GFT instruments / Inventory decisions
Set the buffer.
Know the exposure.
How much stock covers demand uncertainty at a chosen service level? Separate demand error from supplier uncertainty, then test the buffer, reorder point and order size.
01 / An instrument for the decision
How much stock covers demand uncertainty at a chosen service level?
Change the inputs. Read the result. Keep the assumptions visible.
Buffer / order / exposure
Model readyFor indivisible units, round the trigger up to 1227 units. EOQ is an unrounded cost benchmark.
Supplier variance represents 52.6% of combined uncertainty. Reducing residual forecast error by 35% reduces the buffer by 14.8%, with supplier risk unchanged.
The inventory rhythm
The buffer catches uncertainty.
Focus or hover a control to light the part of the trace it moves.
Find the part that can change.
Shares of variance, not additive shares of safety stock. Point at a segment to read it.
Demand error is 47.4% of the variance. Forecast improvement acts on this part only.
Demand error
Supplier lead time
EOQ implies 17.3 demand days between receipts. The service target is 95% of replenishment cycles, not a volume fill rate.
Keep the decision reproducible.
A is kept in this tab's memory. Shared links include A and B when A is saved. Inputs in a link are visible to anyone with that link. CSV files remain on your device.
02 / Compare the policy
Save A. Challenge it with B.
Until you change an input, B is a demonstration: the same case at a 99% service target. Change any assumption and B follows your inputs.
Safety stockunits
Reorder pointunits
Economic order quantityunits
Holding cost avoidedXAF/year
Read the full A against B table, inputs and outcomes
| Measure | A / saved | B / current | B minus A |
|---|
03 / Method, a continuous-review policy
Size the buffer separately from the order.
Four moves take the inputs to the buffer, the trigger and the order. The stage redraws on every step with your current inputs.
Illustrative inputs. Change them above and this stage follows.
Name the two uncertainties.
Demand error is the standard deviation of daily forecast residuals, σd, carried over a lead time of L days. Supplier error is the standard deviation of lead time, σL, times daily demand d. The two variances add, because they are assumed independent.
Scale by the service target.
The service target sets z, a normal quantile. Changing the target changes the protection factor. Changing the forecast changes the uncertainty being protected. They are different levers.
Add what the replenishment will consume.
The trigger is expected demand over the lead time plus the buffer. Order when inventory position reaches it: on-hand plus on-order less backorders, not the shelf count.
Size the order on its own cost.
Annual demand D, fixed order cost S and annual holding cost H, all in CFA franc (XAF), set the order quantity Q. The buffer does not change it. EOQ is a steady-demand cost benchmark, not a stochastic policy: round to pack sizes only after checking constraints.
04 / The tail you accept
Each point of service costs more buffer than the last.
The sawtooth you set above becomes a bell curve here. The service target picks z, the number of standard deviations of lead-time demand the buffer covers. Drag it and watch the covered area, the tail and the buffer move together.
This slider drives the service target in the instrument above. Reading the curve is not a forecast: it is the model's normal approximation.
z, protection factor
Safety stock, units
Cycles that can end short
05 / Worked example, independently checkable arithmetic
A known calculation anchors the instrument.
Step through a four-day teaching case to verify every equation by hand before you trust it on a line. These inputs are a teaching example, not a GFT factory observation, and the model tests reproduce this case at the corresponding normal probability.
Step 4 of 4: the reorder point is the expected demand plus the buffer, 46 units.
With your inputs: ordering cost = cycle holding cost XAF/year.
Economic order quantity check
The default economics are also illustrative: daily demand of 100 units over 300 demand days gives annual demand of 30,000 units. Ordering cost of XAF 15,000 per order and holding cost of XAF 300 per unit per year give Q = √(2 × 30,000 × 15,000 ÷ 300), approximately 1,732.1 units. The annual ordering cost equals the annual cycle-stock holding cost at that quantity. The displayed order quantity remains unrounded so this identity can be checked.
06 / Sensitivity, evidence is conditional
Better forecasts leave supplier uncertainty behind.
Point at or focus a segment to read it.
Buffer before
Buffer after
Reduction
Start from the programme range.
R27 records a forecast-error reduction of 20 to 50 percent in the programme annex. The tool maps it to the standard deviation of unbiased daily residuals. That is an illustrative modelling choice, and no improvement is claimed for GFT or any factory.
Cut only the demand term.
Replace σd with σd × (1 − reduction). Drag the slider: the after bar shrinks, and the buffer follows at a smaller rate than the error.
The supplier term stays put.
Lead-time variance does not move, so the buffer falls by less than the error does. The holding cost avoided is the buffer cut times H. It is not released working capital or a cash-flow forecast.
07 / Assumptions register
Every default has a visible status.
