How to allocate merchandise to stores

• ✓Allocation moves a fixed buy between places; it cannot change the total, so the first allocation is a forecast and the reserve is the correction.
• ✓Door quantity = presentation minimum + weeks of cover × forecast rate, capped at capacity. Units a capped door cannot take go back to the reserve.
• ✓Grade doors for depth and cluster them for size curve and timing — a high-volume door can still sell a large-skew curve.
• ✓Allocate new styles from like-for-like proxies measured over in-stock weeks, or a door that ran out reads as a door that did not sell.
• ✓Read sell-through and rate of sale by door against forecast, not a chain total: equal sell-through can hide one door on plan and another running 30% ahead.

Allocation, replenishment and transfers: three different jobs

Three activities move stock from the DC to doors and between doors, and they can be run by the same team on the same screens. They are not the same decision. Initial allocation is a forecast, replenishment is a reaction, and a transfer is a correction that pays for its own handling twice. Treating them as one job is how a season ends up correcting with transfers what it should have corrected with a reserve.

The sequence matters more than the definitions. A seasonal style gets one initial allocation, then corrections from a reserve held in the DC, then — only if both of those missed — transfers. A continuity style lives on replenishment after its first fill. The replenishment program guide covers the second case in full, including the test a seasonal style has to pass before it graduates onto a trigger.

Allocation also has a fixed place in the planning sequence. The merchandise plan sets the money, the assortment plan sets the styles and the doors that carry them, the buy plan sets the quantities and delivery windows, and allocation takes those quantities as given and places them. The line plan and assortment plan guide covers the decision of which doors carry a style at all, and how to build a buy plan covers the quantity. An allocation that finds the quantity wrong cannot fix it; it can only report it back to the next buy.

What all three share is that none of them can change the buy. An allocation can put 3,000 units in the right doors or the wrong ones; it cannot make them 3,500. That is why an allocation problem and a buying problem produce the same end-of-season markdown line, and why telling them apart takes a door-level read rather than a chain total — the subject of a section further down.

The inputs, and where each one comes from

An allocation is only as good as the seven inputs underneath it, and each one has a different source and a different failure. Grades set depth, clusters set curve and timing, and the two are separate decisions about the same door.

Grades and clusters are where the most information is lost, because a single A/B/C grade can be asked to carry both. A high-volume door can sell a large-skew size curve; a low-volume door in a cold region can need its outerwear three weeks before the rest of the chain. Grade on volume for depth, then cluster separately on size profile and on climate or format for curve and timing, and let a door sit in one group of each.

Capacity is the input most easily treated as infinite. A door’s capacity for a style is the fixture space the category has been given, plus whatever the stockroom holds behind it, and it can bind hardest in the doors that sell most: high-street doors with small stockrooms, and any door during a floor-set change. A capped door does not need its demand reduced; it needs its excess routed somewhere that can hold it — usually the reserve.

The presentation minimum is a merchandising standard, not a forecast. It is the number of units below which the style stops looking like a deliberate offer — a half-empty fixture, or a run with missing sizes — and it is held on the floor rather than counted as cover. Set it per fixture type, with a minimum per size, and write it down, because every other number in the allocation is built on top of it, and because the minimum buy depth guide uses the same figure to test whether a buy can reach a set of doors at all.

Grading doors: cut where the gaps are

A grade is a band of doors expected to sell a category at a similar rate, and it is built from the doors’ own history rather than from their square footage or their reputation. Rank the doors by their sales in the category over the last comparable season, measured over in-stock weeks, and look for the gaps. Cut the ranking where the volume drops, not at a fixed number of doors per grade, so that doors inside a grade are close to each other and doors in different grades are not.

The ten doors in the worked example — labelled by the grade they end up in — sold 40,000 units of the category last season. The two largest gaps in the ranking are between the second and third doors, 7,800 to 4,300, a 45% step down, and between the sixth and seventh, 3,700 to 2,200, a 41% step. Cut there and the A doors hold 40% of the volume, the B doors 40% and the C doors 20%: the same proportions the proxies give the new style further down, which is a useful cross-check rather than something the method relies on. Inside each grade the doors sit within a few hundred units of each other.

Grade per category, not per door. A door can be an A for outerwear and a C for swim, because its climate, its customer and its fixture space decide its category volumes separately, and a single door grade applied to every category sends swim depth to a door that sells outerwear. Re-grade every season from in-stock-adjusted history as well: a door that was under-allocated last season sold less because it had less, and grading it on raw sales locks the error in.

Doors at a boundary deserve a second look. A door just below a cut is allocated like the doors below it, so check whether its rate over in-stock weeks puts it with the grade above, and whether a refit, a new neighbour or a nearby closure has changed its position since last season. The grade is an input to a forecast, not a ranking to defend.

Seven steps, in order

Steps five and six are arithmetic. The other five decide whether the arithmetic puts stock where it will sell — a wrong grade, a rate measured over stocked-out weeks or a missing read date gives a correct division of the wrong numbers.

