unit.
• ▸Cost per unit now vs target: (days ÷ 365) × rate × unit value, for each case.
• ▸Planning method assigned: safety stock, reorder point, EOQ, min/max, build-to-order, or end-of-life.
• ▸Service-level and urgency rule: define what counts as urgent and how it is handled.
• ▸Tracked outcomes: holding cost, write-offs or discounts, service reliability, with the review date.

1. 1Define the outcome as three numbers you track together: on-time delivery percentage, service level, and customer satisfaction. The definition is the right product to the correct customer on time. A shipment that is on time but wrong, or right but late, counts as a miss.
2. 2List how your customers differ before you set any target. The source names four differences: location (one customer is farther away), timing (delivery matters more to one customer), quality requirement, and personality or compatibility. Record which of these applies to each customer.
3. 3Group customers and products into segments instead of managing each individually or treating all the same. At hundreds or thousands of customers, individuals become 'a number on a list', and the usual reaction is rigid, uniform policies. Segments keep standardization where customers share characteristics and differentiate where they don't.
4. 4Check which customers get attention for the wrong reasons. The source notes that big buyers and frequent complainers are treated differently by default. Replace that habit with explicit segment membership based on the differences you listed.
5. 5Assign each segment its own planning and scheduling strategy and service target. The book's toolkit for this includes Optimal Service Level, statistical safety stock, reorder point, EOQ, min/max, build-to-order/make-to-order, and end-of-life planning. Pick the tool per segment, not one tool for everyone.
6. 6Control order-processing accuracy at the source. The source says an incorrect product definition, shipment date, or materials list can delay completion or cause an incorrect product to ship to a customer. Verify these three data items on every order before release.
7. 7Treat people, process flow, and information movement as one system. The source says a failure in any of the three critical resources is a failure of all. When on-time delivery drops, check all three before blaming one.
8. 8Account for outside forces that distort performance: holidays and work schedules, technology availability and change, risk, and budget. Build them into the plan rather than explaining misses afterward.
9. 9Cleanse the data, then build, run, test, and validate the segmentation structure before rolling it out. This follows the book's sequence of analysis, then build, then execute.

The source traces how service breaks down as a business grows. With one customer, one product, and one supplier, segmentation is meaningless. Add a second customer and the two already want different support, such as a farther location, more critical delivery timing, or a tighter quality requirement. At dozens of customers, big buyers become preferred and frequent complainers get more attention, so service is differentiated by accident. At hundreds or thousands, customers become 'a number', and the organization imposes rigid, uniform policies in the name of fairness and efficiency. Add more products and suppliers and the complexity becomes overwhelming. The corrective move is to group customers by the differences that matter (location, delivery criticality, quality requirement), set a delivery and service target for each group, and choose a planning method for each. Then measure on-time delivery, service level, and satisfaction by segment, not as one blended average.

• If you have one customer, one product, and one supplier, then skip segmentation. Once a second customer expects different support, start differentiating.
• If a customer's delivery timing is more critical than another's, then give it a separate segment and a tighter on-time target. Don't average it into a company-wide figure.
• If customers number in the hundreds or thousands, then segment them by shared characteristics. Don't impose one rigid policy, and don't let complainers or large buyers set priorities by default.
• If an order is shipped wrong or late, then check the product definition, ship date, and materials list first. Then check people, process, and information flow together, because a failure in one is a failure of all.
• If a segment's service target is high, then pair it with a planning method suited to it, such as safety stock, build-to-order, or min/max. Choose the method by segment, not by habit.

• Treating all customers the same in the name of fairness. The tell is one policy and one service target for customers with different locations, timing needs, and quality requirements.
• Letting attention follow complaints and size. The tell is that service levels track who calls loudest or buys most, not a defined segment.
• Reporting only a blended on-time percentage. The tell is a healthy average while a time-critical customer group is repeatedly late.
• Ignoring order-data accuracy. The tell is wrong products shipped, or delays traced to an incorrect product definition, ship date, or materials list.
• Optimizing one movement in isolation, such as people, process, or information, and ignoring outside forces. The tell is that an improvement in one area is followed by a decline in another, or by seasonal and schedule-driven misses nobody planned for.

