How to plan for returns

One identity, two plans that look nothing alike

The identity is trivial: net demand equals gross demand less returns received. What is not trivial is that different functions use different sides of it and rarely agree on which one the word “sales” means. Finance means net, because that is the revenue that stays booked. Distribution and allocation mean gross, because a unit has to physically exist in the right place to leave the building whether or not it later comes back.

A plan built on net demand and compared to gross shipments will look wrong for the entire season, and wrong in a way that is almost impossible to diagnose from inside the grid. Every week the actuals run ahead of plan, the team reads it as outperformance, and chases units it does not need. The gap is not performance; it is the return rate, showing up as a bias that grows with volume. The inverse error is quieter and costlier: a gross plan reported against net actuals looks like a persistent shortfall, and the standard response to a shortfall is margin given away against demand that was never missing.

The fix is not sophisticated. Choose the basis deliberately, state it on the plan, and make every derived measure agree with it — including sell-through, a ratio that inherits the error twice. Gross receipts in the denominator and net sales in the numerator is meaningless in both directions. Check a period on the sell-through calculator, and the definition of the rate itself sits in the return rate formula — a definition, and category bands wide enough to orient you, not a number to plan with.

Eight steps, in order

The two steps most often skipped are the third and the fifth — estimating the lag, and separating resaleable from non-resaleable units. A plan with a return rate but neither of those has the right total and the wrong shape.

1. 1

   Decide which demand the plan is denominated in

   Gross demand is what ships; net demand is what stays sold. Both are legitimate bases, but a plan can use only one at a time, because sales, receipts, sell-through and open-to-buy all change meaning with it.

2. 2

   Set the rate by channel, not blended

   Set the return rate by channel and by category. A figure blending a customer-initiated direct channel with a negotiated wholesale one describes neither and cannot be revised, because the components move for unrelated reasons.

3. 3

   Estimate the lag, not just the rate

   Express how long units take to come back as a distribution across periods, not an average. The rate says how many units return; only the lag says whether they land while there are trading weeks left to sell them in.

4. 4

   Phase the inflow against the demand that generated it

   Apply the lag distribution to each period of gross demand and sum the arrivals. The result is an inflow curve that trails demand and spills past the season end. That curve, not a haircut on the total, is what the plan is built against.

5. 5

   Split the inflow into resaleable and non-resaleable

   Only the resaleable share is inventory; the rest is a margin adjustment with no receipt behind it. Set the split by category and by how much of the selling window remains when the unit is processed.

6. 6

   Feed the resaleable share into receipts and open-to-buy

   A resaleable return is a receipt nobody ordered. Subtract it from the receipts still required and the open-to-buy balance falls by the same amount. A plan that skips this reads as open when it is already committed.

7. 7

   Rebuild the size, colour and channel mix on net demand

   Returns are not uniform across sizes, colours or channels, so the mix that comes back is never the mix that went out. Recalculate the curve and the splits on net units before the next buy.

8. 8

   Revise on matured cohorts, and re-solve rather than patch

   Judge the rate on cohorts whose return window has fully elapsed, against like-aged cohorts. When it changes, re-solve the remaining periods rather than adjusting the one that looked wrong.

Phasing the inflow across six periods

One class, one direct channel, six periods, in units. Gross demand peaks in period four. The return rate assumption is twenty percent, and units come back on a lag: half in the period after the one that shipped them, thirty percent the period after that, the last twenty percent one period later again. These inputs were chosen because they divide cleanly.

Illustrative figures, chosen because they divide cleanly. The 20% rate and the 50/30/20 lag split are inputs to the arithmetic, not benchmarks, and are not drawn from any brand or category. Returns generated by the prior season are excluded so the example stays self-contained.

Work period four. Nothing that shipped in period four has come back yet: the 304 units arriving are half of period three’s 360, plus thirty percent of period two’s 280, plus twenty percent of period one’s 200 — 180 plus 84 plus 40. Net demand is 2,000 less 304, or 1,696. The returns landing in the busiest weeks of the season were generated by demand the business was pleased with one, two and three periods earlier.