1. 1

   Fix the allocatable quantity

   Start from the committed buy and take out what is already spoken for: booked wholesale orders, and the ecommerce fulfilment share from the plan. What is left is the store share. Allocation cannot change the total, only where it goes, so this is the number every later step divides.

2. 2

   Grade and cluster the doors

   Grade doors by volume in the category to set depth, and cluster them by the attributes that change what sells — size profile, climate, format — to set curve and timing. A door can be a high-volume grade in a large-skew size cluster; the two are separate decisions.

3. 3

   Forecast each door’s rate of sale

   Use the style’s own history if it has one; otherwise take door shares from like-for-like proxy styles and apply them to the planned chain rate. Measure every rate over in-stock weeks, or a door that ran out reads as a door that did not sell.

4. 4

   Set the presentation minimum

   Decide the units a door needs to present the style properly — a fixture fill, with at least a minimum in every size it carries. The minimum sits on the floor and is not counted as cover, because a door that sells into it starts to look broken.

5. 5

   Allocate floor plus cover, capped by capacity

   Give each door its presentation minimum plus a set number of weeks of forecast demand, and cap the result at what the door can physically hold. Units a capped door cannot take go back into the reserve, not to a door that does not need them.

6. 6

   Split each door into sizes

   Break each door’s quantity into sizes with the curve that door’s cluster actually sells, not the chain curve. Round so the smallest size still meets its minimum, and check prepack ratios against the curve before shipping in packs.

7. 7

   Hold the reserve and set the read date

   Everything not pushed stays in the DC as a reserve. Before the push, set the date of the first sell-through read and the rule for releasing the reserve, so the correction happens on a schedule rather than when someone notices an empty fixture.

One style, ten doors, ecommerce and a reserve

One style-color, a woven shirt in five sizes from XS to XL. The buy plan sized it at 3,000 units for a twelve-week full-price window, 250 a week: 50 a week online and 200 a week across ten doors. There are no wholesale commitments on this style. Ecommerce fulfilment takes its planned share, 600 units, and the ten doors share the other 2,400.

The doors are graded A, B and C on category volume: two A doors forecast at 40 units a week each, four B doors at 20, four C doors at 10 — 200 a week in total. The style is new, so those rates come from proxy styles; the section on new styles works through where they come from. The presentation minimum is 40 units a door: one fixture face across five sizes, with at least two in the smallest. The initial push covers six weeks of forecast demand on top of the minimum. The two A doors are high-street doors with small stockrooms and can hold 240 units of this style; the B and C doors can hold 320 and 200.

The model quantity is the minimum plus six weeks: 40 + 6 × 40 = 280 for an A door, 40 + 6 × 20 = 160 for a B door, 40 + 6 × 10 = 100 for a C door. Capacity binds on the A doors, which take 240 each. The 40 units each A door cannot hold go back into the reserve, not into the B and C doors, whose forecasts do not call for them. The push is 2 × 240 + 4 × 160 + 4 × 100 = 1,520 units, and the DC reserve is 2,400 − 1,520 = 880.

Three things are worth reading off the result. The A doors carry 40% of the store forecast and receive 31.6% of the push, because capacity, not demand, sets their number. The C doors carry 20% of the forecast and receive 26.3%, because the minimum is a fixed 40 units however little a door sells, and it is 40% of a C door’s allocation against 17% of an A door’s. And the minimums together are 400 of the 1,520 units pushed: a quarter of the initial allocation is presentation stock that is not expected to sell in the first six weeks.

Six weeks of cover is a choice, not a rule. It is long enough that no door needs the DC before the first read and the release that follows it, and short enough to leave a reserve worth reading against. At eight weeks the same method pushes 1,760 units and holds 640; at four weeks it pushes 1,200 and holds 1,200. The longer the cover, the less the reserve can correct; the shorter, the more the doors depend on the DC reaching them on time.

The reserve is 36.7% of the store share and 29.3% of the buy. Nothing about that figure is a target. It is the residue of three decisions — six weeks of cover rather than the full twelve-week window, a capacity cap on two doors, and a store share fixed by the buy — and it exists to be spent on the first read, which the section on the first two weeks works through.

Why a flat split fails

The simplest allocation divides the push evenly: 1,520 units across ten doors is 152 each. It ships the same total and puts it in the wrong places. Measured in weeks of cover above each door’s 40-unit minimum, at the forecast rate:

The graded allocation gives every door five or six weeks. The flat split gives an A door 2.8 weeks and a C door 11.2. A flat split is not neutral — it is a decision to understock the doors that sell and overstock the doors that do not, made by not deciding. The A door runs out of its core sizes in the third week; the C door reaches the first markdown holding most of what it received.

Both errors then feed the next season. The A door’s sales are capped by its stock, so its history understates its demand; the C door’s sales are inflated by depth that had to clear, so a share-of-sales proxy overstates it. A flat split this season makes the graded allocation next season less accurate — which is why the fair-share allocation formula on RetailNorthstar starts from proportional demand and adjusts from there, rather than from an even split.

