How to build a margin bridge

IMU, maintained markup and realized gross margin are not the same number

The most common reason a margin conversation goes nowhere is that three people are quoting three different measurements and all calling it margin. They are not synonyms and they are not approximations of each other. They are taken at three different moments in the life of the same units, and each absorbs a different set of events.

Initial markup is a design decision, maintained markup is a trading outcome, and realized gross margin is an accounting fact. Initial markup is retail less cost over retail, struck at costing. It is knowable months before the season and never changes afterwards, because both of its inputs are fixed at the point the order is placed. Maintained markup measures the units that actually sold, at the prices they actually achieved, against the cost of those units — so it absorbs every price reduction and nothing else. Realized gross margin is net sales less the cost of everything consumed, which means it also absorbs the units that produced cost and no sale: the ones stolen, damaged, written off, or returned in a state that could not go out again.

Below is one style-color measured all three ways. The style is bridged in full further down; these are the three endpoints of that bridge.

Seven and a half points separate the first number from the last, and each of the two gaps has a different explanation. The six points between IMU and maintained markup are price reductions plus a cost overrun — trading events, visible weekly, argued about constantly. The remaining point and a half is the cost of forty units that were bought, paid for and freighted and never sold, less a small vendor credit: $22,400 of consumed cost against $21,696 of cost of goods sold, a difference of $704 on $47,190 of net sales. Nobody watches that gap weekly, which is exactly why it needs its own rows on the bridge. The formal definition of the endpoint sits in the gross margin glossary entry, and the entry price that sets the top of the bridge can be checked on the IMU calculator.

Ten steps, in order

The two steps most often skipped are the third and the eighth — publishing the order of the peels, and tying the bottom line to the ledger. Without the first, this season’s bridge is not comparable to last season’s. Without the second, it is a hypothesis.

1. 1

   Start from the costed margin, not from last year

   The top of the bridge is the margin the range was costed to: units bought, at ticket, at costed landed cost. Starting from last year’s realized margin buries the whole of this year’s costing decision inside the opening line, and no later line can recover it.

2. 2

   Fix the base and the denominator, and write both on the bridge

   Choose the value everything is measured against — ticket value of the units bought is the only base that keeps the peels additive — and state the denominator each rate uses. A bridge that changes denominator halfway down still ties in dollars and lies in percent.

3. 3

   Decide the order of the peels and publish it with the bridge

   The sequence is a convention, not a fact. It changes the size of every line it touches and leaves the total unchanged. Fix the order once, write it at the top of the bridge, and use the same order next season or the two are not comparable.

4. 4

   Peel landed-cost variance first, and state the capitalisation policy

   Compare actual landed cost to costed landed cost and multiply by units received, not units sold, because the overrun was paid on everything that arrived. Alongside it, state whether inbound freight and duty are capitalised into inventory or expensed in the period, because that choice moves realized gross margin without changing a single physical fact.

5. 5

   Peel the price reductions that were actually taken

   Value the units that stayed sold at ticket, subtract what they actually rang at, and the difference is the markdown and promotional line. Take it from the sales side only. A reduction is a price event, so nothing about cost of goods moves with it, and a bridge that adjusts both sides has counted it twice.

6. 6

   Peel the units consumed without producing a sale

   Shrink, non-resaleable returns and units sent back to the vendor all leave their cost in the ledger while contributing no revenue. Carry each as its own line, and carry the vendor credit against the RTV line rather than netting it into cost of goods where it disappears.

7. 7

   Peel the deductions the customer took

   Chargebacks, co-op advertising and wholesale markdown money reduce what you collect against an invoice you have already raised, so they sit on the sales side. Accrue the negotiated ones from the contract rather than discovering them in a remittance, and keep compliance deductions on a separate line, because the two have different owners and only one is preventable.

8. 8

   Tie the bottom of the bridge to the ledger before anyone sees it

   The last line must equal realized gross margin as reported, to the dollar. A bridge that lands near the reported number has a missing line in it, and the difference will be argued about instead of the erosions.

9. 9

   Rebuild the same bridge at style-color level, then aggregate upward

   A class-level bridge attributes a total to causes no one can act on, and offsetting errors inside the class cancel before they are visible. Build at style-color, aggregate for the review, and keep the style detail attached to the aggregate.

10. 10

    Put a name against every line before circulating it

    Each line needs one owner who can change the number next season. A line with no owner becomes the plug: every unexplained dollar drifts into it, it grows every season, and nobody is accountable for the growth.