Read the two curves against each other. Gross demand runs 1,000 up to 2,000 and back to 900; the inflow runs 0, 100, 200, 304, 364, 332 — peaking a period later than demand and still heavy once demand has collapsed. That trailing shape is why a return rate cannot be applied as a haircut on the season total. A haircut removes twenty percent from every period evenly. The real inflow removes nothing from the first period and more than a third of the last.

The tail is the part most plans never see. Gross demand totals 8,500 units and will eventually generate 1,700 returns, but only 1,300 arrive before the season closes. The remaining 400 land afterwards — real units, at cost, against a season already reported. In-season net demand reads 7,200 while the true net of the season is 6,800. Neither is wrong; they answer different questions, and a plan that does not distinguish them books the difference as a surprise.

A return is either a receipt or a markdown

Not every returned unit becomes inventory again. Three conditions have to hold at once: the unit must be in saleable condition, processing must put it back into available stock, and enough of the selling window must remain to sell it at a price worth having. The first is a condition question, the second an operations question, the third a calendar question that belongs to the planner.

The third condition changes during the season, which is why the same physical unit is a different planning object depending on when it lands. Early, a resaleable return is an unplanned receipt: it adds to available stock, satisfies demand that would otherwise have needed a purchase order, and consumes open-to-buy. Late, the same unit is an unplanned markdown, because it arrives carrying more weeks of cover than there are trading weeks left.

That gives the plan a date worth naming: the last period in which a returned unit can be received, processed and still sold at full price. It is the return-side equivalent of a last-receipt date, calculated the same way — backwards from the end of the full-price window through the processing lag. Returns generated after it are a margin event with a clearance or disposal route attached, not inventory the next season inherits. So the assumption is really two: how many units come back, and what share re-enter saleable stock in time to matter.

Returns are receipts nobody ordered

Open-to-buy is what remains of the receipt plan after subtracting what is already committed. A resaleable return arrives with no purchase order behind it, so it satisfies part of the receipt requirement and shrinks the balance available to spend. Period four of the same example, shown three ways.

Period M4 of the same illustrative example, in units, with three-quarters of the inflow assumed resaleable. Like the return rate and the lag split, that share is an input to the arithmetic, not a benchmark — set it from your own disposition data. Markdowns and shrink are held out of the identity so the returns line is the only thing moving between the columns.

Ignoring returns, the period shows 200 units of open-to-buy — room to place a top-up order. Counting only the resaleable share, the same period is 28 units over-committed. The swing is the whole of the resaleable inflow, and it inverts the decision: the first version says buy, the second says stop. The middle column matters too. Crediting all 304 units reads as 104 over-committed; crediting the 228 that can actually be sold reads as 28. Both say do not buy, but they imply different exposure, and the difference between them is not inventory — it is a margin adjustment sitting in the wrong line. The identity is the one the open-to-buy calculator uses; returns simply enter it as a receipt source with no order behind it.

What comes back is not what went out

Return plans routinely assume the returned mix mirrors the shipped mix. It does not, for a structural reason: returns concentrate wherever the customer decided with incomplete information — sizes at the edges of a run where fit is least certain, colours that read differently in person, the channel where the product could not be handled before purchase.

If the next size curve is rebuilt on gross units, the plan will systematically over-buy whatever returns most. Gross shipments say a size sold well; net units say a lot of it came back. Rebuilding on net removes the bias, and it is the same arithmetic either way — only the input column changes, so the method in how to calculate size curves runs unchanged on the net column. The same holds for the colour and channel splits, and the correction never appears as a variance, because gross and net are both correct while telling different stories about what to repeat.

Allocation carries a second problem on top: a return is a relocation. Units shipped from many nodes come back to one, so recovered inventory materialises where the demand may not be, and without appearing on any inbound schedule. Two things follow. Replenishment logic has to see processed returns as available supply or the units sit while the plan orders more, and the allocation rules need a route for pushing recovered stock out to locations that can sell it. Without both, a return is not a receipt — it is a slower write-off.