• ▸Segment name: label the customer/product group (e.g., by location, timing criticality, or quality requirement).
• ▸Differentiator: state which difference defines the segment: location, delivery timing, quality requirement, or other.
• ▸Customers/products included: list who belongs in the segment so membership is explicit.
• ▸On-time delivery %: record actual on-time delivery for this segment alone, not the company blend.
• ▸Service level target: set the target for this segment and note how it was chosen.
• ▸Customer satisfaction: record the satisfaction measure used for this segment.
• ▸Planning/scheduling method: name the approach assigned (e.g., safety stock, min/max, build-to-order).
• ▸Order-data check: confirm the product definition, ship date, and materials list are verified before release.
• ▸Outside forces this period: note holidays, work schedules, technology changes, risk, or budget effects on performance.

1. 1Map the three flows first: materials, money, and information. List every step from raw material to customer for your product family and mark where each flow can fail (wrong inventory record, wrong payment amount, wrong ship date). The passages treat a failure in any one flow as a failure of the whole chain, so profitability losses start here.
2. 2Baseline total supply chain cost as a share of revenue. HP's Dick Conrad puts it at 65 cents of every revenue dollar. Compute your own figure, because every improvement is measured against it, and a number this large means small percentage gains move margin.
3. 3Segment products and customers before optimizing. Dell and Walmart are cited as cases where segmentation is managed so it directly affects profitability. Group by the dimensions the passages flag (demand pattern, supplier region, quality expectation, financial capability) and set a different policy per segment instead of one policy for all.
4. 4Build the analytics backbone for each segment. Place your current reporting on the BI ladder: standard reporting, ad hoc, drill-downs, alerts, statistical analysis, forecasting, predictive modeling, optimization. The first four describe the past or present. Move the segment-critical decisions (safety stock, batch size, forecasts, planning and scheduling) into the top four, which address Why and What is Next.
5. 5Convert the cost reductions into cash flow. Compare the payment term with the time to sell. In the Walmart case, a net-30 payment term against a sell-through under 30 days gives positive cash flow. Compute this gap per segment and target the segments where you can shorten inventory-to-sale time below the payment cycle.
6. 6Strip non-value-added controls. List each approval and inspection step and ask what defined standard it checks against. The pharmaceutical cleaning case had four layers of inspection but no documented definition of a clean machine. The oil and gas firm required 16 signatures per purchase order. Remove or define any control without a standard.
7. 7Quantify international exposure on margin. For each foreign supplier or market, test how a small exchange-rate move changes transaction profit. Decide whether to standardize on a stable currency, knowing that it can push the local manufacturer's costs up.
8. 8Track outcomes with KPIs tied to business objectives and review them against the baseline. Report cost-to-revenue, cash cycle, margin, and delivery and price competitiveness per segment, and check that each BI objective maps to a business objective.

Walmart manages its logistics so tightly that it often sells a product before it pays the vendor. Suppose the payment cycle is net 30 days after delivery. If Walmart gets the product to the store and sold in fewer than 30 days, the sale funds the payment and cash flow is positive. Dell shows the same logic from the other side: the passages credit it with a significant cost and cash flow advantage over most computer competitors. Both are described as cases where segmenting products and customers has a direct, significant impact on profitability. Take the same test for your own product: find the days from delivery to sale and the days to payment. A segment where sale comes after payment is where you apply inventory reduction, shorter cycle times, and tighter forecasting first.

• If the days from receipt to sale are less than your payment terms, then the segment self-funds. Protect that flow and extend it to similar segments.
• If your reporting is only standard, ad hoc, or drill-down, then you are describing the past. Move the decisions that drive cost (safety stock, batch size, forecast) into statistical, predictive, or optimization methods before expecting margin gains.
• If an inspection or approval step cannot point to a documented standard, then remove it or define the standard first. Adding another check only adds cost.
• If a foreign supplier's margin is sensitive to a small exchange-rate move, then standardize the currency but test whether the local supplier can survive the cost shift.
• If you remove inventory buffers to cut cost and cycle time, then confirm partner integration and trust first. Without buffers, a chain failure is more catastrophic.

