Four S&OP KPIs that describe how the cycle actually ran
Nov 18, 2026, 9 min read

Forecast accuracy describes the forecast rather than the cycle, so S&OP KPIs have to measure how the process itself ran. Cycle time depends entirely on which two events you measure between, and pulling the executive meeting earlier improves the number while shortening the supply review behind it. Counting decisions works only under a filter, because counting minuted actions rewards a thorough note taker. Plan churn, comparing the plan published this cycle against last cycle's plan for the same target month, is the measure the plant feels and the one least often reported.
Somebody on the executive team asks how the S&OP process is performing. The answer offered is forecast accuracy, which came in at 78 percent, and the conversation moves on. Forecast accuracy measures the forecast. The process could have run brilliantly and produced a poor forecast on a hard month, or it could have been three meetings where nothing was decided and a number that happened to land close.
The measures that describe the cycle itself are different, and almost nobody reports them, partly because they are uncomfortable and partly because each one has three or four definitions that give different answers on identical facts.
Cycle time, measured between which two events
Take a real calendar. Period close starts on day 1 and completes on day 3. The statistical forecast runs on day 4. Demand review sits on day 8, supply review on day 11, the reconciliation session on day 14, the executive meeting on day 17. Decisions get minuted and circulated on day 19. The revised plan loads into the supply system on day 22 and the plant sees a changed schedule on day 24.
Now measure the cycle.
Meeting to meeting is 30 days. That is the repeat interval and it is true by construction, so it measures the calendar rather than the process.
Demand review to executive meeting is 9 days, which is the number most S&OP leads quote, because it is the stretch they personally run.
Close to executive meeting is 14 days, which is the number that appears in maturity assessments.
Close to a plan the plant can act on is 21 days, and that is the only one connected to anything the business feels.
The version worth putting on the wall goes one step further. A demand signal arriving on day 2 waits for the next cycle to pick it up, and with a 30 day repeat the average wait is 15 days. Add the 21 day latency and the plan the plant is executing against reflects information that is on average 36 days old, with a worst case of 51. Cutting the executive meeting from day 17 to day 15 changes that 36 to 34. Moving to a fortnightly cycle changes it to 28 with no change to the meetings at all. The second lever is much larger than the first, and the first is the one every improvement programme pulls.
Decisions, counted under a filter
Most S&OP minutes record actions. Counting actions gives you a number that rises whenever the note taker is thorough.
Apply four tests instead. A decision has a named owner, a date by which it takes effect, a quantified expected effect on volume, margin or cash, and something that changed in a system or in a commitment to a customer or supplier as a result. Anything failing one of the four is an action, a discussion or a note.
Run that filter on a typical set of minutes and the fourteen items on the page become three. That reduction is the finding. A cycle producing three decisions a month is not necessarily unhealthy, and it becomes a problem when you look at what the decisions covered.
Which is the second half of the measure. The gaps entering the cycle carried, say, 19 million of margin exposure across the rolling horizon: volume shortfalls against budget, capacity constraints on two lines, a supplier allocation and a customer listing at risk. The three surviving decisions addressed 2.1 million of that. Eleven percent coverage tells you where the cycle is weak far more usefully than the count of three, and it points at the specific gaps that were discussed and left open.
Assumptions, scored three ways
The plan rests on assumptions, and scoring them later is the closest thing S&OP has to a calibration check. The ledger that holds them belongs to B3; the question here is how to score it.
Take twenty assumptions written into a plan and score them at 90 days.
Binary scoring asks whether each held as stated. Eleven of twenty held, so 55 percent.
Tolerance scoring asks whether each held within a stated band, usually 10 percent. Fourteen of twenty, so 70 percent.
Value weighted scoring asks how much of the plan's exposure sat on assumptions that held. Suppose a customer listing decision governed 4.6 million of margin and went the wrong way, a price increase governed 3.2 million and did not stick, and the fourteen assumptions that held within tolerance governed 7.2 million between them. Then 7.2 of 19 million held, which is 38 percent.
Same twenty assumptions, same outcomes, three scores: 55, 70 and 38 percent. Only the third one tells you whether the plan was resting on anything solid, because assumptions are not equally load bearing and a hit rate that treats them as equal will look healthiest exactly when the two big ones have gone against you.
The scoring choice also decides what people write. Score binary and assumptions get written vaguely so they can never be marked wrong. Score value weighted and every assumption needs a number attached to what it governs, which is a better discipline and a harder one.
Escalation, and where gaps get resolved
Nine gaps enter the cycle. Five get closed inside the demand and supply reviews. Four go to the executive meeting, where two are decided and two are deferred.