Thin history, and revising without whipsaw

A new brand, channel or category has no return history, and the usual response is to reach for a published industry figure. A published band is fine for orientation — it tells you roughly which order of magnitude you are in — but it is the weakest thing to plan on, because return behaviour is driven by things an industry figure knows nothing about: the fit block, the price band, the return policy, the photography. Start instead from the closest analogue you actually own — the same category last season, the same fit block in another colourway, the same product in another channel — and state which analogue was used, so the assumption can be argued with. Where nothing close exists, plan a range rather than a point: commit the buy at the cautious end, phase receipts at the middle, and let the first matured cohort resolve it.

Measure on cohorts, never as returns in a period divided by sales in the same period. The period ratio is contaminated by growth: when demand rises, the returns arriving relate to a smaller earlier base and the ratio understates the true rate; when demand falls it overstates. A team using it concludes that returns improve whenever the business grows and deteriorate whenever it slows, which is backwards as a planning signal.

Revision discipline matters as much as measurement. A recent cohort has not finished returning, so its observed rate is always low and always rising; revising a forward assumption on it produces a cut this month and a reversal next. Two rules keep the plan stable. Compare only like-aged cohorts. And when the assumption does move, re-solve the remaining periods from it rather than adjusting the period that looked wrong — patching a single period breaks the plan’s own season total, and each further patch compounds the drift.

A blended rate hides both channels

A direct-to-consumer return and a wholesale return share a name and almost nothing else. The direct return is customer-initiated and statistically well-behaved at volume, and its timing is largely set by the published return window, so the lag distribution is knowable in advance. It is also inflated at the point of sale by bracketing, where a customer orders more than one size intending to keep one — that raises gross demand and the return inflow together, and it is why a direct channel can show strong gross growth and flat net growth at once.

A wholesale return is a negotiated or contractual event: an authorised return, a stock balancing arrangement, a damage or shortage claim, a chargeback. It is lumpy rather than statistical, arrives as a single movement rather than a distribution, and its size follows the trading relationship rather than customer behaviour. Blending the two produces a rate that describes neither and cannot be revised, because the components move for unrelated reasons. Plan each channel with its own rate, lag and resaleable share, then consolidate for the financial view. The consolidated figure is a reporting output, not a planning input. The wider version of that split — one production buy serving a booked wholesale line and a forecast direct line — is worked through in how to plan wholesale and DTC together.

Two other channels differ again. A store return re-enters saleable stock at the point it is handed over, the fastest recovery available and the one most likely to sell at full price. A marketplace or platform-managed return is the slowest to become visible: condition and disposition data often arrive later than the units, so the resaleable share is knowable after the fact rather than in advance.

Same identity, different mechanism

The gross-to-net identity holds everywhere. What changes by vertical is why units come back, when they arrive relative to the selling window, and whether they become inventory again at all. The principle: the level tracks how much of the buying decision is made after the unit ships, so where fit, scale or shade is judged on arrival the rate is structurally higher. No rates appear below, because a credible one has to come from your own history.

Some of these change the arithmetic rather than the inputs. Where a returned unit cannot re-enter saleable stock — hygiene rules in health and beauty, safety traceability across much of baby and juvenile, freight economics on large home and furniture items — the return is not inventory coming back. It is revenue reversing. There is no receipt, no open-to-buy effect and no unit to reallocate: a sales adjustment, a cost of goods that stays consumed, and often a disposal cost on top. Planning those categories with an inventory credit overstates available stock by exactly the volume that was destroyed.

Sporting goods and outdoor sit between the two patterns. Model-year transitions set the resale clock: a unit returned after the new model year lands is previous-year stock regardless of condition, so the window closes on a date the calendar sets rather than the season. Dealer prebooks add the wholesale pattern on top, with returns arriving as negotiated stock balancing ahead of the changeover.

Seven ways a returns assumption goes wrong

Why this one is hard in a spreadsheet

Returns planning needs three things joined up: the shipment history that generated the returns, the returns themselves with their disposition, and the forward plan they feed. When those three live in separate systems and only meet in a spreadsheet, the return assumption tends to become a single cell nobody revisits after it is typed.

The cost is not that the assumption is slightly wrong. It is that the lag, the resaleable share and the size and colour skew are never measured at all, because measuring them means attributing every returned unit back to the period, size, colour and channel that shipped it. When the plan, the receipts, the returns and their disposition sit on one data model, the inflow phases itself against the demand that produced it, the resaleable share lands in open-to-buy the day it is processed, and the next size curve is built on net units.