• Optimizing one flow only. You fix materials but leave money or information errors; the tell is wrong invoices, wrong ship dates, or bad inventory records erasing the savings.
• Cutting inventory without integrating partners. Buffers vanish but the rest of the chain cannot keep up; the tell is stockouts and failures that are bigger than before.
• Adding controls as the fix for every failure. Signatures and inspections pile up; the tell is long approval times (16 signatures per PO) and checkers who cannot say what they are checking against.
• Treating analytics as reporting. Dashboards multiply but decisions do not change; the tell is that everything on the BI ladder in use is about Who, What, When, Where, and never Why or What is Next.
• Ignoring currency and regional factors in cost comparisons. Cheap labor looks like savings; the tell is margin collapse from a small exchange-rate move, misread specifications (inches vs. meters), or forecasts borrowed from US demand patterns.

• ▸Segment name: product/customer group being assessed.
• ▸Supply chain cost as % of revenue: baseline (HP's is 65 cents per dollar) versus this segment's figure.
• ▸Materials/money/information failure points: list the known error in each flow.
• ▸Days from delivery to sale: average time inventory sits before it sells.
• ▸Supplier payment terms (days): net terms to compare with sell-through time.
• ▸Cash flow gap (terms minus days to sell): positive means self-funding.
• ▸BI level in use (standard through optimization): the highest level used for this segment's decisions.
• ▸Non-value-added controls: count of approvals/inspections lacking a documented standard.
• ▸Currency/regional exposure: effect of a small exchange-rate move on margin.
• ▸KPI and target: the metric tied to a business objective, with baseline and target.

1. 1Start by defining the target set of resources you will segment: customers, products, suppliers, or locations. In the aircraft engine example this splits into two major SKU groups — Supply side and Demand side — each getting its own segmentation structure and goals.
2. 2Set the primary and secondary goal per segment before choosing dimensions. The mail-order pharmacy's primary goal is 'get the correct prescription to the correct customer on time'; the secondary goal is 'being competitive,' which forces you to look beyond revenue and profit to what differentiates your service.
3. 3Pick the segmentation dimension(s) that separate resources by common need: for the Demand side use customer responsiveness and finished-goods inventory minimization; for the Supply side use planning, scheduling, consolidation, and cost reduction of incoming materials.
4. 4Apply retail-derived segmentation lenses where relevant — geographic/demographic (Where), psychographic (Who), positive market (What), behavioral (How), occasions (When), benefits (Why) — to build accurate profiles of each subset.
5. 5Assign different planners and different optimization tools to each segment. Demand side planners have a very different job from Supply side planners; do not force one toolset across both.
6. 6Customize planning and scheduling tools per segment to maximize that segment's goals — on-time delivery, inventory optimization, cost minimization, or customer satisfaction — rather than applying one rigid policy to all.
7. 7Build a differentiation strategy to exploit each segment so the more effective match between target customer and product yields a commercial advantage.
8. 8Re-test that segmentation still preserves standardization gains: keep process standardization where common characteristics exist, so you capture efficiency without treating all resources identically.

A large mail-order pharmacy grew so fast it stopped distinguishing among hospitals, clinics, and individual patients. Having lost the ability to track individual customer information, it defaulted to treating every prescription delivery as critical and overnight-shipped everything, driving shipping costs enormous. Segmentation fixes this on two axes: differentiate urgent from non-urgent deliveries by product (refrigerated drugs need overnight turnaround; renewal prescriptions are not as urgent as new ones), and differentiate customers (a hospital that stocks ahead versus a disabled or elderly patient who needs delivery to the door). Its true differentiator is delivery to the door without the customer leaving home, so speed matters most for that patient segment, not for a hospital pre-stocking ahead of need. By segmenting deliveries and customers, the pharmacy simultaneously reduces cost and improves service instead of forcing every shipment through the same overnight-express policy. The failure was never stepping back to say 'this doesn't make sense — there must be a better way' while too busy fighting fires.

• If you have exactly one product, one supplier, and one customer, then segmentation is meaningless — do not build segments.
• If two resources differ in location, timing criticality, quality requirement, or compatibility, then manage and respond to them differently rather than applying a uniform policy.
• If the resource is a delivery, then classify by urgency of the product (refrigerated = overnight; renewal = non-urgent) before assigning a shipping method.
• If a segment's stated primary goal is fully met by current practice but the practice threatens survival, then add a secondary competitive goal and re-optimize against both.
• If a SKU sits on the Supply side, then optimize for planning/scheduling/consolidation/cost; if on the Demand side, then optimize for responsiveness and finished-goods inventory minimization.