Push, pull and the hybrid a seasonal style needs

Push allocation sets door quantities centrally from a forecast and ships them. Pull allocation lets each door’s own sales decide what it receives, through a reorder point or a model stock. Neither is better in general; each fits a different kind of style.

Pure push has no second chance short of a transfer. Pure pull fails on a new style for a mechanical reason: a door with no stock sells nothing, so its trigger never fires, and a door that receives a token quantity sells it and asks for exactly that again. Pull needs history to start from, which a new seasonal style does not have.

The two also fail in opposite directions when the forecast is wrong. A push that over-forecasts a door strands stock there until a transfer or a markdown moves it; a pull that under-reads a door starves it a little at a time, one reorder cycle after another, without anyone seeing a single large error. The first is visible and expensive; the second is quiet and cumulative, and it shows up only as a door whose history keeps shrinking because its stock did.

The hybrid pushes enough to present and sell from day one, and pulls the rest once the doors have said where the demand is. Its cost is the floor presence the reserve units would have had in their first weeks; its benefit is that the correction is a pick from the DC rather than a second move between doors. It works only when the selling window is long enough for a read to arrive while full-price weeks remain — which is why a gifting-peak or launch-window style, where it is not, leans toward push, and the peak-concentrated season guide sets the reserve against the time a delivery can actually take.

When the push lands

An allocation is also a delivery schedule. The units have to reach each door with enough time to be received, ticketed and put out before the date the style is meant to be on the floor — the floor set or the launch — and the DC can only pick so many in a day. A push to ten doors is one wave; a push to several hundred is a schedule, and the doors at the end of it start selling later than the doors at the front.

Sequence the push by what each door loses by waiting. Climate clusters that sell the category first get it first: cold-weather doors take outerwear before warm ones. Doors with a local event, a reopening after a refit or a launch date go ahead of doors with no date pressure. Delivery days and receiving capacity set the rest: a door that receives twice a week cannot take a push on a third day without stock sitting in its stockroom unworked.

The read date moves with the push. A door that started selling a week late has a week less history at the chain read date, so its rate is measured over its own selling weeks rather than over the calendar since the first door launched. Otherwise the late doors read as slow and lose their share of the reserve to doors that were simply served first.

Allocating to a floor set

A floor set is a planned presentation: which styles go on which fixtures, in which doors, from which date. It turns some of the allocation’s inputs into commitments. A style placed on a front table or a window in a door needs the units to fill that fixture on the set date, so its presentation minimum in that door is set by the fixture rather than by the standard face, and it can be several times the 40 units the worked example uses.

Feature placements are depth decisions made by visual merchandising, and the allocation has to price them: every unit committed to a feature fixture is a unit that is not in the reserve and not in another door. Agree the floor set before the push is calculated, so that its minimums enter the model as inputs, rather than discovering at the set date that the feature fixture needs units the allocation sent elsewhere.

The floor set also fixes the timing. Stock has to be in the door, received and ticketed before the set date, so the push for a floor set is scheduled back from that date by the door’s delivery days and processing time — the same back-scheduling a T&A calendar applies to production, applied to the last mile. When the floor set changes mid-season, styles leaving feature fixtures fall back to the standard minimum, and the units above it become cover the door may not need; that change is a natural point to re-run the release rule.

Holding back a DC reserve

A reserve is the part of the store share kept in the DC at the first push, to be sent where the first sell-through read says demand is. It is the cheapest correction an allocation has, because a unit still in the DC can go to any door at the cost of one pick, while a unit already in the wrong door costs a transfer. Holding units back costs floor presence while they wait and nothing else; every correction available after the push spends handling, freight or margin instead.

No figure is right for every style, and this guide offers none. Four things size it. The first is how uncertain the forecast is: a carryover style with its own door history needs less than a new style allocated from proxies. The second is how quickly the DC can reach a door; a reserve that takes two weeks to arrive has to be released earlier, against a thinner read. The third is how many full-price weeks remain after the read — a reserve released into the last fortnight of a window arrives as markdown stock. The fourth is the presentation minimum: the push has to cover it in every door regardless, so the reserve is drawn only from what sits above it.

In the worked example the reserve is 880 units, 36.7% of the store share. It came out of the method rather than going into it: six weeks of cover rather than the full twelve, plus 80 units the capped A doors could not hold. A planner who wants a larger reserve shortens the cover in the push; one who wants a smaller reserve lengthens it — and the change shows up as a different number of weeks the doors can sell before they need the DC.

The reserve’s size mix deserves a check before anyone promises a top-up. The buy was placed on the core curve, 10/20/30/25/15, but the large-skew doors and ecommerce took their XL first. What is left in the DC is 126, 220, 276, 182 and 76 from XS to XL: the buy was 15% XL and the reserve is 8.6% XL. The large-skew doors that run out of XL first will find the reserve thinnest exactly there — and the reserve’s size mix is, in turn, evidence that the buy curve sat smaller than the doors and channels it was bought for.