10 inputs, none measured. Filter by status to see how many are illustrative, policy, stress or sourced.
| Input | Default | Live now | Status | Basis | Open |
|---|---|---|---|---|---|
| Mean daily demand | 100 units/day | units/day | Illustrative input | Illustrative item profile selected to show both uncertainty terms; no measured factory data. | Open |
| Daily residual standard deviation | 30 units/day | units/day | Illustrative input | Illustrative item profile selected to show both uncertainty terms; no measured factory data. | Open |
| Cycle service target | 95 % | % | Policy target | Illustrative policy target, not an achieved service result. | Open |
| Mean replenishment lead time | 10 days | days | Illustrative input | Illustrative item profile selected to show both uncertainty terms; no measured factory data. | Open |
| Lead-time standard deviation | 1 days | days | Illustrative input | Illustrative item profile selected to show both uncertainty terms; no measured factory data. | Open |
| Demand days per year | 300 days/year | days/year | Illustrative input | Illustrative cost and calendar inputs selected for reproducible EOQ arithmetic; no supplier quotation. | Open |
| Fixed ordering cost | 15000 XAF/order | XAF/order | Illustrative input | Illustrative cost and calendar inputs selected for reproducible EOQ arithmetic; no supplier quotation. | Open |
| Annual holding cost per unit | 300 XAF/unit/year | XAF/unit/year | Illustrative input | Illustrative cost and calendar inputs selected for reproducible EOQ arithmetic; no supplier quotation. | Open |
| Forecast-error reduction case | 35 % | % | Sourced range | Illustrative midpoint of R27’s 20 to 50 percent range; mapping to residual standard deviation is assumed. | Open |
| Adverse cycle, excess demand | 2 σ above mean | σ above mean | Stress choice | Illustrative excess demand expressed in lead-time standard deviations; a stress choice, not a predicted event. | Open |
Showing all 10 inputs. None is a measured value.
08 / Limits, check before an operating change
A service target does not guarantee availability.
The target is the probability of avoiding a stockout in one replenishment cycle, under a normal approximation. It is not the share of units supplied at once. Tick what applies to your case and watch the trace leave the model.
Each ticked limit deepens the stockout notch.
of 6 limits apply
Your caseThe modelCompound stockout notch
No limit ticked. Your case follows the model trace, which is the best case the approximation can show.
A negative end point is a net shortage or backlog, not physically negative stock. Check the real distribution of replenishment demand, price the cost of failure and confirm the policy on held-out cycles before adoption.
Does your case sit outside the model?
Tick to bend the traceNone ticked. The approximation may be tried, then confirmed on held-out cycles before it changes live stock.
09 / Evidence, read alongside the model
Keep the method and its source together.
Point at a source or a policy card below to see which term of the model it backs.
- 01Peter L. King, Understanding safety stock and mastering its equations
APICS magazine, July/August 2011, hosted by MIT. Opened 30 October 2026.
The combined uncertainty method and the cycle-service distinction.
- 02NC State, Safety stock analysis
Published 28 January 2011. Opened 30 October 2026.
Reorder logic.
- 03NC State, Economic order quantity model
Published 28 January 2011. Opened 30 October 2026.
Ordering and holding cost objective.
- 04R27, McKinsey, The State of Fashion 2025
As recorded in GFT's programme annex. The underlying article has not been independently reverified here. See the citation qualification beside the sensitivity claim.
The 20 to 50 percent forecast-error range, used as a sensitivity only.
Four sources. Point at one to read which terms of the model it backs.
Read the questions behind the policy.
Which uncertainty drives the buffer: demand or replenishment time? Which cycle-service target can your operation sustain? Check forecast residuals, supplier delays and the cost of a stockout before choosing a policy.
Compare with the OEE instrument to distinguish demand uncertainty from production losses. The measurement approach explains how these models relate to factory decisions.
The model stays visible.
SS = z × √(L × σd² + d² × σL²). Reorder point = d × L + SS. EOQ = √(2 × annual demand × ordering cost ÷ annual holding cost per unit). Continuous review; independent uncertainty; normal approximation.
Method source: Peter L. King, APICS (2011), hosted by MIT; NC State inventory tutorials (2011). ↗
A planning aid. Not investment, legal or tax advice. Verify inputs and boundaries before an operational decision.
10 / Use the result in its boundary
A smaller buffer needs evidence.
Validate the forecast on unseen replenishment periods, measure supplier delays and review achieved service before changing a live stock policy.
Nothing validated yet: the buffer line shows the current model at 95.0%, 226.7 units. Release no stock on it.
Model buffer at the chosen levelunits
Reorder pointunits
Service level on the linecurrent instrument setting
Release stock only after the new policy proves itself.
The sensitivity case is conditional. Better forecasts cannot remove supplier variance, and a service target is a probability under assumptions. Confirm pack sizes, storage limits and critical-material consequences before placing an order.