• Treating all customers the same in the name of efficiency and fairness — the tell is rigid policies, systems, and procedures with no way to identify or distinguish customer types.
• Losing individual customer information as you scale to hundreds or thousands — the tell is customers becoming a number on a list and every delivery defaulting to 'critical.'
• Optimizing only the primary goal (on-time service) while ignoring cost — the tell is overnight-shipping everything and watching shipping costs balloon toward bankruptcy.
• Forcing one segmentation structure and one toolset across both Supply and Demand sides — the tell is planners using identical tools despite having very different jobs and goals.
• Being too busy fighting fires to step back and question the process — the tell is nobody asking 'this doesn't make sense; there must be a better way.'

• ▸Resource type: which set are you dividing (customers / products / suppliers / locations)?
• ▸Segment name: label this subset by its common characteristic or need.
• ▸Common need: state the shared characteristic that binds this subset.
• ▸Segmentation dimension: which lens defines it (geographic, psychographic, behavioral, occasions, benefits, urgency, supply-vs-demand)?
• ▸Primary goal: the single outcome this segment must achieve (e.g., correct delivery on time).
• ▸Secondary goal: the competitive/survival constraint (e.g., cost, being competitive).
• ▸Optimization tool: the tailored planning/scheduling tool assigned to this segment.
• ▸Owner/planner: who plans this segment separately from other segments.
• ▸Differentiator: what makes serving this segment a commercial advantage.

1. 1Classify your current capability against the eight BI levels from Davenport and Harris, lowest to highest: standard reporting, ad hoc reporting, queries with drill-downs, alerts, statistical analysis, forecasting/extrapolation, predictive modeling, optimization. The first four are access and reporting; the last four are analytics. Mark each level as in use, partial, or absent for each supply chain decision area.
2. 2Sort questions by time orientation. Standard reports, ad hoc reports and drill-downs answer Who, What, When, Where and How about past events. Alerts answer "What is happening now?" The top four levels answer Why and What is Next. If your recurring questions are all Why or What-is-next but your tools stop at level four, you have a capability gap, not a reporting gap.
3. 3Define success by the business decision, not the analysis step. The source says analytics is a methodology whose ultimate goal is business decisions. For each analytic, write the decision it changes (a forecast, a safety stock level, a batch size, a planning and scheduling rule) and the action it recommends or triggers.
4. 4Build the statistical backbone where the source places it: forecasting calculations, inventory safety stock and batch size calculations, and planning and scheduling processes. Start with demand profiling, a two-by-two plot of demand volume against variability (coefficient of variation) for each SKU within a defined region, market or category.
5. 5Use the demand profile to identify the key supply chain segments (Godsell's step 6), then develop tailored practices for each segment (step 7). Do not apply one planning tool to the whole facility; select the planning and scheduling tool that is optimal for each segment.
6. 6Cleanse the data before analyzing it. The source sequence for applied SCS includes a Cleanse and Analyze Data step ahead of building segmentation logic. Treat Big Data (large volumes of detailed, generally unmanaged data) as raw input, not as an analytic asset.
7. 7Add visual BI for communication. Analytics is the discovery and communication of patterns, so pair each model with a visual or report that the decision owner actually uses, and add collaboration with sales for the demand forecast.
8. 8Run the Build/Run/Test/Validate loop on each segmented structure before committing, then assign a leader to own the decision culture. The source notes a growing number of business leaders driving analytical projects and putting their stamp on decision-making.

The preface describes a COO of an extremely complex supply chain who had been through dozens of consulting companies, each fixing only part of the planning and scheduling problem, and who asked which solution was best. The author answered that all of them are best, within the right segment: the facility does not need a one-size-fits-all solution but a segmented one, where a different planning and scheduling tool is optimal in each segment. Applying the capability steps: the team would profile demand by plotting volume against coefficient of variation for each SKU, and identify segments. It would then select the tool suited to each segment, and execute with that tool, with analytics supplying the forecasting, safety stock and batch size calculations inside each. The source states the result as "Then you will finally achieve optimality," and that SCS optimization is impossible without an analytics backbone. The passages give no numbers for this engagement, so the test of success is that each segment has a named tool and a named decision it drives.

