Ecommerce performance analytics: from metrics to SKU actions

Ecommerce performance analytics is measuring what each visit, order and product contributes after the variable costs of selling it. Revenue tells you what sold. Contribution tells you what was left to pay for ads and overhead.

That gap is widest at product level: below, three products share a ROAS of 4 and only two of them cover their ads. Every number here is a dated public source or arithmetic you can redo.

What is ecommerce performance analytics?

Ecommerce performance analytics is the discipline of tying every metric to a decision: which product gets more ad priority, which offer needs fixing, which checkout step leaks orders. Revenue starts the investigation; contribution profit settles it.

Revenue on its own misleads because it has no denominator.

U.S. retail ecommerce reached $300.2 billion in the first quarter of 2025, 16.2% of all retail sales, according to the U.S. Census Bureau’s quarterly ecommerce estimates. If your sales grew more slowly than that market, you lost share while reporting growth.

I use one test for every metric in a report: can you name the product decision it should change? If not, it belongs in an appendix.

A weekly review then answers three questions per product:

Which ecommerce performance metrics should you track?

The ecommerce performance metrics worth tracking fall into three groups: traffic and conversion, basket and margin, and marketing efficiency. Before calculating any of them, check that the exports can support the formulas.

Use net revenue excl. VAT, with discounts and refunds handled the same way everywhere. Contribution profit here means revenue after product and variable order costs, before advertising.

Core ecommerce analytics metrics and their formulas

Nine ecommerce analytics metrics cover traffic, basket, margin and marketing efficiency, and each needs one fixed formula before it can drive a product action.

An abandoned cart is a cart without a completed purchase inside your chosen window. Dividing unrelated purchase events by add-to-cart events does not measure the same shoppers.

For basket context, Statista puts the 2025 global average online order value at $117.40 after discounts and returns in its online-shopping order-value series. It is a useful check on how you define order value; your own category sets the target.

Which metric should drive which decision?

Each decision needs one primary metric, read against its own threshold:

A ROAS of 4, for example €4,000 of attributed revenue on €1,000 of ads, says nothing about what remains after costs. POAS adds that cost information.

Breaking even on ads is not breaking even as a company. Fixed costs still need paying from what sits above a POAS of 1.

MER checks whether channel reports add up for the business as a whole (MER vs ROAS). Gross-profit LTV supports acquisition planning but is not a spend allowance; the LTV:CAC ratio leaves room for service costs and profit.

How do attribution models split sales credit?

Attribution models split the credit for one order across the ad interactions before it, and each model splits it differently. Say a shopper finds a jacket through a Shopping ad, comes back through a Meta retargeting ad and buys after searching for the shop’s name.

Show the dataHide the data

Who gets the €100? One order, three touches, four models

Attribution model	Shopping ad (first touch)	Meta retargeting ad (middle)	Branded search (last touch)
Last click	€0	€0	€100
First click	€100	€0	€0
Linear	€33	€33	€34
Position-based	€40	€20	€40

Last-click reporting gives that order to branded search. A merchandiser who cuts the Shopping budget because it “converts nothing” has confused where the order finished with what made it happen.

Google’s model weighs time to conversion, device type, number of ad interactions, the order of exposure and the creative type, per its data-driven attribution documentation.

The model is specific to the advertiser and the conversion event, so a purchase and a subscription renewal do not share one interpretation.

What should you do when attribution models disagree?

Disagreement between models tells you where to investigate. Compare last click with data-driven attribution per channel:

Before a large budget shift, run a holdout or geo experiment, because moving observed credit around does not show that the ads caused the sale. Incrementality testing for Google Shopping walks through both designs.

Store campaign and product IDs, timestamps, device and conversion value with each journey, and keep the clicked product apart from the purchased products.

Pick one primary model for routine comparison, keep last click as a baseline, and log disagreements instead of choosing whichever model flatters a channel.

Why does product-level analysis beat the account average?

Product-level analysis beats the account average because one ROAS can hide products that lose money on every sale.

Take three products, each with a ROAS of 4 on €1,000 of ads, and margins of 20%, 30% and 50% (illustrative data).

Show the dataHide the data

Three products with a ROAS of 4: POAS at their own margins

Product (contribution margin)	POAS
A (20%)	0.8
B (30%)	1.2
C (50%)	2

POAS = ROAS × contribution margin, as long as revenue, contribution and spend cover the same products and attribution basis. Together the three return +€1,000, which makes A’s €200 loss invisible in the campaign report.

Costs are moving faster than the ratios that report on them.