There are styles that should carry no reserve. A launch with a fixed date and a short window has no time for a read. A door far from the DC on a slow delivery route may need its full allocation up front. And a style bought so thin that the minimums consume most of it has nothing above the floor to hold back — the minimum buy depth guide is the test for that case, and the answer there is fewer doors, not a smaller reserve.

When the buy is short: fewer doors or less cover

Not every buy covers every door. A style cut at the buy stage, or chased into a smaller second delivery, reaches allocation with less than the doors’ combined forecast, and the allocation has to decide what to give up. Take the same style bought at 1,500 units instead of 3,000: ecommerce takes 300 and the ten doors share 1,200.

Three options, and none is free. Spreading 1,200 units across all ten doors at four weeks of cover uses every unit and leaves no reserve, and a third of the push is minimums. Dropping the four C doors and holding four weeks in the six that remain leaves 320 units in the DC to correct with, and gives up the C doors’ 40 units a week of forecast demand — 20% of the store forecast — unless the reserve is later sent there. Keeping all ten doors at two weeks of cover holds 400 units back, but half of what is pushed is presentation stock that is not expected to sell.

The choice is between breadth, depth and the ability to correct, and a short buy cannot have all three. Which one to give up depends on what the doors are for. If the C doors carry the style mainly to complete an assortment, dropping them costs little; if they sell a size cluster nobody else does, it costs that whole cluster. The minimum buy depth guide sets the floor below which a size run breaks in every door, and below that floor, fewer doors is the only option that keeps the style presentable anywhere.

Reading the first two weeks and releasing the reserve

The read date was set before the push: end of week two. By then the A doors are selling 40 a week, exactly on forecast; the B doors 26 a week, 30% ahead of their 20; the C doors 6 a week, 40% behind their 10. The release rule, also set in advance: top each door up to its minimum plus six weeks at its actual rate, capped at capacity, and send nothing to a door already holding more than that.

Read the sell-through column alone and the A and B doors look the same: 33.3% and 32.5%. They are not. Sell-through measures units sold against units received, so it cannot tell a door selling to forecast on a capped allocation from a door outrunning its forecast on a thin one. The rate against forecast and the weeks of cover separate them: the A doors hold 3.0 weeks above their minimum at a rate that was expected, the B doors 2.6 weeks at a rate that was not.

The release follows the rule. Each B door’s target is 40 + 6 × 26 = 196 units; it holds 108, so it takes 88, and the four take 352. Each A door’s target is 40 + 6 × 40 = 280, capped at its 240 capacity; it holds 160 and takes 80, so the two take 160. The C doors hold 88 against a target of 40 + 6 × 6 = 76 and take nothing. The reserve sends 512 units and keeps 368. After the release the A doors have 5.0 weeks above the floor, the B doors 6.0 and the C doors 8.0.

A two-week read is noisy, and the rule has to be robust to it. A promotion, a local event, a spell of weather or a delivery that arrived a few days late can each move a door’s rate for a week. Read the rate over the door’s own selling days, set aside any week with a known one-off, and treat a door whose read sits close to its forecast as on plan rather than chasing the decimal. The release is a correction to the first allocation, not a second forecast made from two weeks of data, and the second read exists to correct the release in turn.

The C doors are the part the reserve cannot fix. They hold 12 units a door above six weeks of cover, which they will sell slowly or carry into markdown. The reserve’s value was that it never sent them more: had all 880 units been pushed on day one in proportion to the B and C doors’ forecasts — the A doors had no room — each C door would have received about 73 more, about 173 in all, and would now hold about 161 units against a rate of 6 a week — about 20 weeks of cover above its floor. The weeks of supply calculator runs the cover figure door by door, and the in-season reforecast is where a style running this far from its plan gets its remaining weeks re-projected.

Choosing the release rule

The rule that releases the reserve decides where it goes as much as the read does, and it has to be chosen before the read, or it will be chosen to fit whatever the read shows. Three candidates, each run on the same two-week read and the same 512 units:

Release in proportion to sell-through. Weighting each grade’s doors by their sell-through sends 139.5 units to the A doors, 272.0 to the B doors and 100.4 to the C doors: about 25 units each to four doors that need none, and about 68 to each B door against the 88 it needs. Sell-through rewards a door for selling a share of what it received, so a door that is selling slowly still collects.

Release in proportion to rate of sale. Better, because it follows units sold per week, so the B doors’ 26 a week counts for more than the C doors’ 6. But it ignores what each door already holds, and it still sends stock to a C door holding eight weeks of cover, because that door’s rate is above zero.

Top up to cover at the actual rate — the rule used above. It asks how many weeks each door can sell from what it holds, at the rate it is actually selling, and sends the difference to a target, capped by capacity. It is the only one of the three that sends nothing to a door that already has enough, and the only one that says how long the release will last. Its weakness is that two weeks is a short base for a rate: a door that had a promotion, an event or a weather spike in those weeks is topped up to a rate that will not hold, which is why the second read, a week or two later, runs the same rule on a longer window.

