Returns reserve — the most-mis-modeled line in DTC finance.

01 Why the line breaks — what flat-percent gets wrong

Most DTC and CPG brands set the returns reserve as a single percentage of trailing-period sales. The percentage is anchored to a historical return rate that is itself blended across every SKU, every channel, every season, and every customer cohort the business has ever shipped. The reserve calc looks defensible because the historical rate is real — it is just real at a level of aggregation that does not match the obligation the brand is sitting on at any given quarter-end. The reserve produces a smooth number that the controller can defend to the audit committee and that nobody on the operating side pushes back on. That smoothness is exactly what breaks in diligence.

Returns are not flat. They are SKU-mix-dependent, channel-mix-dependent, season-dependent, and customer-cohort-dependent — and the four layers move at different speeds. A DTC apparel brand running a 22% blended trailing-12 return rate that ships into a Q4 holiday cohort weighted 70% toward new customers will run a true return rate against that cohort closer to 32%-36% over the forward 90-day window. The reserve at year-end is supposed to cover the forward 90 days of return obligation against the orders shipped in the trailing 90 days. If the reserve is anchored to a 22% rate against a cohort that will actually return at 33%, the under-reserve is roughly 11 points of revenue on the holiday-cohort pool. On $15M of Q4 net sales, that is a $1.65M reserve gap that surfaces as a one-time EBITDA hit in the quarter the actuals catch up — usually Q1 of the following year, exactly when the brand is opening a sale process.

The pattern repeats across every DTC and CPG sub-category we work in. Beauty brands at a 7% blended rate book the reserve at 7% and run into 14% on the new-customer post-promo cohort. Footwear brands at 19% blended book the reserve at 19% and run into 26% on the holiday cohort. The reserve was modelled at the wrong level of aggregation and the diligence team rebuilds it at the correct level.

02 The right model — SKU mix × channel × season × cohort

The reserve we now run with engagements has four layers. Each layer is independently calibrated against the brand's own data and applied multiplicatively against the outstanding-sales pool to produce a forward-90-day return-obligation estimate. The four layers are SKU mix, channel mix, season, and customer cohort — in roughly that order of materiality for most DTC and CPG brands.

Layer 1 — return rate by SKU or category group

The foundation layer is the return rate by SKU, computed off the trailing-12 by-SKU returns history, with the rate refreshed quarterly. For brands with tens of thousands of SKUs the right unit of aggregation is the category group — apparel tops, apparel bottoms, dresses, outerwear, footwear, accessories, beauty, electronics — but the principle is the same: the rate is computed against the unit the customer actually returns, not the line item the brand books revenue against. Industry priors that calibrate well as starting points where you lack history: apparel runs 20-40% returns online (Coresight has the US blended apparel-plus-footwear rate at 24.4%); footwear at 17-30%; home and furniture 15-23%; accessories and jewellery 12-15%; electronics 8-15%; beauty and personal care 4-12%. The Richpanel 2026 synthesis and the NRF / Happy Returns 2025 Retail Returns Landscape both anchor the all-ecommerce return rate at 19-20% versus 8-10% in-store — roughly 2-3x the offline channel.

Layer 2 — channel adjustment

The second layer is channel: DTC, marketplace, retail wholesale, and retailer-ecommerce all return at materially different rates, and the channel mix in the forward 90 days drives the blended rate even when the SKU mix is held constant. Working priors for apparel: DTC mid-20s%, marketplace high-20s to low-30s%, wholesale to brick-and-mortar low- to mid-teens (10-15%). For accessories and beauty the spread is tighter. The point is not the exact level — every brand needs to calibrate against its own data — the point is that the spread between channels is real and large enough to move the reserve by 3-8 points when channel mix shifts. A brand that adds a marketplace channel mid-year or expands its retail wholesale mix has to re-weight the reserve, not just refresh the blended rate.

Layer 3 — seasonal adjustment

The third layer is season. NRF / Happy Returns put holiday-cohort return rates at roughly 17% higher than the annual average across all retail; for apparel and footwear DTC the holiday lift runs 3-6 percentage points above base months in absolute terms. January-shipped cohorts return at a smaller lift over base (roughly 1.2x the base rate) driven by gift returns and post-promo behaviour. The seasonal layer is multiplicative against the SKU-and-channel layers and applies to the outstanding-sales pool weighted by ship date. The reserve at year-end on Q4 holiday cohorts is not the reserve at mid-year on Q2 base cohorts even if the brand has not changed its SKU or channel mix.

Apply the three layers — SKU × channel × season — to the outstanding-sales pool and the reserve produces a number that is defensible inside a QoE walk. The math is not complicated; the discipline is in maintaining the data infrastructure that lets the brand pull return rates at the right level of aggregation on demand. Most brands we work with can stand the model up inside a finance-and-data-engineering sprint of four to six weeks once the order and returns data is sitting in the same warehouse table.