The erosions, in the order they hit

The bridge below has seven lines, and they arrive in a rough chronological order: cost lands, price is reduced, units go missing, units come back, the customer deducts. That is not the only defensible sequence, but it is the one that maps to the calendar, and a sequence that maps to the calendar is easier to argue about with the people who own the lines.

Line one, landed-cost variance, is the difference between what a unit cost to get into stock and what the costing sheet said it would. It is inbound freight, duty, brokerage and any charge that attaches to the goods before they are available to sell. The variance is actual landed cost less costed landed cost, multiplied by units received — not units sold, because the overrun was paid on everything that arrived. It carries a second question that is not a trading question at all: whether those inbound costs are capitalised into inventory or expensed in the period they are incurred. Capitalise them and they depress gross margin as the units sell; expense them and gross margin rises while operating expense rises by the same amount. Total profit is identical either way, which is precisely why the policy has to be written on the face of the bridge.

Line two is markdown and promotional dollars taken: the ticket value of the units that stayed sold, less what those units actually rang at. It is a sales-side line only. Nothing about cost of goods moves when a price is reduced, and a bridge that adjusts both sides for a markdown has counted it twice. Keep permanent markdowns and temporary promotional reductions separable within the line — one is a decision to exit, the other a decision to accelerate, and they are usually made by different people for different reasons. What a reduction has to sell to pay for itself is a separate calculation, and one worth doing before the reduction rather than after it on the break-even markdown calculator.

Lines three, four and five are the units that produced cost and no sale. Shrink and inventory adjustments leave the cost of the unit in the ledger with nothing against it. Customer returns reverse a sale: where the unit re-enters saleable stock the erosion is limited to the reduction it will later need, and where it does not — damaged, opened, out of season, out of date — the sale reverses and the cost stays consumed. The refund allowance for returns still in flight belongs here too, net of the units expected to be resaleable. Return to vendor is the mirror image: units travel back up the chain and a credit comes down, but the credit is normally struck at FOB, so the freight and duty already paid on those units stay with you. The RTV line is almost never a wash, and it should never be netted into cost of goods, where the size of the underlying defect problem disappears.

Lines six and seven are the wholesale half: chargebacks, and co-op advertising and markdown money. Both reduce what you collect against an invoice you have already raised, so both sit on the sales side. They should not share a line. A chargeback is a penalty for a compliance failure — a late delivery, a wrong carton mark, a routing or ASN error — it is preventable, it is disputable inside a window, and it is owned by logistics and sales together. Co-op and markdown money are negotiated terms: known in advance, accruable from the contract, owned by whoever agreed them. Booking them together makes the preventable half invisible.

The arithmetic on those two lines is worth spelling out, because it is where bridges most often go wrong. Take a wholesale style invoicing 2,000 units at $30.00, so $60,000 of gross invoiced sales. Co-op runs at a contractual two percent of invoiced value: $1,200, accruable the day the invoice is raised. The account then runs a promotion at ten percent off on 1,200 units and the brand funds it — 1,200 units at $30.00, times ten percent, is $3,600 of markdown money. Compliance deductions across the season came to $1,800. The three lines total $6,600 against $60,000 invoiced, eleven points of invoiced value, and only one of the three was known when the order was booked. These are illustrative inputs chosen to divide cleanly, not typical rates; the split between the negotiated lines and the preventable one is the part worth carrying across.

One style-color, from costing sheet to ledger

A single style-color, sold through the brand’s own retail and direct channels. One thousand units received, one ticket price, one season. Start with the costing sheet and what it turned into.

The top of the bridge is the costed margin on everything bought: 1,000 units at a $55.00 ticket is $55,000 of ticket value, against 1,000 units at the $22.00 costed landed cost, or $22,000 of cost. Costed margin is $33,000, and expressed as a rate it is exactly the initial markup, 60.0% — which is the point of choosing this base: the top of the bridge and the costing sheet agree by construction.

Now the unit account, which has to reconcile before the money does. Of the 1,000 units received, 960 shipped and stayed sold; 10 were lost to shrink and found at the physical count; 20 shipped, came back from customers and were not resaleable; 10 were found defective and returned to the vendor for credit. Those four numbers sum to 1,000, and every unit sits in exactly one bucket. A bridge whose unit account does not close will not close in dollars either, and it is far cheaper to find the discrepancy in units.