• If your questions are Who/What/When/Where/How about the past, then standard reports, ad hoc reports or drill-downs are sufficient; if they are Why or What is Next, then you need statistical analysis, forecasting, predictive modeling or optimization.
• If a question is "What is happening now?", then build alerts rather than another report.
• If one planning or scheduling tool is being applied across SKUs with very different demand volume and variability, then segment first using the volume-versus-CoV profile and choose a tool per segment.
• If an analysis has no named business decision attached, then do not build it; analytics is judged by the decision it supports, not by the analysis step.
• If the data is large but unmanaged, then cleanse it before modeling rather than feeding Big Data straight into forecasts or safety stock calculations.

• Mistaking reporting for analytics: dashboards and drill-downs are labeled "analytics" while only the first four levels exist. The tell: every output describes the past and nobody can state a Why or What-is-next answer.
• Tool-centric analysis: teams optimize individual analysis steps instead of the methodology. The tell: models run but no decision or recommended action changes.
• One-size-fits-all planning: a single solution is applied across the whole operation, as the COO's consultants did. The tell: each fix helps one area and leaves others missing the mark.
• Skipping data cleansing and demand profiling: forecasts and safety stocks are computed on unmanaged data. The tell: results that planners distrust, and SKUs with high variability treated like stable ones.
• Avoiding math and statistics in decisions, as organizations traditionally did. The tell: forecasting, service level and inventory choices rest on judgment alone and no one owns the analytics.

• ▸Decision area: name the supply chain decision (forecast, safety stock, batch size, planning and scheduling tool selection).
• ▸Question type: mark as past (Who/What/When/Where/How), now (alerts), or future (Why/What is Next).
• ▸Current BI level: choose from standard reporting through optimization.
• ▸Target BI level: the minimum level that answers the question type above.
• ▸Data source and cleansing status: list data used and whether it is cleansed.
• ▸Demand profile: record volume and coefficient of variation per SKU group, region, market or category.
• ▸Segment and tool: name the segment and the planning or scheduling tool assigned to it.
• ▸Decision and action owner: name the person who acts on the output and the action recommended.
• ▸Validation result: record Build/Run/Test/Validate outcome and the date of next review.

1. 1Reject the single-tool premise first. The source's history shows each tool optimizes one problem: EOQ for order quantity, MRP for material schedules from routings, JIT for materials movement, TOC for machine bottlenecks. Any facility with two or three of those problems is mis-served by one tool. Write down which problem each part of your supply chain actually has.
2. 2Name the critical resource for each candidate segment. The source says it may be labor, materials, machinery, energy (aluminum) or logistics (apparel, bottling). A tool built around the wrong resource cannot fit. Record one resource per segment before you look at any tool.
3. 3Place each segment on the product lifecycle (the 'Life of a Product' concept). New or growing, mature and end-of-life items call for different tools. EOL planning and BTO/MTO planning are separate tools in the source's key-concepts list, so lifecycle stage is a selection input, not a label.
4. 4Describe each segment's demand and customer behavior with the source's quantitative concepts: forecast analysis, Coefficient of Variation (CoV), and touch/ERP integration. Use them to choose among the candidate tools: S&OP, MRP, EOQ, OIL, Min/Max, ROP with Statistical Safety Stock and Optimal Service Level, BTO/MTO, EOL. The passages name the tools but do not supply selection thresholds, so set your own and record them.
5. 5Build the segmentation logic map and map one planning and scheduling strategy to each segment. This follows Chapter 6's sequence: design the logic map, define and map strategies per segment, then build, run, test and validate. Each segment gets exactly one primary tool plus a stated reason tied to its resource and lifecycle stage.
6. 6Check each tool's known flaw against its segment. MRP, for example, assumes infinite worker capacity and produces scheduling shortages. If a segment's critical resource is labor, MRP alone fails that test unless capacity is handled some other way.
7. 7Rate current fit as an outsider would. The source warns that one of 10 departments having MRP does not make the organization proactive. Score fit per segment, not per company, and do not over-rank.
8. 8Run, test and validate the segmented structure, then re-test as products move through lifecycle stages. A segment's tool is no longer right once its lifecycle stage or critical resource changes.