03 The customer-cohort layer and the lag curve

Layer four is the customer-cohort overlay, and it is the marginal improvement on top of the three-layer base — not the foundation. New customers return at higher rates than repeat customers, and the cohort mix in the outstanding-sales pool is a meaningful adjustment when the brand is acquiring aggressively or shifting cohort mix. Working priors for fashion DTC: new-customer return rates run 1.2-1.5x repeat-customer rates. On DTC apparel base season, if the blended rate is 24%, new might be 28-32% and repeat might be 18-22%. For beauty and replenishment products the spread is narrower at roughly 1.1x. Brands that have the customer-cohort tagging to apply this layer cleanly can run reserves that are 10-15% sharper than brands that cannot. Most brands do not yet have the tagging infrastructure; if you do, use it.

The other half of the cohort discipline is the lag curve — the timing of when returns actually arrive against the cohort that originated the sale. For fashion DTC, roughly 70-80% of total returns arrive in the first 30 days, 90-95% by day 45-60, with a small tail through day 90. Beauty and electronics curves are tighter at 30-45 days; furniture and home goods are wider out to 60-90 days. The lag curve matters for the reserve because the reserve at any given quarter-end is the expected returns on the outstanding-sales pool minus the returns that have already been realised against that pool. The reserve formula at a layer L for an order cohort t is: (Ultimate Return Rate[L] minus Realised Return Rate[L,t]) times Net Sales for cohort t. Sum across all in-scope cohorts that are still inside the return window and that is the period-end reserve. Most brands compute this implicitly inside a single percentage; the layered model computes it explicitly and produces a number that survives a buy-side rebuild.

The inventory disposition — the sister liability

The returns reserve has a sister liability that most brands carry at roughly half the level the actual economics would imply: the inventory disposition reserve. The returns reserve is the customer-credit side — what the brand owes the customer in refunds and credits against expected returns. The inventory disposition reserve is the asset side — the expected loss on the returned physical inventory once it is back in the warehouse and routed to its final disposition path: restock, refurbish and repackage, resell as B-stock or liquidation, donate, recycle, or destroy. Both reserves are triggered by the same return event; both are distinct economic effects; both should be modelled separately and netted only at the consolidated EBITDA level.

Apparel disposition mix for a reasonably mature operation: restock 50-70%, B-stock / resale / liquidation 15-30%, destroy / write-off / recycle / donation 10-20%. Optoro's widely cited benchmark is that processing a return costs 20-39% of the item's original price all-in once you include inbound freight, inspection, grading, repackaging, and the resale or liquidation gap on the proceeds. For a $100-cost apparel SKU with a typical mid-mix — 60% restock, 20% B-stock at 35% recovery, 20% destroy at zero recovery — gross recovery is $67 on the unit, processing and freight run roughly $14, net recovery is $53, and the disposition reserve per returned unit is $47. That $47 is on top of the customer-refund reserve. Brands that book a single combined reserve at the customer-refund level are under-reserving the disposition side by exactly that amount on each unit, and the diligence team will rebuild both halves separately in the QoE.

04 What a QoE provider will test for

The QoE walk on the returns reserve is mechanical and the buy-side provider will run it the same way every time. Knowing the steps lets the seller pre-empt the rebuild. The diligence team will: (1) pull the trailing-12 returns history at the SKU-or-category and channel level from the brand's order and returns systems; (2) compute the trailing-12 return rate by SKU and by channel — the un-blended rates, not the headline; (3) build the lag curve by cohort to estimate the ultimate return rate by layer; (4) apply the rate against the outstanding-sales pool weighted by ship date and channel; (5) compare the rebuilt obligation to the booked reserve on the balance sheet at the latest period-end; (6) report the gap as the proposed adjustment.

The adjustment has two distinct flavours and the QoE will separate them. The first is the one-time catch-up — the correction of the historical under-accrual on the balance sheet as of the latest period-end. If the rebuilt reserve is $2.0M and the booked reserve is $0.5M, the $1.5M shortfall is a one-time hit. Buyers treat the catch-up as a debt-like item or a working-capital peg adjustment — it comes out of the purchase price at close, not out of the ongoing EBITDA multiple. The second is the recurring run-rate — the adjustment to ongoing EBITDA reflecting the difference between the historical assumed return rate and the steady-state rate that the rebuilt model implies. If the seller modelled an 8% return rate but the steady-state from the rebuilt model is 12%, the recurring adjustment is the 4-percentage-point delta applied across the trailing period and projected forward. The recurring adjustment is the line that compresses the EBITDA multiple, because it changes the pro-forma EBITDA the buyer is paying against.

The split matters because the two adjustments are priced differently. The one-time catch-up is dollar-for-dollar against equity value — a $1.5M catch-up is a $1.5M price reduction. The recurring run-rate adjustment is multiplicative against the EBITDA multiple — a $0.8M recurring adjustment at 10x EBITDA is an $8.0M valuation reduction. The recurring line is what hurts. Brands that arrive with a clean sell-side QoE that has already moved the reserve onto a layered model — and can show the rebuilt reserve has been running for at least two quarters with actual returns tracking the model within tolerance — clear without either adjustment. That is the goal of the sell-side reserve rebuild: not to game the diligence, but to land the reserve at the level the diligence team would land it anyway, twelve months before the data room opens.

