Demand, service and the analytical grain
6 min read
Lesson objectives
- Build a due-date cohort and calculate order and unit service without double-counting split shipments.
Please sign in to use all the tools
Once you sign in, the reading tools are yours: larger or smaller text, typeface, line spacing, focus mode, highlighting and bold, and notes saved to your account and gathered in My study.
A warehouse dashboard may report that every shipment left on schedule while customers say that half their orders arrived late. Both reports can describe the same operation. The warehouse may count dispatch events; the customer may count complete deliveries. Before calculating a percentage, decide what one row represents. This is the analytical grain: a whole order, an order line, or an event affecting a shipment. Counting event rows does not give an order count. The distinction determines what question a result can answer. A dispatch improvement, for example, does not necessarily resolve a delay between the loading bay and the receiving customer.
An order records requested quantity and an agreed commitment. A shipment is a physical grouping: it can serve several orders, and one order can require several shipments. Delivery records arrival; acceptance adds a judgment about whether the quantity is usable. Keep an order ledger with order and line identifiers, item, quantity, unit and original commitment. Keep an event ledger with an event identifier, the linked order, movement type, event time and accepted quantity. Use a separate allocation table when relationships are many-to-many. Repeating the full ordered quantity on every movement row and summing it would multiply demand by the number of recorded events.
ASCM's SCOR description treats perfect orders as broader than quantity delivery alone, including timing, condition and documentation. GS1 describes EPCIS visibility data as events concerning an object, time, place and business context. These ideas help distinguish a customer commitment from its supporting events. For this lesson, define a local on-time-in-full measure narrowly: an order succeeds only when its entire accepted quantity has arrived by its original due date. Documentation is outside this fictional dataset. Consequently, the measure must not be presented as a complete perfect-order score or as an implementation of every requirement in either framework.
Choose the cohort by orders due within the measurement period, not by shipments completed during it. A completed-only selection omits the most delayed orders and can make poor service look stronger. Fix the reporting cutoff, the period boundaries, the time zone and whether days mean calendar or working days. Our case uses calendar days, one time zone and a reporting period ending on day seven. An order due on day eight stays outside the denominator even if it arrived early. An order due on day five remains inside, including when it arrived on day six. Arrival performance must not redefine membership in the cohort being evaluated.
| Order | Ordered | Accepted by its due date | Complete and on time | In measured cohort |
|---|---|---|---|---|
| 1 | 100 | 100 | Yes | Yes |
| 2 | 80 | 0 | No | Yes |
| 3 | 50 | 40 | No | Yes |
| 4 | 40 | 40 | Yes | Yes |
| 5 | 60 | 60 | Yes | No; due on day eight |
Splitting Order 1 into two deliveries neither penalizes it twice nor rewards it twice. Sum its accepted quantity by its commitment and then evaluate the order once. If an order contains two lines, order success requires both lines to satisfy the agreed definition; average line performance is a different measure. Also cap the quantity credited to each line at the quantity ordered. An unexplained receipt of 110 units against a 100-unit requirement must not compensate for a shortage on another order. Retain the excess in an exception view rather than deleting it. The cap controls the service calculation; it does not establish that the excess transaction was legitimate.
Use two reconciliation checks before presenting the result. Eligible distinct orders must equal successful plus unsuccessful orders. Required quantity must equal quantity accepted on time plus the on-time shortfall, after preventing overdelivery from offsetting another order's shortage. Here, 4=2+2 and 270=180+90. The shortfall contains 80 late units and ten damaged units. The first points toward an investigation of release or transport timing; the second points toward packaging or handling. These are investigative directions, not proven causes. Report the denominator alongside the percentage: changing the outcome of one order in a four-order cohort moves the order success rate by 25 percentage points, which limits how confidently short-term fluctuations should be interpreted.
Preserve both event time and recording time. A receipt that occurred on day five but was entered on day six can correct the delivery history without showing that the planner knew about it on day five. Likewise, an agreed new due date must not erase the original promise. Report revised-commitment performance separately, including how many orders changed and why. A commercially legitimate revision is still different from meeting the first commitment. To reconstruct a dashboard as it was known when a decision was made, use records available by that decision time. To calculate eventual service using complete evidence, use the verified event times. Distinguishing these questions prevents later information from being attributed to an earlier planner.
Lesson summary
Define the grain, cohort and commitment before calculating service; order success, on-time units and cutoff coverage answer different questions.