The Preface's COO says dozens of consultancies each fixed a small part of his planning and scheduling problem and left other areas missing the mark, and he asks which solution is best. The author answers 'all of them are best' and recommends a segmented approach: define the segments, select the appropriate tool for each, and execute with it. The Introduction gives the logic behind that answer. EOQ was the original one-size-fits-all tool. MRP followed and scheduled materials from routings, but it assumed infinite worker capacity and was plagued with shortages. JIT then optimized materials movement and TOC optimized machine bottlenecks. Each tool is fit only where its problem is the segment's problem. Where the critical resource is energy (aluminum) or logistics (apparel, bottling), none of the four was built for that constraint, so the fit has to be tested rather than assumed. Applied to a facility like the COO's, you would split it by critical resource and lifecycle stage, map one tool to each segment, and validate the mapping before running it. The source does not give the COO's final segment-to-tool table, so the specific tool for each of his segments is not documented here.

• If a segment's critical resource is labor, then do not rely on MRP alone, because it assumes infinite worker capacity and yields scheduling shortages.
• If the binding constraint is a machine bottleneck, then use a bottleneck-focused tool such as TOC. If it is materials movement, use JIT. If it is energy or logistics, then none of the legacy tools fits by default and you must select deliberately.
• If a product is at end of life, then move it to EOL planning rather than keeping it under its growth-stage tool. If it is made only against customer orders, use BTO/MTO planning.
• If two segments need different tools, then do not average them into one compromise. The source's premise is that each segment gets its own optimal tool.
• If one department runs a tool well, then do not rate the whole organization as fitted. Score fit per segment, as an outsider would.

• Defaulting to the tool you already own, usually MRP, for every segment. The tell is recurring scheduling shortages in labor-constrained areas.
• Choosing by tool fashion or the last consultant's pitch rather than by segment characteristics. The tell is that you cannot state the segment's critical resource.
• Ignoring lifecycle stage, so end-of-life items stay under growth-stage planning. The tell is a segment whose tool has not changed while its product has.
• Over-ranking your current fit because one department has a sophisticated system. The tell is a high self-score with no segment-level evidence.
• Applying a tool in the wrong role. The source's example is SPC used as an after-the-fact control, where it only confirmed that defects had occurred. The tell is a tool that reports problems but does not help fix them.

• ▸Segment name: define the product/customer group being assessed.
• ▸Critical resource: labor, materials, machinery, energy, logistics or other. Name one.
• ▸Lifecycle stage: new/growth, mature or end of life, with the evidence for it.
• ▸Demand character: forecastability and CoV notes, and whether orders are customer-driven (BTO/MTO).
• ▸Current tool: S&OP, MRP, EOQ, OIL, Min/Max, ROP/SSS/OSL, BTO/MTO, EOL or other.
• ▸Known flaw of current tool vs. this segment: for example, MRP assumes infinite worker capacity.
• ▸Proposed tool and reason: tie the choice to the critical resource and lifecycle stage.
• ▸Fit score (1-10) and rater: score as an outsider would, and name who scored.
• ▸Validation result and re-test trigger: record the test outcome and the event that forces re-evaluation, such as a lifecycle change.

1. 1Define the unit of analysis. Return on Availability starts from the source's definition of segmentation: divide a target set of resources (customers, products, suppliers, locations) into subsets with common characteristics or needs. Pick one set, usually products or SKUs, and list every item in it.
2. 2Tag each item with its lifecycle stage and product characteristics. The source's segmentation types use product attributes, lifecycle stage, seasonality, and demand variation as focus areas. Add the lifecycle concepts from the SCS key-concepts chapter: life of a product, build-to-order/make-to-order, and end-of-life planning. These tags show where availability is worth paying for and where it is not.
3. 3Cost the availability of each item. Record what it takes to keep the item available: inventory held, lead time on the supply and demand side, channel complexity, and order size and volume. Use the source's warning that timeliness can be prohibitively expensive and therefore impractical, so cost is measured against the service level you promise, not in isolation.
4. 4Record the benefit of availability for each item. Use the benefit language the source attaches to segmentation: effective product planning, improved lead time and margin, reduced inventory, increased customer loyalty, and targeted sales. Note whether the benefit is a service level, margin, or retention effect.
5. 5Cluster the items into segments by return. Group items with similar lifecycle stage, demand pattern, and cost-to-benefit profile. If the groupings are not obvious, run the post-hoc statistical techniques the source names (K-means or cluster analysis, latent class analysis, or Random Forests) and then interpret the clusters.
6. 6Map the behavior toward each segment before you plan it. For every segment, answer the source's questions: what is the customer service level, what is an acceptable response lead time, and what is our tolerance for cost overruns or losses? These answers fix how much availability each segment gets.
7. 7Assign a planning and scheduling method to each segment from the source's toolbox. Options are statistical safety stock, optimal service level, reorder point cycle, EOQ, optimal inventory level, min/max, build-to-order, and end-of-life planning. Each segment gets the method that serves its goal, whether that is customer satisfaction, on-time delivery, inventory optimization, or cost minimization.
8. 8Test, validate, and revisit. The source ends the build phase with run/test/validate. Run the segmented plan, compare cost and benefit per segment against the baseline, and move items between segments as lifecycle stage changes. An item is not permanently in one segment.