05 What drives the size of the QoE hit

The 14% median EBITDA adjustment across our 18-QoE sample is exactly that — a median. The dispersion around the median is wide, and four operating factors explain most of the variance.

Channel concentration

Brands with DTC concentration above 60-70% of the channel mix run the largest reserve gaps. DTC return rates are structurally 2-3x in-store; long return windows (60-90+ days), free returns, "try at home" programs, and aggressive satisfaction guarantees compound. CPG brands with retail wholesale concentration and minimal DTC exposure run smaller gaps on the returns line but larger gaps on the trade-deduction line (which we cover in the trade-spend deduction recovery playbook). The channel mix is the single most consequential variable for the size of the adjustment.

Seasonality misalignment

Brands using a single annual blended rate against a highly seasonal Q4 holiday cohort run the second-largest gaps. NRF / Happy Returns put holiday-cohort return rates ~17% above annual average across all retail; for apparel and footwear DTC the seasonal premium is larger. A brand that books the year-end reserve at the annual blended rate against $15M of Q4 holiday net sales is under-reserving by roughly $0.4M-$0.9M before any other adjustment.

Growth-rate-driven lag

Hypergrowth brands run the third-largest gaps because the trailing-12 return rate is computed against a smaller historical revenue base than the outstanding-sales pool. If revenue has doubled in the trailing 12 months, the historical return rate is anchored to last year's shipments while the forward 90-day obligation is sitting against this year's shipments. The blended rate is mathematically stale even if the methodology is sound. Brands growing 50%+ year-over-year need the cohort-level reserve more than slower-growth brands — the lag effect alone can move the reserve by 4-8 percentage points.

Accounting policy maturity

Brands running on cash-basis or quasi-cash accounting through Shopify plus a basic accounting stack frequently book the reserve only as returns actually post — there is no accrual layer at all. The QoE then has to build the reserve from first principles, and the catch-up adjustment is the full obligation rather than a delta. This is where we see the >20%-of-EBITDA adjustments in our sample. The fix is not complicated; the work is shifting from cash to accrual on the returns line with a documented, auditable layered model.

06 The sell-side rebuild — what we do in the prep window

The sell-side reserve rebuild that survives buyer diligence is an 8-12 week workstream. The math is not the bottleneck; the data infrastructure is. The sequence we run with engagements:

1. 01
   Week 1-2 — data assembly: Pull order and returns data into the same warehouse table at the order-line and SKU level. Tag every order with ship date, channel, customer ID, and new-vs-repeat cohort flag. Tag every return with original-order linkage, return date, reason, and disposition outcome. Brands on Shopify plus a basic accounting stack — this is the longest part of the workstream. Brands on NetSuite — usually two days.
2. 02
   Week 3-4 — rate calibration: Compute trailing-12 return rates at SKU/category × channel × season × cohort. Build the lag curve using cumulative-return-percent-by-days-since-ship (7d, 14d, 30d, 45d, 60d, 90d). Calibrate the ultimate return rate per layer against the 60-90 day cumulative. Validate against historical realised returns in the prior period.
3. 03
   Week 5-6 — model build and back-test: Build the reserve formula in the warehouse or BI tool: Reserve = sum over cohorts of (Ultimate RR by layer minus Realised RR by layer) × Net Sales by cohort. Back-test against the last 4-6 quarters. Tolerance is typically ±10%; outside that band the model needs another layer or the lag curve needs recalibration.
4. 04
   Week 7-8 — disposition reserve and policy memo: Build the parallel inventory disposition reserve using the same cohort logic against the disposition mix (restock / B-stock / write-off) and the recovery economics. Write the accounting policy memo documenting the methodology, data sources, refresh cadence (quarterly), and audit trail. The memo is what the QoE team reads first.
5. 05
   Week 9-12 — run it live, document variance: Switch the financial system to the new methodology and run it for at least two quarters before the data room opens. The diligence team needs to see the model running cleanly with actuals tracking inside tolerance. Two clean quarters of variance documentation is the deliverable that clears the QoE walk without an adjustment.

07 Three questions for this quarter

The returns reserve is one of the rare diligence lines where the seller can fully control the outcome with disciplined operating work in the 12 months before a process. The work is not glamorous and it does not show up on the headline pitch. It shows up at the closing table, when the buyer's QoE team passes on the adjustment because the layered model is already in place and the math is already done. Three questions to run against the current reserve this quarter:

1. Is the returns reserve running on flat-percent of trailing sales, or on the layered SKU-mix × channel × season × cohort model? If still on flat-percent, the rebuild is the highest-ROI single piece of sell-side preparation work in our practice.
2. How does the trailing-90 actual return rate compare to the booked reserve as a percent of outstanding sales — is the gap meaningful at the margin? A gap of more than 200 basis points against the outstanding-sales pool is a flag that the reserve will rebuild materially in diligence.
3. Does the inventory-disposition reserve track the actual cost of restock, B-stock, and write-off — including the 20-39% per-unit processing cost that Optoro benchmarks — or is it a separate rounding-error line at half the true level?