The Defensible Profit Model: how MarginOS grades cost accuracy

The Defensible Profit Model is how MarginOS makes a margin number you can act on: every cost behind it is graded by the strength of the evidence supporting it, and that grade travels with the number. A cost taken from a supplier document is treated differently from one you typed in, which is treated differently again from one MarginOS had to model. And when a cost genuinely cannot be established, MarginOS reports it as unknown rather than quietly calling it zero — because a zero cost turns an unprofitable product into a star.

What is the Defensible Profit Model?

The Defensible Profit Model is MarginOS’s approach to cost accuracy: every cost behind a margin is graded by the class of evidence supporting it — measured, declared or modeled — and a cost that cannot be established is reported as unknown rather than assumed to be zero. It exists because a profit figure is only useful if you know how much to trust it.

Most profitability tools present a single margin number with no indication of where its costs came from. That is comfortable until you try to make a decision with it: discontinuing a SKU, raising a price, or cutting a channel is an expensive move to make on a number quietly built from a store-wide guess. MarginOS takes the opposite stance — it would rather tell you a cost is missing than show you a confident figure it cannot defend.

Before you start

• A connected Shopify store, which supplies orders, product records and any cost held against your variants.
• Optionally, a connected 3PL or shipping provider, or warehouse invoices forwarded by email — these are what turn fulfillment, inbound and storage costs from estimates into measured figures.
• Optionally, per-unit landed cost overrides for SKUs where you know the true cost better than any system does.
• Optionally, store-wide cost defaults, which act as a clearly-labelled safety net rather than a source of truth.

Where to see cost provenance in MarginOS

1. Open Catalog Profit and review the Data Trust column to see, at a glance, which SKUs have margins backed by strong evidence.
2. Click any SKU to open its detail drawer, where the landed cost is broken out into its separate layers.
3. Read the labelling on each layer to see whether that cost was measured, declared by you, or modeled.
4. Check your Margin Briefing for data-quality signals — these name the specific SKUs whose costs need better evidence, and what to supply.

The layers of a landed cost

MarginOS does not treat “product cost” as a single number. The effective landed unit cost is assembled from separate layers, each graded independently, so a strong figure in one layer is never hidden behind a weak figure in another.

This is why a MarginOS landed cost is usually higher than the supplier invoice line: it reflects the full cost of making a unit sellable, not just its purchase price.

How MarginOS grades cost accuracy

Every cost MarginOS resolves falls into one of three classes. The class is the honest answer to “how do you know that?”

Declared — you told us

A per-unit landed cost you set yourself. This outranks every inferred source, deliberately: an operator who knows their true cost should outrank an inference drawn from a system. Setting an override replaces the base product cost only — freight, inbound handling and storage remain separate layers on top, so an override never silently erases the rest of your cost stack.

Measured — a document or record says so

A cost established from evidence: supplier, freight and 3PL documents, the cost held against the product in your store, or the cost carried on the order line itself. Where several sources describe the same cost, MarginOS prefers the more primary one — a warehouse or supplier document outranks a platform-reported figure — and a weaker source is never allowed to overwrite a stronger one that is already established. Measured costs are the only class MarginOS keeps as the SKU’s cost of record.

Modeled — a labelled assumption

Where evidence is genuinely absent, MarginOS may model a cost so a SKU is not left entirely unpriced — for example, estimating inbound handling when no inbound document exists, or applying a store-wide average cost assumption if you have configured one. Three rules make this safe:

• It is always labelled. A modeled cost is never presented as a measured one, and the SKU carries a data-quality signal naming the assumption in play.
• It never becomes the record. A store-wide assumption is applied for display and calculation only; it is not kept as the SKU’s cost of record, so it cannot harden into apparent fact or contaminate later figures.
• It yields immediately. The moment any real evidence arrives for that SKU, the evidence takes over. Clear the store-wide default and the SKU reverts to reporting a missing cost.

What happens when a cost is genuinely unknown

If no class produces a cost, MarginOS reports the cost as missing and raises a signal against the SKU. It does not fall back to zero. This is the single most important rule in the model, because a zero cost does not look like an error — it looks like a fantastically profitable product, and it will quietly distort every roll-up that SKU belongs to.

The same principle governs costs that arrive but cannot be matched to a SKU. Rather than discarding them or spreading them somewhere convenient, MarginOS quarantines them with a reason, so the money stays visible and accountable instead of vanishing from your P&L. A cost MarginOS cannot place is a question for you to answer, not a number for it to invent.

Where MarginOS is deliberately approximate — and says so

Some costs are billed to your brand as a whole rather than per SKU; warehouse storage is the clearest example. MarginOS allocates these across SKUs by their share of the relevant driver, and is explicit about what that means: the brand-level total is exact — every dollar billed is accounted for — while the per-SKU split is a reasoned allocation, not a measurement. Allocations are constrained so the parts always sum back to the billed total, which is what stops apportioned costs from being duplicated or lost along the way.

One consequence is intentional: slow-moving stock carries its share of storage cost. MarginOS does not flatter dead inventory by excluding what it costs you to keep holding it.

Example

A cold brew concentrate sells for $28.00. MarginOS assembles its landed cost from four layers:

MarginOS shows this SKU as costed, with the one modeled layer called out — so you know the $8.87 is well supported, and you know precisely which document would make it airtight. Contrast a second SKU with no supplier document, no store cost and no override: if you have set a store-wide assumption it is applied and clearly labelled as an assumption; if you have not, the SKU reports a missing cost and appears in your briefing, rather than showing a $28.00 profit that would be entirely fictional.

FAQ

Where does MarginOS get my product costs?

From the strongest evidence available for each SKU: a per-unit cost you set yourself, costs read from supplier, freight and 3PL documents, the product cost held in your store, or the cost carried on the order line. MarginOS prefers primary documents over platform-reported figures.

What happens if a SKU has no cost data at all?

MarginOS reports the cost as missing and raises a data-quality signal against that SKU. It does not treat the missing cost as zero, because a zero cost would silently show the product as far more profitable than it is.

Does MarginOS ever estimate my costs?

Yes, in two clearly labelled cases: an inbound handling estimate when no inbound document exists, and a store-wide average cost assumption if you have set one and no evidence exists for that SKU. Both are marked as modeled, never presented as measured, and are replaced the moment real evidence arrives.

Why is my landed cost higher than my supplier invoice?

Landed cost is the full cost of getting a unit sellable, not just the product price. MarginOS adds inbound freight, inbound handling and storage as separate layers on top of the base product cost, so the figure is deliberately higher than the supplier line alone.

Can I override a cost that MarginOS calculated?

Yes. A per-unit cost you set takes precedence over every inferred source, because an operator who knows their true landed cost should outrank an inference. Freight, inbound handling and storage remain separate layers on top of your override.

How do I know whether a margin is based on measured or estimated costs?

Open the SKU’s detail drawer to see the cost stack broken out by layer, with the class of evidence behind each one. SKUs relying on assumptions carry a data-quality signal telling you exactly what to supply to upgrade them.

Related

• How MarginOS calculates Gross Profit (CM1)
• Data Trust levels: how MarginOS grades margin reliability
• Why a cost landed in Unmapped Costs
• Set your store-wide COGS, shipping and pick-pack fallbacks