• ✓Net demand equals gross demand less returns received. Pick one basis, label every derived line with it, and make the reporting match — a net plan measured against gross actuals looks like outperformance all season.
• ✓A return rate is not a haircut. Returns arrive on a lag, so the inflow trails the demand curve and part of it lands after the season closes.
• ✓Only resaleable units are inventory. The rest is a margin event with no receipt behind it, and the same unit stops being resaleable on a date you can calculate.
• ✓Resaleable returns are receipts nobody ordered, so they consume open-to-buy. A period that reads open on gross can already be over-committed.
• ✓The mix that comes back is not the mix that went out. Rebuild the mix on net units — size curves where size drives the return, colour and channel splits everywhere.
• ✓Set the rate by channel from your own closest analogue, measure it on cohorts rather than period ratios, and revise only on cohorts whose window has elapsed.

• How to plan receipt flow — where the returns inflow lands →
• How to set open-to-buy — the budget returns quietly spend →
• How to read a WSSI — the weekly view net demand is traded against →
• How to calculate size curves — rebuild the curve on net units →
• How to plan wholesale and DTC together — two return mechanisms, one buy →
• Sell-through calculator — keep numerator and denominator on one basis →
• Open-to-buy calculator — check a single period →
• Weeks of supply calculator — cover once returns are counted →
• Return rate formula on RetailNorthstar — the definition behind the assumption, and bands that orient rather than set it →

Frequently asked questions

What is the difference between gross demand and net demand?

Gross demand is the units that ship to customers in a period. Net demand is gross demand less the units that come back, so it represents what stayed sold. The identity connecting them is simply net demand equals gross demand less returns received. The distinction matters because a receipt plan, an open-to-buy position and a sell-through rate all mean different things depending on which basis sits underneath them, and a plan that mixes the two looks wrong all season without anyone being able to say where the error is.

Should a sales plan be built on gross or net demand?

Either works, provided the choice is explicit and every downstream line uses the same basis. Net demand is the better basis for margin, cash and financial planning, because it reflects revenue that stays booked. Gross demand is the better basis for receipts and allocation, because units have to physically exist to ship whether or not they later come back. Planning both and holding the return assumption as the reconciling line between them keeps the two views comparable. What does not work is planning in one basis and reporting in the other.

How does return lag change the plan?

Return lag is the delay between a unit shipping and that unit arriving back, and it is what makes returns a phased inflow rather than a percentage haircut. Because returns arrive after the demand that generated them, the inflow curve trails the demand curve and part of it lands after the season has closed. A plan that applies the rate to the season total gets the right annual number with the wrong shape: too little inventory early, too much late, and a tail of units arriving with nothing left to sell them into.

Do returned units count against open-to-buy?

The resaleable ones do. A returned unit that re-enters saleable stock is a receipt nobody ordered, so it reduces the receipts the plan still needs and reduces the open-to-buy balance by the same amount. Non-resaleable units do not, because no inventory is created; those are a sales and margin adjustment instead. Leaving the resaleable inflow out of the calculation makes the position look more open than it is, which is how a season ends up over-bought without a single unplanned purchase order.

How do you set a return rate assumption with thin history?

Start from the closest analogue you actually own rather than an external figure: the same category in a prior season, the same fit block, the same channel, the same price band. If nothing close exists, plan a range instead of a point, commit the buy at the cautious end, phase receipts at the middle, and treat the first matured cohort as the event that resolves the range. Measure on a cohort basis from the start, attributing returns to the period that shipped the units, so the first revision has something meaningful behind it.

Why do wholesale and direct-to-consumer returns need separate assumptions?

They are different mechanisms. A direct return is customer-initiated, statistically well-behaved at volume, and its timing is largely set by the published return window. A wholesale return is a negotiated or contractual event, arriving as an authorised movement, a stock balancing or a damage claim, and it is lumpy rather than statistical. Blending the two produces a rate that sits between two unrelated behaviours, understating the volatility of the wholesale side and misstating the direct side, and it cannot be revised because the components move independently.