Multiple deliveries: when the second flow is the reserve

The worked example lands the whole buy at once and holds the reserve in the DC. A style bought in two or more deliveries gets a similar correction from its later flows without the DC holding anything. Had the woven shirt arrived as 1,800 units at launch and 1,200 four weeks later, the first flow would cover the 1,520-unit push and 280 units for ecommerce — about five and a half weeks at 50 a week — and the second would be allocated on a four-week read rather than on the proxy forecast.

The trade is timing and commitment. A reserve can be released any day after the read; a later flow arrives on the date its purchase order set months earlier, whether the read is ready for it or not. A flow can be re-phased only while the order is still open, and once it ships its total is fixed, although its door split is not. A later delivery is a reserve with a fixed arrival date, so its date has to fall after the first useful read and before the fastest doors run out — the window how to plan receipt flow sets out for phasing intake across a season.

Flows change the size question as well. A second delivery can be bought on a corrected curve if the first read arrives early enough to reach the vendor, which a reserve cannot do: the reserve’s sizes are whatever the first delivery brought. Where lead times allow it, the second flow’s size split is the cheapest place to fix a buy curve that sat too small or too large.

Carryover stock at the start of a season

A carryover style starts its new season with stock already in some doors and in the DC, and the allocation starts from that position rather than from zero. Each door’s opening stock counts against its new allocation: a door holding 30 units of a style whose model quantity is 160 needs 130, not 160, and a door holding more than its model takes nothing until it sells down.

Carryover stock is unevenly spread by construction. It is what did not sell, so it sits heaviest in the doors and sizes that sold slowest last season. Netting it door by door is what stops the new allocation from piling new receipts on top of old stock in exactly those doors. Consolidating it first — moving broken runs into fewer doors before the new receipts land — is the scheduled transfer the transfers section describes, timed before the push rather than after it.

The carryover’s position by door and size is an input, not a footnote, and the carryover’s own history is the best proxy the style will ever have: the doors where it sold through last season are the doors to grade it highest in this one, measured, as always, over the weeks it was in stock.

From allocation to replenishment

Some seasonal styles turn out to be steady sellers, and at some point continuing to allocate them by reads and releases is the wrong tool. A style that has sold at a stable rate across several reads, holding in-stock on its core sizes, is a candidate for a pull trigger: a reorder point per door, refilled from the DC as it sells, with the reserve becoming the DC’s replenishment stock.

The handoff needs a test rather than a feeling. The replenishment program guide sets out the graduation test a seasonal style has to pass before it moves onto a trigger, and the phase-out rule that ends it, because a style moved onto replenishment too early carries its launch weeks into the reorder point, and one moved too late spends weeks being corrected by hand. Allocation decides where a fixed quantity goes; replenishment decides how much more to send, and a style should be on one or the other at any moment, not both.

The worked example is not a candidate yet. Two weeks is one read, and the three grades are selling at three different rates against their forecasts. A third and fourth read that settle the B doors’ rate and confirm the C doors’ slow sell-down would be the evidence the graduation test asks for.

Allocating a new style with no history: like-for-like proxies

The forecast rates in the worked example had to come from somewhere, and the woven shirt is new: it has never sold in any door. Its door rates come from proxies — prior styles similar enough that the doors which sold them are likely to sell this one in similar proportion. A proxy supplies shares, not volumes: the plan supplies the chain rate, and the proxies say how it splits across doors.

Match proxies on several attributes at once — class, price band, silhouette, fabric weight, color family and launch window — rather than on one. A proxy matched only on class can carry a different price band into a door whose customer will not pay it; one matched only on price can carry a different season. Two or three proxies averaged are steadier than one, because any single style carries its own accidents: a late delivery, a missing size, a door that got a feature fixture.

The first proxy, a field jacket from the prior spring, sold 42% of its store volume in the A doors, 38% in the B doors and 20% in the C doors; the second, a chore coat from the prior fall, sold 38%, 42% and 20%. Averaged, 40%, 40% and 20%. Applied to the planned 200 units a week: 80 a week across the two A doors, 40 each; 80 across the four B doors, 20 each; 40 across the four C doors, 10 each — the rates the worked example used.

Every proxy rate has to be measured over in-stock weeks. A door that ran out of a proxy’s core sizes sold whatever was left, and its sales for those weeks describe its stock, not its demand. Door B2 sold 160 units of the first proxy over a ten-week window but was out of its core sizes for four of those weeks: 16 a week across the calendar, 26.7 a week across the six weeks it had stock. Taken at 16, B2’s share is understated, it is allocated less, it runs out again, and the error repeats into every season that uses this one as a proxy. The forecast accuracy guide covers the same censoring from the forecasting side.

