Decomposing a Sales Plan Miss: Volume, Rate, Mix and Timing

What a sales bridge is

The word that matters in that definition is attributes. A variance number is a measurement; a bridge is an attribution. Plan versus actual variance tells you a class finished the month down 19 percent. It does not tell you whether fewer people bought, whether the same people bought cheaper things, whether the assortment shifted underneath the total, or whether the demand is simply four weeks behind schedule. Those four situations produce the same headline and demand four different responses, and three of the four responses are wrong in any given case.

A bridge exists to make that choice explicit. It converts one number into a short ordered list of numbers that sum back to it, where each line has an owner and a lever attached. The output of a good bridge is not an explanation, it is an assignment — this many dollars belong to the buy, this many to the price, this many to the assortment, this many to the calendar.

This guide covers the sales and demand side only. A gap in realized margin is a different bridge with a different set of effects, and building the two into one waterfall is how a review ends up unable to tell a pricing decision from a costing one. The margin half of the question — initial markup against realized margin, and the effects that sit between them — is a margin bridge, and it is a separate artifact with its own effects, its own sequence and its own owner. Run it separately and reconcile the two at the end. Nothing below builds one — the margin bridge is built line by line, with its arithmetic and its owners, in how to build a margin bridge on retail-plan.com.

The four effects

Three of these four are computed inside the period. Timing is different in kind: it is not a component of the gap so much as a question about whether the gap is real yet. It is included in the list because leaving it out is the single most expensive omission in variance work, and because the first thing a planner should do with a bridge is ask whether the effects it just named are permanent or merely early.

The three in-period effects are not independent of each other, and this is the source of most of the trouble that follows. Units and price interact. Mix is a statement about how units are distributed, so it overlaps with volume by construction, and because different options carry different prices it overlaps with rate as well. A bridge is a procedure for cutting through those overlaps in a defined order, not a set of independent measurements.

Building the bridge, one step at a time

The worked example below runs on one class for one month. All figures are illustrative, chosen because they divide cleanly. They are not benchmarks and are not drawn from any brand.

The class carries three styles. The month plan and the month actual look like this.

The headline is a gap of 81,000 minus 100,000, or minus 19,000 dollars, which is minus 19.0 percent against a plan of 100,000. Units are down 200 on 2,000, or 10 percent. Class average unit retail fell from 50.00 to 45.00.

Note already that the headline conceals more than it reports. Style A beat its unit plan by 200. Style B missed by 200 and Style C missed by 200, and only one of the three took a price cut. None of that is visible in minus 19 percent.

Step 1 — Volume

Volume is the effect of selling a different number of units in total, valued as though nothing else had changed. Hold the planned mix and the planned prices constant and rescale the whole plan to the actual unit total.

That is 1,800 minus 2,000, or minus 200 units, times the planned AUR of 50.00 — a volume effect of minus 10,000 dollars. Equivalently, the plan rescaled to 1,800 units at the planned mix and planned prices is 100,000 times 1,800 over 2,000, or 90,000. The running total moves from 100,000 to 90,000.

Step 2 — Mix

Mix is the effect of selling a different distribution of things at the same total. The mechanism is a constant-mix baseline: take the actual unit total and split it according to the planned shares, then compare each style's actual units against that baseline at planned prices.

Planned shares are A 50 percent, B 25 percent, C 25 percent. Applied to 1,800 actual units, the constant-mix baseline is A 900, B 450, C 450.

The mix effect is minus 6,000 dollars, and the reason is legible in the table: the class sold 300 more of a 40-dollar style than its share entitled it to, and 150 fewer of a 100-dollar style. Selling more units of a cheaper thing and fewer of an expensive one costs money even when the total unit count is held constant. Running total moves from 90,000 to 84,000.

Step 3 — Rate

Rate is what remains once volume and mix are removed: the change in achieved price per unit, style by style, on the units that actually sold. This is where markdown, promotion, and any structural price change land.

Only Style C moved on price, from 20 down to 10. That is 300 units times minus 10, a rate effect of minus 3,000 dollars. Running total moves from 84,000 to 81,000, which is actual. The bridge closes.

Read that as three separate conversations. Minus 10,000 is a demand and availability question. Minus 6,000 is an assortment question — the class was bought with too much weight on the wrong option. Minus 3,000 is a pricing question, and it is the smallest of the three, which is the opposite of where a markdown-anxious review would have started.