Escalation rate is 44 percent. Resolution at source is 56 percent. Deferral rate among escalations is 50 percent. Each of those describes something different about the process, and the third is the one that predicts trouble, because a gap deferred twice has been open for sixty days and the window to act on it has usually closed by then.
There is a fourth number nobody collects, which is the gaps that were never raised. You can approximate it by comparing the plan gap at the start of the cycle against the gap in the published plan: exposure that disappeared without a decision was either resolved by luck or absorbed by somebody quietly. Where that figure is large, the functional reviews are filtering rather than escalating.
Measure escalation rate as a fault and you will get fewer escalations within one quarter, along with the same number of problems arriving later and with less time to fix them. Pair it with the value of unresolved gaps carried into the next cycle, and the incentive reverses.
Plan churn at the level the plant feels
Compare the plan published this cycle against the plan published last cycle for the same target month.
At total company level the number moved 1.8 percent. Across 42 product families, the volume weighted mean absolute change was 9.4 percent. Across 8,600 item-location combinations it was 27 percent. Same two plan versions, three churn figures, and the ones the factory and the distribution centre live with are the second and third.
Then split by horizon. Of that 27 percent, the changes falling inside the four week frozen window affected 310 item-locations, and every one of those forced a schedule change, a material call-off or an expedite. Churn outside the freeze costs almost nothing. Churn inside it is the number to report, and reporting total churn hides it inside a much larger denominator.
Churn is also the measure most easily bought with the wrong currency. Freeze earlier and churn falls, while the volatility moves out of the S&OP process and into in-month expediting, which is Y1 territory and has its own measures. Report churn alongside the count of unplanned changes made after the freeze, or you will reward moving the mess rather than reducing it.
What these measures invite people to do
Cycle time invites moving the meeting earlier. The executive meeting shifts from day 17 to day 13, the number improves, and decisions get made on an incomplete supply picture because the supply review had four days rather than seven. Pair cycle time with the count of decisions later reversed or revisited in the following cycle.
Decision count invites minuting trivia. Pair it with decision value and with gap coverage.
Assumption hit rate invites vagueness, unless the ledger requires a number, a date and a tolerance for every entry.
Escalation rate invites silence at the functional reviews. Pair it with unresolved gap value.
Plan churn invites earlier freezes. Pair it with post-freeze changes.
The pattern across all five is the same: each cycle measure has a cheap route to improvement that damages the outcome, and each needs a partner measure that closes the route. Publishing the pairs together, in the same table, is what makes the set usable.
Where the data for these already lives
None of this needs a new system, and three of the five need nothing you do not already produce.
Cycle time comes from calendar invites and system timestamps. The close date is in the general ledger, the forecast run has a job log, the meeting dates are in Outlook, and the plan load into the supply system has an execution timestamp. Assembling them into one table per cycle takes an hour a month by hand and is worth automating once the argument about which pair of events to use has been settled.
Decision count and value need the minutes restructured into a table with four columns: owner, effective date, quantified effect, and the system or commitment that changed. Writing the minutes that way is a change to how one person takes notes, and it produces the measure as a by-product rather than as a reporting exercise.
Escalation counts come from the same table, with a status field that distinguishes decided from deferred and carries the value of anything unresolved.
Plan churn is the one that needs something you may not have, which is a stored snapshot of every published plan version at item-location. Where those snapshots have never been kept, churn is unmeasurable retrospectively and you start the clock from the next cycle. Storing them is cheap and the storage decision has a habit of being deferred for years, so it is worth making this week rather than as part of a project.
Assumption scoring needs the ledger, and B3 has both the structure and the discipline that makes it work.
Where this stops
None of these measures says whether the plan was any good. A cycle that runs in 14 days, produces eleven qualified decisions and holds 80 percent of its assumptions can still be planning to the wrong number. Cycle health has to be read next to an outcome measure, which usually means forecast bias at the lag the business commits at (BB15 has the horizon argument, BB1 the bias argument) and gross margin against the plan the cycle published.
External benchmarking is close to worthless here. Cycle time from close to executive meeting varies by two weeks depending on whether the business counts from close start or close complete, and nobody publishes their definition. Compare against your own history on a frozen definition and treat other companies' figures as a source of questions.
There is also a limit on precision. Measuring a cycle to the day encourages a lot of attention to the meeting calendar, and in most businesses the binding constraint is one person's availability or one data feed that arrives on day 3 instead of day 1. Fix the constraint before you optimise the schedule around it, and expect the cycle time measure to be flat until you do.
Take last cycle's minutes, apply the four part decision filter to every item on the page, and count what survives.
More on this in metrics on the DemandLabs platform.