The 960 units that stayed sold rang at three prices: 600 at the $55.00 ticket, 210 at $44.00 after a twenty percent reduction, and 150 at $33.00 after a forty percent reduction. That is $33,000 plus $9,240 plus $4,950, so net sales realized are $47,190. At ticket those same 960 units would have been $52,800, so the markdown and promotional line is $52,800 less $47,190, or $5,610. It checks from the other side: 210 units at $11.00 off is $2,310, and 150 units at $22.00 off is $3,300, the same $5,610.

The remaining lines follow from the unit account. Landed cost came in sixty cents above the costed figure on all 1,000 units received, so the landed-cost variance is $600. The 10 shrink units forgo their ticket value, $550, while their cost stays in the ledger. The 20 non-resaleable returns reverse $1,100 of ticket value on the same basis. The 10 RTV units forgo $550 of ticket value and bring back a $200 vendor credit struck at FOB — the credit does not reach the $0.60 per unit of freight and duty already paid on them — so the RTV line is $350 net.

The bridge closes at $24,790 of realized margin, and that figure can be reached independently: net sales of $47,190 less cost of goods of $22,400 — 1,000 units at $22.60, less the $200 credit — is $24,790. The two routes to the same number are what make it a bridge rather than a narrative. If they do not agree, a line is missing, and the missing line is always more interesting than the ones already on the page.

One more thing has to be said out loud, because it is the single most common source of argument in a margin review. The points column runs to 45.07%, because every peel is expressed against the $55,000 ticket base. But realized gross margin, as the ledger reports it, is $24,790 over net sales of $47,190 — 52.5%. Those two rates differ by seven and a half points and both are correct. The gap is not an erosion anybody caused; it is the denominator changing from the ticket value of everything bought to the net sales actually realized. A bridge that shows only one of them invites a meeting about arithmetic. Show both, label them, and the meeting is about the erosions instead.

The order of the peels changes the size of every line

A margin bridge shares a property with every other decomposition: the total is invariant to the order of the peels, and the individual lines are not. Change the sequence and the same dollars move between lines. Nothing about the business has changed, and the story the bridge tells has.

The example above makes it concrete. The 20 non-resaleable returns were carried at their $55.00 ticket value, $1,100, because the base itself is carried at ticket. Suppose instead the returns line is valued at what was actually refunded — those units had sold on promotion at an average of $47.00, so the refunds were 20 times $47.00, or $940. The returns line falls to $940. The missing $160 does not vanish: it is a price reduction that was taken on those units, so it belongs in the markdown line, which rises from $5,610 to $5,770. In points of the base, the markdown line moves from 10.20 to 10.49 and the returns line from 2.00 to 1.71 — the pair still totals 12.20 points, and the bottom of the bridge is unchanged at $24,790.

Neither convention is more correct than the other. What is wrong is not stating which was used. A bridge circulated without its sequence gets rebuilt from memory next season, the convention drifts, and a line moves by three tenths of a point for a reason that has nothing to do with trading — which somebody then explains as a returns improvement. The remedy costs one line of text: write the order of the peels and the valuation basis of each at the top of the bridge, and keep both stable across seasons. Where the same erosion could plausibly land on two lines, say which one owns it and why.

The same warning applies on the sales side of a plan miss, where volume, rate, mix and timing are peeled in a sequence that is equally conventional and equally load-bearing. Publish the sequence there for the same reason. A business with one documented order for its sales decomposition and another for its margin bridge can at least reconcile the two. A business with neither is comparing this season’s attribution against last season’s method and calling the difference performance.

The class bridge ties, and both styles inside it are wrong

The second example is the reusable mechanism of this page, and it is why the level a bridge is built at matters more than the lines it contains. Take a class holding two styles. Style A is the style-color bridged above, selling through the brand’s own channels. Style B is a wholesale style in the same class, planned to a lower rate because it sells at wholesale price and carries the deduction lines. Each has its own plan, line by line, and each is compared against it.

Read the class column and the season looks controlled. Landed cost is a hundred dollars favourable, markdowns a hundred and fifty adverse, shrink and returns immaterial, wholesale deductions fifty adverse. Total variance to plan: zero, on $40,990 of planned realized margin. There is nothing in that column to act on and nothing to escalate. Every line passes, and the two styles inside the class are wrong on four lines each.

Style A landed $1,800 below its own plan — the $24,790 the bridge above closed at, against a plan of $26,590, a shortfall of 6.8% of its planned margin. It gave back $2,400 more in markdowns than it planned and $900 more on returns, offset by $1,200 of favourable landed cost and $300 of shrink. Style B landed $1,800 above plan — $16,200 against $14,400, or 12.5% better — with the signs reversed on every line. The two are not related. They sell to different customers, they were bought by different people, and their errors have no common cause. They simply happen to be the same size and the opposite sign, and to sit in the same class.