The three things every naive bridge gets wrong

1. A bridge is sequence-dependent, and the sequence has to be on the output

The bridge above peeled volume, then mix, then rate. Nothing forces that order. Peel them in a different order and every number changes, while the total stays fixed at minus 19,000.

Run the same month as rate first, then volume, then mix. Rate at planned units and planned mix is Σ plan units times the price change: only Style C moved, so that is 500 units times minus 10, or minus 5,000. Running total 95,000. Volume then rescales that base to 1,800 units: 95,000 times 1,800 over 2,000 is 85,500, a volume effect of minus 9,500. Mix is the remainder to actual, 81,000 minus 85,500, or minus 4,500 — which also computes directly as Σ of the constant-mix differences valued at actual prices: plus 12,000 for A, minus 15,000 for B, minus 1,500 for C.

The rate effect grows by two thirds depending on where it is peeled, and the mix effect shrinks by a quarter. Two analysts can produce two honest bridges of the same month that disagree on every line and agree on the total, and if neither states its sequence the disagreement is unresolvable. The rule that follows is mechanical: print the peel order on the output, next to the numbers, every time.

The reason the numbers move is that each effect absorbs its interaction with the effects peeled after it. In the simplest two-effect case this is visible directly. Volume measured alone against the plan base is minus 200 units times 50.00, or minus 10,000. Rate measured alone against the plan base is minus 5.00 of AUR times 2,000 units, or minus 10,000. Those sum to minus 20,000 against a real gap of minus 19,000. The extra 1,000 is the interaction term — minus 200 units times minus 5.00 of AUR — and in a chained bridge it does not appear as a line, it is silently folded into whichever effect went second.

There are two defensible treatments and one indefensible one. You can chain the effects in a stated order and accept that the later effects carry the interaction, which is what most planning teams do and what the tables above show. Or you can measure every effect independently against the plan base and report the leftover as an explicit residual line. What you cannot do is measure independently and then quietly scale the effects so they add up, which is the standard spreadsheet fix and which destroys the one property that made the bridge useful — that each line meant something specific.

2. Mix at what level

A mix effect is only a number relative to a level. Run the bridge at class level with a single row and the mix effect does not exist at all: with nothing inside the level to shift between, the entire gap resolves into volume and rate. That is not a clean result, it is a blind one.

The same logic keeps applying downward. The minus 15,000 that Style B contributed to the mix effect is a style-level statement, and it is not yet an action. Open Style B into its colors.

At style level, Style B is one failed option and the obvious response is to cut it. At style-color level it is one under-bought color and two wrong colors, and the responses are opposite: buy navy deeper next season, and drop rust and cream. Cutting Style B on the style-level read removes the one color in the class that was actually working.

That is the general rule. A mix effect is actionable only when it is computed at the level someone commits money at — the decision level of the hierarchy, which is style-color in apparel, model-color bought as a size run in footwear, and a model in a specific finish in home and furniture. Anything computed above that level names a symptom; anything computed below it is diagnostic detail with too little volume behind each cell to trust. Run the bridge at the decision level and roll it upward, rather than running it at the reporting level and hoping the detail can be reconstructed later.

3. Returns lag, and why a week-4 bridge is not a week-12 bridge

The bridge above ran on gross sales. Returns arrive on a lag, so the same month looks different depending on when you open it.

Suppose this class returns at 12 percent of gross units by the time the window closes — 216 units on 1,800 — and that 90 of them have landed four weeks after month close, with the remaining 126 arriving over the following two months. Illustrative figures again, chosen because they divide cleanly.

Nothing about the month changed between those three rows. The month got worse only because more of it finished happening. A team that runs the bridge four weeks after close and files the result has recorded a number that will not survive contact with the return curve, and a trend line built from bridges run at inconsistent lags measures the reporting calendar rather than the business.

Returns distort mix as well as total, because return rates are not uniform across an assortment. If Style A returns at 15 percent, Style B at 8 percent and Style C at 4 percent, then the 1,800 gross units become 1,020, 276 and 288 net — a net total of 1,584. Style A's share of the class falls from 66.7 percent gross to 64.4 percent net. The style that beat plan beat it by less than the gross bridge says, and a buy decision taken off the gross view buys too deep into the option with the worst keep rate. The relationship between the two bases is worked through in the gross to net sales formula.