CommerceIQ’s Q3 2025 ecommerce benchmark found ordered product sales up 9% year on year and retail media spend up 22%, while unit margins fell 2–3% and CPC and ROAS stayed broadly flat. A steady ROAS can sit on top of worsening economics.

Which costs belong in the product calculation?

Every variable cost of an order belongs in the product’s contribution; fixed costs and stock do not.

The ecommerce return rate guide shows how much returns alone can move contribution.

Bundles and substitutions need care: a shopper can click one jacket and buy another, or buy it in a bundle. Keep the acquisition item and the purchased order lines separately, then calculate contribution from what was sold. Product journeys show which products bring buyers who purchase something else.

A low-return acquisition product can still be worth it when its buyers go on to generate enough contribution. That needs observed cohort data and a payback rule; without both, expected repeat purchases are a justification, not a measurement.

How do you read one product through the funnel?

Read a product’s funnel one transition at a time, because each drop-off points to a different cause.

Take an illustrative jacket with matched item IDs over one observation window: 2,000 product views, 300 carts, 120 checkout starts and 60 purchases.

Show the dataHide the data

One jacket's funnel: 2,000 product views to 60 purchases

Funnel stage	Shoppers at each stage
Product views	2,000
Add to cart	300
Checkout started	120
Purchases	60

That is a 15% view-to-cart rate, 40% of carts reaching checkout and 50% of checkouts completing. Overall, 3% of viewers buy and 80% of carts are abandoned (1 − 60 ÷ 300).

Baymard Institute’s checkout research puts average cart abandonment at 70.19%.

In its survey of U.S. shoppers, 42% had abandoned because they were just browsing or not ready to buy, and 17% because checkout was too long or complicated. Those shares describe shoppers in general, not why this jacket loses its carts.

High cart activity shows interest; it says nothing yet about whether the price or product page is right. Checkout drop-offs likewise justify an investigation before anyone concludes the payment form is broken.

Are your funnel events counting the same shoppers?

Build the funnel from view_item, add_to_cart, begin_checkout and purchase, with item IDs that match the feed, deduplicated purchases and stages joined per journey.

Comparing independent event totals produces a funnel no shopper went through.

Repeat views and cart edits inflate the stages: a shopper who changes the quantity is not a second abandoned cart. Write down whether each stage counts users, sessions, carts or events.

Judge any checkout fix by the extra contribution it brings after its own costs. A conversion lift paid for with heavier discounts can leave less money than the original funnel did.

How do you turn SKU data into break-even actions?

Turn SKU data into actions by comparing each product with its own break-even, then giving it more or less priority. Break-even ROAS is 1 ÷ contribution margin, so a 25% margin needs a ROAS of 4; work it out per product rather than applying one store-wide target.

Show the dataHide the data

Break-even ROAS by contribution margin

Contribution margin	Break-even ROAS
20%	5
25%	4
30%	3.33
40%	2.5
50%	2

The break-even ROAS calculator gives that floor for any margin. In POAS terms the same floor is always 1, which is why POAS vs ROAS puts every product on one scale.

Thirty days and 20 products are a practical starting point. Extend the window where conversion lag, returns or low volume leave the evidence thin.

Which action fits which product?

Each product’s action follows from two signals, clicks and POAS against your operating target. That target usually sits above 1, because the ads also have to leave money for overhead and profit.

These rules describe patterns; they do not prove cause.

Apply labels deliberately, keep a change log, and avoid restructuring campaigns after every short-term swing, so the next review can tell an offer change from a priority change.

For Performance Max, placement reporting is not product explainability.

Google’s April 2025 channel-reporting announcement added channel and placement visibility, but a product decision still needs campaign spend, feed items, purchased items and contribution reconciled.

A spreadsheet handles this for a few dozen products. Across hundreds, Product Segmentation does it for the whole range.

Product Metrics ML places each product in one of six segments by ad clicks and ROAS or POAS, and writes the segment to Merchant Center as a custom label for Shopping and Performance Max. It is built for shops with 500 to 50,000 products and is free up to €10,000 of Google Shopping spend over a rolling 30 days (pricing).

Final thoughts

The number that decides a product’s budget is not its ROAS but the distance between its POAS and 1. An account can meet every target and still fund products that lose money on each sale, as product A did above.

So pick the 20 products taking the most ad spend this week, put their real margins next to their ROAS, and lower priority on the ones below break-even before you scale anything else.

If you also sell through Amazon, fulfilment fees change the maths at business level as well as per product.

For that wider exercise, splitting fixed costs from contribution, break-even planning with agentcentral shows how to calculate the break-even point for Amazon FBA. Keep it separate from the narrower question of whether one SKU covers its ads.