Two further corrections. A door that opened after the proxy season has no proxy history at all; borrow the share of a sister door in the same grade and cluster, and mark it as borrowed so the first read is weighted accordingly. And a proxy allocated flat — or from a proxy of its own — carries that allocation’s errors in its door shares, which is one more reason the reserve exists: however good the proxies, the first allocation of a new style is a forecast, and the reserve is what turns the first two weeks of real selling into the second allocation.

Size-level allocation: the curve each door actually sells

A chain size curve is an average of doors that do not sell the same curve. Applied to every door, it is right on average and wrong in any door whose customers differ from the average, and the error does not show in the totals. Cluster the doors by the size curve they sell, and give each cluster its own curve. In the worked example, six doors sell the core curve — 10/20/30/25/15 from XS to XL — and four sell a large-skew curve, 5/15/30/30/20. Ecommerce sells its own, 8/17/28/27/20.

Every door and every size is a whole number, and the smallest position — 5 units of XS in a large-skew C door — still clears the two-unit minimum per size. The ten doors and ecommerce take 2,120 units between them; the buy was 3,000 on the core curve; the difference by size is the reserve described above.

Send a large-skew door the core curve instead and the failure is invisible until the fixture is looked at. Door B3 sells the large-skew curve at its forecast 20 a week. Sent 160 units on the core curve, it sells six weeks of demand — 120 units at its own curve — and is left with:

Forty units left: exactly the presentation minimum in total, and broken by size. The XL is sold out and the L is at 4 against a minimum of 12, while XS and S each hold 8 more than the door needs. The door will sell its remaining M and its stray L, hold its XS and S, and read at the season close as a style that stopped selling in week six — when the units it needed were in the four core-curve doors and in the DC.

Prepacks make the same error by construction. A prepack of 1-2-3-2-1 ships 11.1%, 22.2%, 33.3%, 22.2% and 11.1% from XS to XL in every pack. Eighteen packs to B3 is 162 units — 18, 36, 54, 36 and 18 — against the 8, 24, 48, 48 and 32 its curve calls for: 10 too many XS and 14 too few XL. A pack ratio is a size curve chosen for the DC’s convenience, so check it against each cluster’s curve before shipping in packs, and ship eaches, or a second pack ratio, to the clusters it does not fit.

The reserve is released by size on the same rule as by door. Each door’s target in each size is its minimum in that size plus six weeks of its own curve at its actual rate, and the DC sends the difference where it holds stock. With 76 XL left in the reserve, the large-skew doors’ XL top-up is the first line to run short — which is the point at which the second flow’s size split, or the next buy’s curve, is where the correction has to happen.

Rounding needs a rule written down rather than left to a spreadsheet. Round each door’s sizes so they sum to the door’s quantity, give any remainder to the sizes with the largest fractional parts, and never round the smallest size below its minimum — if the minimum cannot be met, the door takes a shorter run or the style goes to fewer doors. The size curve guide builds the curves, the size curve calculator splits a quantity by one, and the size curve template carries a size-percentage row that drives a size-units row per style and colour, for channel or region differences.

Allocating to ecommerce fulfilment alongside the doors

Ecommerce is a node in the allocation with its own forecast, its own size curve and its own failure modes, and it has to be allocated with the doors rather than after them. In the worked example it takes 600 units, 20% of the buy, from a planned 50 a week, on a curve of 8/17/28/27/20 — 48, 102, 168, 162 and 120 units from XS to XL.

The first decision is whether that share is ring-fenced or shared. A ring-fenced ecommerce share is simple to run and wrong by exactly the amount the online forecast misses; a shared pool corrects for that and needs a priority rule written down before the season. In a shared pool, the 880-unit reserve and the 600 online units are one 1,480-unit position in the DC, and the rule decides who is served first when it runs short: the online order already placed, the store top-up, or a split by forecast share.

Ship-from-store changes the arithmetic again, because a door’s stock becomes online availability. The door’s presentation minimum should not be: if ecommerce can sell a store’s last units, the fixture empties for online demand. Set the units a door exposes online as its stock on hand less its presentation minimum. A B door at the two-week read holds 108 units against a minimum of 40, so 68 of them are available to sell online — the available-to-sell figure, set by the allocation rather than left to the order system.

Online returns come back to stock at a lag, and some of them come back to a door rather than to the DC. A resaleable return is a receipt the allocation did not plan; counting it as available on the day it is processed, and to the node it lands at, is what keeps the reserve from being released twice for the same demand. The returns planning guide treats the inflow in full.

Wholesale commitments, where a style carries them, come off the top before ecommerce or any door: a booked order with a ship window is a promise, and an allocation that shorts it to protect a door has made a sales decision on the way past. The wholesale and DTC planning guide covers the scarcity rule for when the buy cannot cover both.

Telling an allocation error from a buying error

At the season close, a style bought right and allocated wrong produces the same markdown line as a style bought wrong. Both leave units unsold at the exit; both read as a style that underperformed. Separating them takes sell-through by door against the units each door received, never a chain total.