Cancellation is not a rare accident; it is the expected behaviour of aggregation. Errors at style level are largely independent, so summing them shrinks them relative to the base — and the more styles in the class, the more thoroughly the individual errors disappear into each other. A class bridge is therefore at its most reassuring exactly when it is least informative. The class in this example would pass a review, take a green status, and hand both buyers the same instruction for next season: repeat what you did.

There is a second, subtler version of the same problem. Even when a class-level variance is large enough to notice, it attributes the total to a different cause than the style-level bridge does. The $150 class markdown variance in the table is the residue of a $2,400 error and a $2,250 error pointing in opposite directions; a bridge built directly at class level cannot see either of them, so it names markdowns as the class’s issue at a size that implies no action. The rule that follows is simple and expensive to ignore: build at style-color, aggregate for the review, and keep the style detail attached so whoever reads the class number can open the styles behind it. The class view is still how the number reaches finance — but it is an aggregation of bridges, not a bridge.

One owner per line, or the line becomes the plug

A bridge that names mechanisms but not people is an interesting document that changes nothing. The reason to size each erosion separately is that each one is produced by a different function, on a different timescale, using a different lever. Sourcing cannot do anything about markdown depth; planning cannot do anything about carton marking; nobody at all can do anything about a line with no name against it.

The failure mode is specific and it is always the same: one line ends up unowned, and that line becomes the residual. It starts as the honest remainder — timing differences between the inventory system and the ledger, an adjustment posted to the wrong period, a batch of deductions that could not be matched to shipments. Because nobody owns it, nobody reduces it. Because nobody reduces it, everything else unattributable drifts into it. Within a few seasons it is one of the larger lines on the bridge and the only one with no explanation, at which point the bridge as a whole stops being trusted and the organisation goes back to quoting two numbers and a sentence.

Two rules keep that from happening. First, no line goes on the bridge without a name against it, even when the assignment is arguable — an arguable owner produces an argument, which is a better outcome than silence. Second, cap the unexplained residual explicitly: if it exceeds a threshold the business sets in advance, the bridge is not published until the excess is traced. That threshold is a governance decision rather than an arithmetic one, and agreeing it before the first bridge is built is the single most useful thing a finance and planning team can do together.

Same waterfall, different dominant line

The structure of the bridge does not change between categories. What changes is which line carries most of the erosion, how volatile that line is, and — in one case below — whether an additional line exists at all. No rates appear here, because which line dominates is a structural property of the category while its size is a property of your business.

Apparel and footwear share a shape: the wholesale half of the bridge does most of the damage, and it does it through lines that were agreed rather than lines that went wrong. Markdown money and co-op are negotiated at the account months before the season, and they scale with invoiced value whether or not the product sells — which means they are forecastable, and a business that treats them as a surprise has chosen to. Chargebacks are the exception in that pair: they are failures, not terms. Footwear carries more of them than apparel because prepacked size runs multiply the number of things a carton can get wrong, and the practical consequence is that a footwear brand should size its chargeback line per shipment and per door rather than as a percentage of sales. Expressed as a percentage it looks stable while the door count doubles underneath it.

Accessories and bags invert the emphasis. Freight per unit is small, because the goods are dense and light relative to their value, but duty is not small, and the duty rate turns on material composition and construction — precisely the things still being decided in development after the ticket price has been set. A substitution made to hit a cost target can change the classification and move the landed cost of a style already costed, and a reclassification applied retroactively lands as a variance against units that have already sold. For this vertical the landed-cost line needs a duty sub-line and a second review point after the final materials are locked, not just the one at costing. The mechanics of assembling that number sit in the landed cost calculator, and the same discipline applies when the range is developed in-house rather than bought, which is the case worked through in how to plan a private label range.

Home and furniture makes freight the volatile line rather than the stable one. Freight is charged on cube, so a large low-value item can carry a freight cost that rivals its FOB, and the costed allowance has to be set per style rather than as a blended percentage of cost — a blended percentage systematically over-costs the small dense items and under-costs the bulky ones, which then arrive as a variance that looks random. Damage is the second difference. A damaged sofa is not a markdown candidate the way a damaged tee shirt is, because it usually cannot be brought back to first quality, so it lands on the write-off or RTV lines instead of the markdown line. Customer returns behave the same way: the reverse leg is freight-constrained, and where recovery costs approach recoverable value the unit is disposed of, so the bridge carries a revenue reversal and a disposal cost with no inventory credit at all. How much of that inflow is inventory and how much is pure margin adjustment is the question worked through in how to plan for returns.