Two rules fall out. State the basis on the output, and make the plan and the actual agree on it — a net plan compared against gross actuals flatters the business by exactly the return rate. And fix the lag at which bridges run, so that month-over-month comparisons are comparisons.

Timing is not a miss

This is the part of the analysis that separates a planner from a reporting tool, because the two situations that require the most opposite responses are the two that look most alike.

Consider the same class from the other direction. The month is weeks 5 through 8 of a twelve-week season. The weekly figures below are illustrative in the same way the bridge example was — constructed so both scenarios land on the same week-8 number. They are not benchmarks and not drawn from any brand. Through week 4 the class ran exactly on plan. At week 8 the month is 1,800 units against a plan of 2,000 — the same minus 200, the same minus 10 percent. Two entirely different things can produce that.

Scenario A — the demand arrived four weeks late. The second delivery slipped, so weeks 5 through 8 continued at the pre-build run rate of 450 units a week instead of climbing into the planned peak. The build then happened in weeks 9 through 12.

Scenario B — the demand was not there. The class ran at 90 percent of its planned weekly shape from week 5 onward, and kept doing so.

At week 8 the two scenarios are indistinguishable. Both sit at 3,600 cumulative units against a cumulative plan of 3,800, both report a month of 1,800 against 2,000, and both produce exactly the same volume effect in the bridge. Every point-in-time comparison available at week 8 returns the same answer for a slip and for a shortfall.

The test that separates them

Do not compare the point. Compare the shape of the cumulative curve against the phased plan, and watch what the gap is doing rather than how large it is.

• →Timing. The cumulative gap opens while the driver is missing, stops opening once the driver lands, and then closes. In Scenario A the gap runs plus 50, zero, minus 100, minus 200, minus 350, minus 350, minus 250, minus 100. It widens, plateaus at week 10, and reverses. Weekly actual crosses above weekly plan — 550 against 450 in week 11, 550 against 400 in week 12 — which is a thing a genuine demand shortfall does not do.
• →Demand. The cumulative gap opens continuously and at a roughly constant proportion of the cumulative plan. In Scenario B it runs minus 40, minus 90, minus 145, minus 200, minus 255, minus 305, minus 350, minus 390. It never plateaus and never reverses, and weekly actual never exceeds weekly plan in any week.

Confirm the read against a physical cause before acting on it, because the curve is evidence and not proof. A timing story should have an event behind it that someone can name and date: a purchase order that landed in week 9 instead of week 5, a launch that moved, a marketing beat that shifted, a store that opened late, a weather turn that came a month behind the prior year. A timing explanation with no identifiable cause is a hope, not a diagnosis, and it is the most comfortable thing to believe in week 8.

Note also what Scenario A costs even when it is genuinely only a slip. Season-end cumulative is 5,600 against a plan of 5,700 — the shift pushes the tail of the curve off the end of the selling window, and 100 units of it never sell at full price. A pure timing variance is rarely worth exactly zero. It is worth much less than the point variance says.

The cost of getting it wrong

The two errors are not symmetric.

Treat a shortfall as a slip and you hold a plan that is not going to happen. Receipts keep flowing against a forecast the demand no longer supports, the class carries too much inventory into the exit weeks, and the correction arrives as markdown at the worst possible point in the season. The loss is real and it is bounded by the size of the remaining buy.

Treat a slip as a shortfall and you cut the plan. The reforecast comes down, the open-to-buy is reallocated to other classes, the chase order is cancelled — and then the demand shows up in week 9 against an inventory position that was reduced to meet it. The response destroys the sales it was measuring. The class stocks out at the moment it starts working, the following season's plan is built from a suppressed actual, and the hindsight review records a style that failed when what actually failed was a delivery date. That error compounds into next year in a way the first one does not, which is the practical reason the timing test runs before the reforecast rather than after it.

The corollary for cadence is that a bridge is worth running more often than a plan is worth changing. Running the analysis weekly costs little and builds the curve you need to read; acting on it weekly converts every delivery wobble into a plan cut. Where that line sits, and how to derive it from the last date a decision can still change, is set out in how often to reforecast a merchandise plan. The phased plan the whole test depends on — and what makes a weekly shape credible enough to compare against — is covered in how to phase a sales plan.

What to do with each finding

A bridge that ends in a table of dollars has done half its job. Each effect points at a different lever, a different owner, and a different window in which the response is still possible.