The chain total at the two-week read is 416 units sold of 1,520 pushed, 27.4%. No door is at 27.4%: the A doors are at 33.3%, the B doors at 32.5% and the C doors at 12.0%. The chain figure is an average of doors doing different things, and it describes none of them. A buyer reading only the chain figure sees a style selling a little slowly; a planner reading the doors sees a style selling to plan or better in six doors and failing in four.

The test is whether the doors that received the style sold it at the rate their forecast expected. If most doors sold to forecast and a few did not, the buy was about right and the allocation sent too much to the few — an allocation error, and next season’s grades and proxies are what change. If every door sold below its forecast in proportion, the allocation spread the style correctly and the chain demand was not there — a buying error, and next season’s buy is what changes. If doors sold to forecast but ran out of particular sizes while holding others, the error is in the size curve, and it is neither of the first two.

The three diagnoses need three numbers per door — units received, units sold, and the rate against its own forecast — plus sizes for the third. A planning report that holds only a chain sell-through and an end-of-season markdown cannot make the distinction, which is the same blindness the GMROI guide describes from the inventory side: a ratio over a total cannot see where the stock is.

When a transfer is worth it

A transfer moves stock that is already in one door to another. It pays for a pick and pack at the sending door, freight between doors, receiving at the other end, and the days the units spend in transit rather than on a fixture. It is the costliest of the corrections described here, and the only one left once the reserve is spent.

In the worked example, the C doors hold 12 units a door above six weeks of cover at their actual rate — 48 units in all — while the B doors need 352 to reach six weeks. A transfer of the 48 would cover about a seventh of the need and handle each unit twice; the reserve covers all of it from the DC in one move. A transfer is worth it when the reserve is empty and the units will earn more full-price weeks in the receiving door than they lose in transit, and not otherwise.

Transfers have a legitimate scheduled use late in a season: consolidating broken size runs from several doors into fewer, so the doors that keep the style present complete runs for the markdown. That is a planned move with a date, not a reaction. A rising count of ad hoc transfers in the middle of a season is a different signal — an allocation being corrected door by door because there was no reserve, or because the reserve was released on a rule that did not fit the doors.

Allocation at the end of the season

The presentation minimum is held on the floor while the style is selling at full price; by the end of the window it is the stock being cleared. Across the ten doors in the worked example that is 400 units — the minimums — plus whatever cover remains. When the doors stop holding a minimum and start selling it down is an exit decision, and it belongs to the markdown plan, which works its exit dates back from the season end.

Three allocation decisions sit at the exit. Which doors keep the style through its markdown, and which send their remaining units to them, so the doors that stay present complete runs. Whether the reserve still in the DC goes out to the doors for the markdown or goes to an outlet, a clearance channel or the next season’s carryover. And which sizes are worth consolidating, because a door holding three XS and nothing else has no presentable offer at any price.

The end of the season is also where the next allocation’s inputs are made. Door sell-through against units received, rates measured over in-stock weeks, the sizes each door ran out of first — all of it becomes next season’s grades, clusters and proxies. An allocation that kept those records by door and size gives the next one something to start from; one that kept only a chain total starts the next season from a flat split again.

Scoring the allocation afterwards

An allocation is judged twice: at the first read, where the reserve corrects it, and at the season close, where its inputs for next season are made. The close needs measures that separate the allocation from the buy and from the size curve, kept by door. Six measures, each read by door rather than as a chain total, are enough to tell the three apart.

Score at two points. At the first read the scorecard is short — sell-through, rate against forecast and cover by door — because its job is to drive the release. At the season close it is complete, because its job is to correct the inputs. A scorecard kept only at the close cannot show whether the reserve was released in time; one kept only at the first read cannot show whether the size curve held to the exit. Keeping both, by door and by size, is what lets a season’s allocation teach the next one anything.

None of the six has a target in this guide, and none should be borrowed: each is read against the plan the allocation was built on and against the same style’s other doors. The scorecard’s job is to send each miss to the input that caused it — the grade, the proxy, the cluster curve, the reserve, the release rule — so that next season’s allocation starts from a corrected input rather than from the same one with a note attached.

Who owns which input

An allocation draws on inputs owned by different people, and an input nobody owns is the one that goes stale. Written down, the ownership looks like this:

• Grades and clusters — merchandise planning, re-cut each season from in-stock-adjusted history.
• Presentation minimums — visual merchandising, per fixture type, with a minimum per size.
• Capacity — store operations, per door and category, updated when fixtures or stockrooms change.
• Size curves — planning, per cluster, measured over in-stock weeks.
• Forecast rates and proxies — planning for the chain rate, the allocator for the split across doors.
• Read date, release rule and reserve — the allocator, written down before the push.
• Transfers and consolidation — the allocator with store operations, on a schedule.

Writing the owner next to each input is what turns an allocation from a spreadsheet one person maintains into a plan a team can check. When a door runs out, the list says which input to question first: a capacity figure that was stale, a minimum set too low, or a curve that was never re-cut for that door’s cluster.

