Period P&L: the margin cascade

Run by scripts/pnl.py. This is the period-level companion to the per-unit model in margin.py, and it is a genuinely different calculation — not the same model summed up.

Net margin = Net Operating Profit ÷ Net Revenue. Precisely, this is net operating margin: before financing costs and income tax. It is not the after-tax net profit margin an accountant reports, and should not be compared to one. Always state the numerator and the denominator.

Neither number appears in Seller Central. Amazon shows gross sales that overstate revenue and has no idea what the products or the team cost. The waterfall has to be built.

The eight steps and four checkpoints

Gross Ordered Revenue            everything ordered, at the price ordered
(-) Cancelled Orders
  = Gross Shipped Revenue            what actually left the warehouse
(-) Refunds
(-) A-to-Z Claims
(-) Chargebacks
  = Net Product Revenue              the product money actually kept
(+) Shipping Revenue
(+) Gift Wrap Revenue
(-) Shipping Refunds
  = NET REVENUE                      ← checkpoint. The denominator for everything below
(-) COGS (landed)
  = PRODUCT MARGIN                   ← what the product earns before any selling cost
(-) Amazon Fees
  = CHANNEL MARGIN                   ← what survives the channel
(-) Advertising
(+) Reimbursements
  = GROWTH MARGIN                    ← what survives paying for growth
(-) Allocated Overheads
  = NET OPERATING PROFIT

The final number says how much was made. The checkpoints say why. Report the checkpoints, not just the bottom line — the gap between Product Margin and Channel Margin is the cost of selling on Amazon, in points, and it is worth tracking period over period because fee creep is real and silent.

Express each cost as points of net revenue. That is what makes one period comparable to another, and pnl.py emits it as points_of_net_revenue. Supply a prior block and it also reports each checkpoint's movement as deltas_in_points — in points, so −2.9 means the checkpoint gave up 2.9 points of net revenue. That is the fastest way to answer "why did margin drop."

A rough read on the result: operators commonly treat 15%+ as healthy, 20%+ as strong, and under 10% as fragile, since one fee increase erases it. These are rules of thumb, not laws. The more useful habit is watching your own checkpoints move.

Date basis — decide this first

One decision shapes every number: which date a unit of currency belongs to. Three defensible answers, each right for a different job:

All three are needed for different jobs. Mixing them in one calculation is how spreadsheets quietly break. pnl.py warns when the basis is not stated.

Related trap at Step 2: a matured order cohort, where every order has either shipped or cancelled, reconciles exactly. A current-period shipment report contains orders placed earlier and is missing orders placed now that have not shipped. State which view is in use and where the cutoff sits.

VAT stays outside

VAT is money collected on behalf of a tax authority. It is never revenue and never cost. Put it inside the P&L and it either inflates revenue (VAT-inclusive prices) or invents a phantom expense line — both distort every margin below.

Keep it as a memo, outside the waterfall, clearly marked as not affecting profit, because sellers still need to know the cash amount involved. Pass it as vat_memo.

Same logic on the cost side: recoverable input VAT is not a cost. Where VAT is not recoverable — because of registration status or the nature of the expense — it does form part of that cost, and belongs in the line it attaches to, not in a VAT line.

This is consistent with the per-unit model, which divides the price by (1 + VAT rate) to reach net revenue. Both keep VAT out of the profit math.

Failure patterns to check for

These five inflate margins constantly. Check for each before trusting a result:

1. Starting from the wrong revenue. Seller Central's headline "sales" is gross ordered revenue, including orders that will be cancelled and refunded. A margin on that base is inflated before a single cost is counted.

2. Forgetting the refund family. Refunds are the visible leak. A-to-Z claims and chargebacks also claw revenue back and live in different reports, so spreadsheets routinely omit them.

3. Ignoring reimbursements. Amazon owes money for inventory it loses and damages. This is a positive line. Omitting it understates profit and removes the incentive to claim it.

4. Mixing VAT in. See above.

5. Stopping at fees. Revenue − COGS − Amazon fees is Channel Margin, not net margin. Advertising and overheads still have to come out. Calling channel margin "net margin" is the most common overstatement in the category.

Two more worth watching:

Double counting in the refund family. Match each claw-back to its order. An A-to-Z claim that Amazon settles by refunding the order appears once — as a refund or as a claim, not both. A later reversal reverses the original line rather than adding a new one.

Prep cost booked twice. Prep belongs in landed COGS per unit or in overheads as a period cost. Never both.

ACoS vs TACoS

ACoS is ad spend ÷ ad-attributed sales. TACoS is ad spend ÷ total revenue.

For margin purposes TACoS is the honest number, because the P&L does not care which sales the ads claim credit for. The default denominator here is net revenue; some tools use gross sales, which produces a smaller and flattering percentage. pnl.py reports which denominator it used — state it whenever comparing against a benchmark.

This is the same convention the per-unit model uses when it divides ad spend by total units rather than ad-attributed units.

Where each number comes from

The reason so few sellers have a true net margin: the inputs live in six places.

See references/sourcing.md for the specific calls.

Cadence

Weekly for the trend, monthly for the full picture with overheads allocated, quarterly for pricing and product-line decisions. Investigate any checkpoint that moves more than a couple of points.

Validation

pnl.py reproduces the published Clarisix worked quarter exactly — Net Revenue €27,224.4, Product Margin 66.8%, Channel Margin 36.4%, Growth Margin 29.2%, Net Operating Margin 24.2%, TACoS 8.0%. It ships as scripts/fixtures/period-worked-quarter.json. Re-run it after any change:

python scripts/validate_fixtures.py