The source's Dell e-commerce value chain (eVCM) shows availability being priced and allocated item by item. A customer configures a laptop online and sees a price. The system then pushes that configuration through the Supply Chain to check product availability down to the vendor's vendor, which is Available to Promise (ATP) validation. If product is not available, the system checks capacity, which is Capable to Promise (CTP) validation. It then allocates product and capacity through information collaboration across the chain. Read through Return on Availability, each configuration has a different cost to make available: stock that already exists, or capacity that must be committed. Each also carries a different benefit, a customized sale the customer has already priced. A segmentation team would place standard, stocked configurations in a segment planned with stock-based methods. It would place unusual configurations that depend on capacity in a build-to-order segment. The source gives no financial figures for Dell, so the cost and benefit measures per segment must come from your own data.

• If an item is in end-of-life or a short, high-demand lifecycle stage, then plan it with end-of-life or tightly bounded methods and do not extend its availability, because the cost of holding availability is no longer matched by future benefit.
• If a segment's required lead time or service level would be prohibitively expensive to meet, then lower the promised service level for that segment, because the source states that timeliness can be impractical when cost is ignored.
• If customers configure or order to specification, then use build-to-order/make-to-order planning with ATP/CTP checks, not stock-based reorder methods.
• If demand is stable and predictable, then use reorder point, EOQ, or min/max, since the segment can hold availability cheaply. If demand is highly variable (high CoV), then use statistical safety stock and optimal service level to set how much buffer is worth its cost.
• If an item or market area carries higher risk (unpredictable acceptance, vendor or competitor reactions), then treat it as its own risk segment with different planning and marketing mechanisms, according to your organization's risk appetite.

• Treating availability as uniformly good and applying one service level to every item. The tell is high inventory in slow or declining items alongside stock-outs on items that matter to customers.
• Segmenting on a single attribute, such as volume, and ignoring lifecycle stage. The tell is that items shift stage but stay in the same segment and plan, so end-of-life products still get growth-stage availability.
• Defining segments but never mapping behavior toward them. The tell is that no one can state the service level, acceptable lead time, or cost-overrun tolerance for a given segment.
• Optimizing one resource flow in isolation, for example chasing manufacturing efficiency while logistics erases the gain (Toyota builds a car in four hours but transit takes weeks or months). The tell is a good plant metric alongside poor delivered availability.
• Assuming one ideal segment structure exists and copying another firm's. The tell is segments that do not reflect your own priorities and goals, so planning tools are customized to the wrong objectives.

• ▸Segment name and members: list the SKUs or items grouped together and the shared characteristic that defines the group.
• ▸Lifecycle stage: record the stage (growth, mature, end-of-life) and any seasonality that affects the segment.
• ▸Demand pattern: note demand variation, using coefficient of variation if you have it, and whether items are stocked or ordered to specification.
• ▸Cost of availability: list inventory held, lead time, channel complexity, and order size and volume, with your own measured figures.
• ▸Benefit of availability: state the service level, margin, loyalty, or targeted-sales effect that availability produces for this segment.
• ▸Target service level and lead time: set the acceptable response lead time and the service level promised to customers.
• ▸Tolerance for cost overrun or loss: state how much cost above plan the segment may absorb before you act.
• ▸Risk level: mark the segment low, medium, or high risk and name the main source of unforeseen exposure.
• ▸Planning method: choose from safety stock, optimal service level, reorder point, EOQ, min/max, build-to-order, or end-of-life planning, and give the reason.
• ▸Review trigger: name the event (stage change, cost shift, service miss) that moves an item to another segment.