Seven ways allocation errors show up

None of these appears as an allocation line on any report. Each shows up as something else — a stock-out, a transfer, a markdown — in a door and a size, weeks after the allocation that caused it.

What the unit of allocation is in each vertical

The method is the same everywhere. What changes is the unit being allocated, what sets the presentation minimum, and what binds first — demand, capacity or the calendar — so an allocation rule is written per vertical rather than borrowed. Apparel is the reference case. The allocation and replenishment by vertical guide on RetailNorthstar works the replenishment side the same way.

• Sell-through calculator — the first number the reserve release is decided on →
• Weeks of supply calculator — cover by door at the actual rate →
• Size curve calculator — split a door quantity by its curve →
• How to calculate size curves — the curves each cluster runs on →
• Minimum buy depth for a size run — when a buy cannot reach every door →
• How to plan a replenishment program — continuity stock after the first fill →
• How to plan wholesale and DTC together — one buy, two demand signals →
• How to plan for returns — the inflow the allocation did not plan →
• How to build a buy plan — the committed quantity allocation starts from →
• Merchandise allocation — glossary definition →
• Size curve — glossary definition →
• Replenishment — glossary definition →
• Size curve template →
• Allocation and replenishment for apparel on RetailNorthstar →
• The fair-share allocation formula on RetailNorthstar →

Frequently asked questions

What is merchandise allocation?

Merchandise allocation is the process of splitting a committed buy across stores, sizes and channels — which doors receive each style, in what depth, in which sizes, and when. It happens after the buy is placed, so it moves units between places and never changes how many there are. The first allocation is a forecast, because no sell-through exists yet; the reserve held back from it is what corrects that forecast once the first weeks of selling have been read.

What is the difference between allocation and replenishment?

Allocation distributes a new buy before any sell-through exists, so it runs on a forecast: grades, proxies, presentation minimums and capacity. Replenishment refills continuity stock to a model or reorder point as it sells, so it runs on the actual rate of sale. A seasonal style uses both in sequence — an initial push, then a reserve released on the first read — while a core style lives on replenishment.

How do you allocate merchandise to stores?

Take the committed buy, carve out wholesale commitments and the ecommerce share, then give each door its presentation minimum plus a set number of weeks of forecast demand, capped at what it can hold, and split that quantity into sizes with the door’s own size curve. Keep the rest in the DC as a reserve. In this guide’s example, 3,000 units become 600 for ecommerce, 1,520 pushed to ten doors and 880 held in the DC.

How much stock should you hold back in a DC reserve?

There is no universal share. The reserve should be large enough to correct the first allocation — sized by how uncertain the forecast is, how quickly the DC can reach a door, and how many full-price weeks are left after the first read — and small enough that the doors can present and sell from day one. In this guide’s example the reserve is 880 of the 2,400 store units, 36.7%, and the first read releases 512 of it.

How do you allocate a new style with no sales history?

Borrow door shares from like-for-like proxy styles — the same class, price band, silhouette and launch window — and apply them to the new style’s planned chain rate. Measure each proxy over in-stock weeks only: a door that sold 160 units over ten weeks while out of its core sizes for four of them sold 26.7 a week when it had stock, not 16. Then hold a reserve, because a proxy is still a forecast.

What is the difference between push and pull allocation?

Push allocation sets door quantities centrally from a forecast and ships them; pull allocation lets each door’s own sales trigger what it receives, through a reorder point or model stock. Push fits new styles, launches and short windows where no sales exist yet; pull fits continuity styles with steady demand. A seasonal style with a selling window long enough for one read can use both: push an initial allocation, then pull from a DC reserve once sell-through is visible.

How do you allocate sizes to stores?

Split each door’s quantity with the size curve that door actually sells, grouping doors into size clusters rather than applying one chain curve. In this guide’s example, a door that sells a large-skew curve but is sent the chain curve ends six weeks holding the right total — 40 units — with its XL sold out and its L at 4 against a minimum of 12.

What is a presentation minimum in allocation?

A presentation minimum is the number of units a door needs to present a style as a deliberate offer — a full fixture face, with at least a minimum in every size it carries. It sits on the floor rather than counting as cover. In this guide’s example it is 40 units a door, with at least two in the smallest size, and the ten minimums are 400 of the 1,520 units pushed.

How do you grade stores for allocation?

Rank doors by their sales in the category over a comparable season, measured over in-stock weeks, and cut the ranking where the volume drops rather than at a fixed number of doors per grade. Grade per category, because a door can be high-volume in one category and low in another, and cluster separately on size profile and climate. In this guide’s example, ten doors fall into two A doors, four B doors and four C doors at the two largest gaps in the ranking.

How do you know if an allocation was wrong?

Read sell-through by door against the units each door received, and the rate of sale against each door’s forecast, rather than a chain total. In this guide’s example, A doors sell through 33.3% and B doors 32.5% in two weeks — nearly identical — but A is on forecast and capacity-capped while B is running 30% ahead of forecast with less stock behind it. The chain figure, 27.4%, describes no door at all.