The last row is a control, not a finding. A residual that is a large share of the total gap means the effects are overlapping too much to be separated at the level you are working at, which is nearly always a signal to run the bridge one level lower rather than to argue about the allocation.

Where the bridge changes by vertical

The four effects are the same everywhere. What changes is which one dominates, and what the mix effect has to be computed on before it means anything.

Apparel. The running example. Mix is computed at style-color, and the size dimension sits below it as diagnostic detail: a style-color that missed on units while its core sizes were stocked out in week 3 is a depth failure wearing a demand failure's clothes. Rate effects concentrate in the exit weeks, so a bridge run before the markdown cadence starts and one run after it are describing different halves of the season.

Footwear. The mix effect has to be computed on pairs and against the size run, not on aggregate units. A model that sold 80 percent of its planned pairs while breaking size in weeks 3 through 12 is a size-curve failure, and at model level it presents as an ordinary volume miss. Because footwear is bought as a run and sold as individual pairs, a model can be nominally in stock for the whole season and unavailable to most of the demand — the broken-run weeks have to come out of the denominator or the bridge attributes an availability failure to demand. Size-level detail hangs under the model-color line rather than replacing it; the size curve allocation formula covers the mechanics.

Accessories and bags. There is no size axis, so the entire mix effect concentrates in the variant axis — material, hardware, color, and the body itself. That makes the decision level the SKU in most ranges, and a mix effect computed at a grouping level like handbags is close to meaningless because everything that moved, moved inside it. Gifting seasonality also compresses the timing question: when a category earns a large share of its season in a handful of weeks, a one-week shift in a peak is worth more than a four-week shift in a year-round program, and the cumulative test has to be run at weekly resolution rather than monthly to see it at all.

Home and furniture. Timing is usually the dominant effect rather than a check on the others, because lead times run in months and because the sale and the delivery are separate events. The first decision is which one the plan was written against: a bridge on written orders and a bridge on delivered revenue for the same month can point in opposite directions when a container slips, and a business that plans in one and reports in the other will never reconcile them. Long replenishment horizons also mean a timing variance identified in week 8 frequently has no available response before the season ends — which raises rather than lowers the value of naming it correctly, because the finding belongs to next season's flow plan.

Sporting goods. Two structures distort the bridge. Weather-driven categories produce genuine timing variance every year, and the cumulative test has to be run against a restated calendar or a late cold snap reads as a demand collapse followed by an unexplained recovery. Model-year changeover produces the other distortion: in changeover months the outgoing and incoming models are different options in the mix, so the dominant term is a mix effect between them rather than a volume effect on the category. Read as a volume miss, the response is to discount the outgoing model harder, which spends margin to accelerate a crossover that was going to happen anyway.

Health and beauty. In a launch month the mix effect swamps everything else, because a new franchise or a new shade extension moves the distribution far more than it moves the total. Mix has to be computed at shade level within a franchise, since a shade range is where the assortment decision actually lives and a franchise-level mix effect hides the two or three shades carrying the range. Replenishment cadence also means volume effects resolve faster here than in seasonal categories — a core SKU that misses one month and recovers the next is usually an availability story, and worth checking against on-hand before it enters a bridge as demand.

The common thread is the second of the three failure modes above: the level you compute mix at is a vertical-specific decision, because the level at which someone commits money is different in each. The level names for each category are set out in merchandise hierarchy by vertical.

Who runs it, and when

The bridge is a planner's artifact. It is built weekly in-season, at the decision level, and rolled up for review — not built at review level and drilled into afterwards, because the drill-down is only available if the calculation was done at the bottom in the first place.

Three outputs come off it. In-season, the bridge is the input to the reforecast and to the open-to-buy decision: volume and mix effects tell the buyer where the chase should go and where the open buy should be cut, and the timing test tells the planner whether to change the receipt flow or the plan. At month close, it is the variance narrative — the reason the class missed, stated in four numbers with a stated sequence, a stated level and a stated sales basis. At season end, the accumulated bridges are the spine of the hindsight analysis: a season of mix effects at style-color is a direct read on which options the next line plan should carry, and a season of timing effects is a read on whether the phasing or the delivery calendar was wrong.

The three uses share one requirement. A bridge is only comparable to another bridge if the sequence, the level and the sales basis are the same, and none of the three is recoverable after the fact. Print them on the output.