Health and beauty is the one vertical in this list that needs an extra line. Short-dated and expired stock is an erosion with no equivalent in apparel: past its date the unit cannot be sold at any price, so the clearance route that recovers something from aged apparel recovers nothing here. That changes the bridge in two ways. The write-off line has to be forecast from date profiles rather than treated as a residual, because unlike shrink it is knowable in advance — you know at receipt when a batch stops being sellable. And the markdown line acquires a deadline: a reduction taken late enough is not a reduction, it is a write-off with extra steps. Returns behave the same way for hygiene reasons, since an opened unit reverses the sale and yields no saleable inventory, so it belongs with the non-resaleable returns line and never with the resaleable share.

Seven ways a margin bridge goes wrong

Why teams go back to comparing IMU against gross margin

A margin bridge is easier to build once than to keep. The default it decays back to is familiar: quote IMU at the start of the season, quote realized gross margin at the end, describe the gap in a sentence. Four forces pull it that way, none of them laziness, and knowing which one is acting on you is the difference between a bridge that survives a season and one that survives a meeting.

The first reason is that the bridge needs data from systems that do not know about each other. Costed landed cost lives in the costing sheet, actual landed cost in the freight and customs paperwork, markdowns in the pricing system, shrink in the inventory adjustments ledger, returns in the ecommerce and warehouse systems, and deductions in accounts receivable. Joining those six for one season is a project. Joining them at style-color, in a form that can be rebuilt every month, is a different order of work — and it is exactly the work that gets deprioritised the moment a season goes badly and everybody is busy trading.

The second reason is political, and it is the one nobody writes down. Every line has an owner, and the owner is being measured by it. A document that sizes each function’s contribution to a margin shortfall in dollars is, from inside the organisation, a document that assigns blame with decimal places. The predictable responses are to argue the sequence, to argue the allocation of shared costs, and to argue that the line is not controllable — all of which are sometimes true, and all of which are cheaper than reducing the line. A bridge circulated without prior agreement on the sequence and the ownership map produces one meeting and no second version.

The third is timing. Bridges are usually built at season close, when the season is finished and nothing on it can be changed, so the exercise feels like an autopsy rather than a control. A bridge built monthly in season is a different instrument: landed-cost variance is known at receipt, the markdown line accumulates weekly, the deduction accrual can be posted at shipment, and the only genuinely late line is the returns tail. Most of the bridge is knowable while there is still something to do about it, and the version built at close inherits none of that value.

The fourth is that nobody owns the reconciliation itself. Each line has an owner, but the bridge as an artefact usually does not, so when the person who built the spreadsheet moves on, the method leaves with them. Next season it is rebuilt from memory, the sequence changes, the base changes, and the new version cannot be compared to the old one — at which point the comparison that survives is the one needing no method at all: IMU at the start, gross margin at the end, a sentence in between. The way out is unglamorous. Name an owner for the bridge itself, fix the sequence and the base in writing, build it monthly rather than at close, and keep it at style-color so the number it produces is attached to something a buyer can act on. Where the plan, the costing, the receipts, the price history, the returns disposition and the deductions sit on one data model, most of that happens without anyone maintaining a spreadsheet — which is the only version of this that survives more than two seasons.

• ✓A margin bridge runs from the margin you costed to the margin you banked, one line per erosion and one owner per line. It is only a bridge if the last line ties to the ledger to the dollar.
• ✓IMU, maintained markup and realized gross margin are three measurements taken at three moments. On one style they can be seven points apart and all three correct.
• ✓Take the peels in the order the erosions hit: landed-cost variance, price reductions, units consumed without producing a sale, then the deductions the customer took.
• ✓The sequence changes the size of every line it touches and never changes the total. Publish the order and the valuation basis, or next season is not comparable to this one.
• ✓Build at style-color. At class level offsetting errors cancel — the class bridge can tie on every line while both styles inside it need a different decision.
• ✓Which line dominates is a property of the category: markdown money and chargebacks in apparel and footwear, duty in accessories and bags, freight and damage in home and furniture, and short-dated write-offs in health and beauty, which is a line apparel does not have.
• ✓Every line needs a name against it. The unowned line becomes the residual, absorbs everything unattributable, and grows until the bridge is no longer trusted.