1. 1Start with the goal, not the metric. Write the organization's goals and the customer's expectations in one sentence each (for example, on-time delivery to the customer's initially requested date). Every candidate measure has to trace back to one of them.
2. 2Inventory what you measure and reward today. For each role, ask 'what is your performance measurement, and what does the bonus pay for?' The Provo plant surfaced its quality failure with exactly this question. The answer was 'piece parts produced.'
3. 3Test each current measure for what it motivates. Assume employees will make the number look good, because they are skilled at it. Ask whether making the number look good produces the result you want. If it detracts from that result, as piece-count did for quality, mark it for elimination.
4. 4Screen candidate measures against four criteria: the organization's goals, the customer's expectations, the response employees or suppliers will have to the measure, and the measure's accessibility and reliability. Start from the three focus areas: cycle time, on-time performance to customer expectations, and quality.
5. 5Cut to as few measures as possible, ideally ones that pull everyone's response toward the same optimization. Do not keep a measure out of tradition. Eliminate any measure that does not add value.
6. 6Audit control systems separately. List every approval, checkpoint, and signature step. For each, ask whether it acts 'as it happens' or 'after the fact,' and whether it fixes the error or only relocates or confirms it. Remove controls that add steps, cycle time, or failure opportunities without improving the output.
7. 7Tie pay and recognition to the chosen measures. Training, banners, and meetings do not change behavior if the paycheck rewards something else.
8. 8Use a feedback or reporting mechanism so the measures keep working. In segmentation work, choose motivation-based metrics (cycle time, inventory turns, operating costs) and customer metrics (on-time delivery, satisfaction, quality, touch points) per segment, then revisit them as the structure is tested.

A company needed to close one plant and chose the one with the lowest quality, in Provo, Utah. Managers later asked the author for a post-mortem. The plant had Six Sigma, TQM, other quality initiatives, banners, extensive training, and meetings, yet still had the worst quality of all the plants. The author asked what the performance measurements were for employee performance. The answer was 'piece parts produced.' Bonuses were paid on units produced, not on whether the parts were good. Quality belonged to the quality department and was often fixed after the fact, so the line employees 'madly pumped out parts.' The fix is to change what is measured and paid, because quality training did not help the paycheck and employees did not care about 'all the quality stuff.' Measuring good output or quality at the line would aim the same effort at the goal.

• If a measure draws effort away from the result you actually want (piece count vs. quality), then eliminate it rather than add training around it.
• If employees' pay depends on a number, then assume they will optimize that number, and check that the number equals the goal.
• If a control or checkpoint is 'after the fact' and only confirms that an error occurred, then it will not improve output. Replace it with measurement of the process as it happens.
• If several approvers sign each item and each defers to the others, then the controls are producing less control. Cut the signature chain and assign real accountability.
• If you are choosing among candidate measures, then pick the fewest that satisfy goals, customer expectations, employee response, and accessibility and reliability. Never keep one because of tradition.

• Quality programs layered on a contradictory incentive. The tell: banners, training, and Six Sigma are present, but results are worst-in-company and the bonus pays on volume.
• Too many measures. The tell: no one can say which number matters most, and employees optimize different ones.
• Stacked approvals that nobody checks. The tell: signers say 'I trust the others,' and the purchase order gets delays but no scrutiny.
• Using a good tool in the wrong place. The tell: Statistical Process Control is applied after the fact, defects are identified and counted, and quality output does not change.
• Controls that move the error rather than fix it. The tell: added steps, longer cycle time, wasted floor space and capacity, and the same errors showing up elsewhere in the process.

• ▸Organizational goal: state the result you want in one sentence.
• ▸Customer expectation: state what the customer expects (e.g., the initially requested delivery date).
• ▸Current measure: name the metric used for this role or process.
• ▸What the pay or bonus rewards: write exactly what is paid on.
• ▸Behavior the measure actually motivates: describe what employees do to make the number look good.
• ▸Gap versus goal: note whether that behavior helps or detracts from the goal.
• ▸Four-criteria check: score goals fit, customer fit, employee response, and accessibility/reliability.
• ▸Control step review: list each checkpoint and mark it 'as it happens' or 'after the fact,' and whether it fixes or only relocates the error.
• ▸Decision: keep, replace, or eliminate, with the replacement measure (cycle time, on-time performance, or quality).