• IMU calculator — set the number at the top of the bridge →
• Landed cost calculator — build the cost the bridge starts from →
• Markdown calculator — size the largest peel on most bridges →
• Break-even markdown calculator — what a reduction has to sell to pay for itself →
• How to plan for returns — the assumption behind the returns line →
• How to plan wholesale and DTC together — where chargebacks and markdown money come from →
• How to plan a private label range — where IMU is set rather than inherited →
• Retail math formulas — the definitions the bridge lines are built from →
• Planning margin on a moving cost base, on RetailNorthstar — costing a margin when the cost itself will not sit still →

Frequently asked questions

Why is actual gross margin lower than planned margin?

Because the planned figure is an initial markup struck at costing, and a sequence of reductions sits between it and the ledger: cost landed higher than it was costed, price reductions were taken to clear the units, some units were lost to shrink, some came back and could not be resold, and on wholesale business the customer deducted chargebacks and allowances from the invoice. Each of those is a separate mechanism with a separate owner. A single number for the gap says nothing useful. A bridge that names each reduction, sizes it in dollars and ties to the reported margin turns the gap into a list of decisions.

What is the difference between IMU, maintained markup and gross margin?

They are three measurements taken at three points in the life of the same units. Initial markup is retail less cost over retail, computed at costing, before anything happens — it says what the range was designed to earn. Maintained markup is what the units that actually sold earned at the prices they actually got, so it reflects markdowns and promotional reductions but not the units that never produced a sale. Realized gross margin is net sales less the cost of everything consumed, so it also absorbs shrink, write-offs, non-resaleable returns and vendor credits. On a single style these three can be six or seven points apart and every one of them is correct. The error is using them interchangeably, because they answer different questions and belong to different owners.

What erodes gross margin in retail?

In the order the erosions usually hit: landed-cost variance, where inbound freight and duty come in above the allowance carried at costing; markdown and promotional dollars, the reductions taken against the ticket price to move the units; shrink and inventory adjustments, where cost stays in the ledger and no sale appears against it; customer returns, where the sale reverses and only the resaleable share returns as inventory; wholesale chargebacks for compliance failures; co-op advertising and markdown money negotiated into the trading terms; and return-to-vendor activity, where the credit received rarely covers the freight and duty already paid. Which of these dominates is a function of the category rather than of how well the business is run.

Should freight be in cost of goods or in operating expense?

Inbound freight and duty are ordinarily capitalised into inventory cost, so they reach the ledger as cost of goods when the units sell; outbound and last-mile shipping is ordinarily an operating expense. The point that matters for a bridge is that the choice moves gross margin percent without changing profit. Capitalise the inbound cost and it depresses the margin rate; expense it and the rate rises by the same amount operating expense does. Confirm the policy with finance, write it on the face of the bridge, and never compare a margin rate across years, entities or systems without checking that both sides treat freight the same way.

How do chargebacks and markdown money show up in margin?

Both reduce what you collect against an invoice you have already raised, so they sit on the sales side of the bridge rather than in cost of goods, and both usually arrive as deductions on a remittance weeks after the shipment. They are different animals and belong on separate lines. A chargeback is a penalty for a compliance failure — late delivery, wrong carton marking, a routing or ASN error — and it is disputable within a window, so it is owned by logistics and sales together. Co-op advertising and markdown money are negotiated terms, known in advance, and should be accrued from the contract at the point of shipment rather than discovered in a deduction. Booking both into one line hides the fact that one of them is preventable and the other is priced in.

Who owns a margin variance?

Nobody owns the variance; each line of the bridge has an owner, and that is the point of building it. Landed-cost variance belongs to sourcing and logistics. Markdown dollars belong to planning in season and to buying for the depth and entry price that made the reduction necessary. Shrink belongs to warehouse and retail operations. Returns split three ways — planning owns the assumption, merchandising owns the driver, operations own disposition. Chargebacks belong to logistics and sales jointly, allowances to sales alone, RTV credits to sourcing and quality. A line with no name against it becomes the residual: it absorbs everything unattributable and grows every season with nobody accountable for it.

At what level should a margin bridge be built?

Build it at style-color and aggregate upward for the review. A bridge built directly at class or department level attributes the same total to different causes than the style-level bridge does, and it hides offsetting errors: a style that over-marked and a style that under-marked cancel inside the class, so the class line shows a small variance while both styles need a different decision next season. Only the style-color bridge names something a buyer or planner can act on. The class view is still worth having — it is how the number reaches finance — but it should be an aggregation of style-level bridges rather than a bridge of its own.