Retail marketing runs on Cost Per Click. The business runs on retained margin per cart. The space between those two numbers is where DTC brands burn through working capital, big-box buyers misallocate retail media spend, and CPGs lose the trade-vs-media argument they keep promising the CFO they will win next year.
Open the dashboard at any retail or DTC marketing team — Allbirds, Target, P&G, the brand on the next floor — and the spine of the reporting is roughly the same. Impressions, clicks, click-through rate, conversion rate, cost per click, cost per acquisition, ROAS. Maybe a lifetime-value column tucked off to the side that nobody fully trusts. The whole structure is inherited from a decade of Google Shopping, paid search, and the click as the unit of accountability.
That structure made sense when the funnel was visible. The shopper ran a query, scanned ten organic results, clicked one, landed on a product page, added to cart, checked out. Every step was countable. Every click had a cost. Optimizing toward a cheaper click was, for a long time, the same thing as optimizing toward a more efficient business.
That funnel barely exists anymore. AI Overviews now answer the query before the click happens. Performance Max bids on signals the marketer never sees. Demand Gen places product images in YouTube and Discover before the shopper has formed an intent. ChatGPT and Gemini shopping experiences are pulling consideration into a surface where there is no SERP, no organic result, and no click in the traditional sense. The agent is the shopper now, and the agent does not click around comparing prices the way the persona doc imagines.
The KPI hierarchy did not get updated. Most retail teams are still optimizing the metric at the top of an inherited stack while the platform underneath has rewritten itself. That is the gap. The Retail Decision Stack closes it.
What Cost Per Click is actually measuring in 2026
The marketer pays for a click. Performance Max routes it. Sometimes the click is from a high-intent shopper at the bottom of a cart funnel; sometimes it is from a YouTube viewer Google decided to push a product image at because the algorithm modeled them as a likely converter; sometimes it is from a Discover feed surface the marketer cannot even isolate in reporting. Three different shoppers, three different downstream behaviors, one shared CPC line in the dashboard.
The platform does not care about that CPC. Google's AI is bidding on its own internal estimate of expected value, calculated from signals the marketer mostly does not provide and partly cannot see — the auction-time bidding behavior documented in Smart Bidding. If the marketer feeds it a flat conversion event — purchase — the platform optimizes for cheapest path to any purchase. If the marketer feeds it a flat ROAS target, the platform optimizes for revenue without distinguishing between a one-time cart and a repeat customer. The platform does what it is told. It is the input that is wrong.
Last-click attribution had this same shape. The metric was cheap, defensible, and quietly destructive. Meridian MMM exists because retail finally got tired of the Search-attribution illusion. Cost Per Click is the next layer of the same illusion: the metric is fine for an in-channel diagnostic, but it is the wrong basis for budget allocation in an AI-mediated retail funnel.
The Retail Decision Stack
The Retail Decision Stack is a five-layer KPI hierarchy. Each layer corrects a measurement failure in the layer above it. The whole stack runs in the units the business actually earns in — gross margin, retention, customer lifetime margin — rather than the units the platform produces in its default reports.
Read the right column straight down and the stack is doing one thing: pricing every layer of the funnel in the same units the P&L is priced in. Margin, mix, retention, lifetime. Cost Per Cart is the load-bearing metric in the middle — it is the layer that connects the click-level reporting the platforms produce to the business-level economics the CFO actually cares about.
"Cost Per Click was a measurement metric. Cost Per Cart is a budget allocation metric. The two are not the same metric, and they should never have been doing the same job."
How this looks across DTC, big-box, and CPG
DTC and digitally-native brands
The DTC playbook of the last decade was: drive cheap clicks through Meta and Google Shopping, capture email, retarget into a first purchase, hope retention covers the gap. That model broke in 2022 and has been quietly insolvent ever since. Cost Per Cart is the metric that exposes which acquisition cohorts actually pay back the working capital that funded them. Brands running PMax with a flat purchase-conversion signal are spending the same dollar to acquire a one-time discount shopper and a high-AOV repeat customer. The Retail Decision Stack reweights the conversion value at the cart level so Google's AI bids toward the customer the brand actually wants. Within four weeks of the switch, the audience mix shifts and the working-capital math starts to recover.
Big-box and omnichannel retail
Big-box retail has the additional problem of running a retail media network and a brand media operation as if they were two different businesses, measured in two different units. The brand side reports on ROAS and CPC. The RMN side reports on closed-loop ROAS to the suppliers buying the inventory. Neither side is measuring in retained margin. The Retail Decision Stack restates both in the same units, which is the only way to settle the perennial argument about whether RMN spend is incremental or just intercepting demand the store already had. Once both sides report in Cost Per Cart and CLM, the cross-functional argument becomes a math problem rather than a political one.
CPG brands
CPG marketing leaders spend half their year defending media budgets against trade promotion budgets, with the trade side carrying a measurement advantage because trade outcomes are visible at the SKU level in the retailer's data. Media outcomes have been stuck at the campaign level. Meridian MMM and the Retail Decision Stack together change that — Meridian does the cross-channel incrementality work and the Decision Stack does the cart-level economic work, so the trade-vs-media conversation moves into the same units. The CPG brands that build this measurement layer first will win the next three planning cycles against the brands still running the argument on ROAS.
Feeding the stack into Performance Max and retail media
The technical implementation is more straightforward than it sounds. Performance Max already supports value-based bidding through dynamic conversion values. The piece most teams are missing is a model that produces a defensible per-cart value at the moment the conversion fires.
The shape of the model: take cart contents at conversion, calculate gross-margin contribution from product-level margin data already sitting in the feed, weight by an expected-retention coefficient calculated from acquisition source and customer cohort, and pass the resulting value to PMax as the conversion value. Google's AI will bid toward higher predicted retained margin within two to four weeks of the data being fed back through. The product feed becomes a strategic asset rather than a data file, because the feed-level margin and retention attributes flow directly into the bidding signal.
Retail media networks are a parallel build. Walmart Connect, Amazon Ads, Target Roundel, Kroger Precision, and Instacart Ads each have their own conversion-value mechanics, but the principle is the same: send them retained margin as the optimization signal, not gross sales. Brands that do this on RMN before their competitors do it will see the same audience-mix shift on those platforms that they see on PMax, and the cross-channel comparison will start producing decisions instead of arguments.
What the closer-loop on the P&L looks like
The financial argument for the Retail Decision Stack is the same shape as every argument on this site: the marketing decision and the business decision become the same decision.
A brand running on CPC and ROAS allocates budget to whatever produces the cheapest reported conversions. The cohort comes in. Some fraction of the cohort is high-margin and high-retention; some fraction is low-margin and one-and-done. The blended numbers look fine. The unit economics, calculated honestly six months later, are weaker than the marketing report suggested. The brand adjusts inventory, raises price, blames the agency, or all three. Nobody is wrong. The business loses working capital anyway.
A brand running on the Retail Decision Stack allocates budget against retained margin at the point of the bid. The cohort comes in already weighted toward the right shopper. The retroactive reconciliation still happens, but the variance is information rather than damage — feedback that improves next quarter's retention model rather than evidence that this quarter's marketing was misallocated. The same marketing dollars produce more retained margin, more working-capital velocity, and a higher LTV/CAC ratio. Cost Per Decision drops because the marketing system is bidding on the right shopper the first time, instead of bidding on the cheapest shopper and correcting later. Health insurance has the same shape of metric problem in its own vertical — the lesson generalizes.
The Retail Decision Stack is one vertical application of a broader argument — every marketing KPI in common use measures activity, not decisions, and agentic AI is the first technology that makes the correction operable rather than aspirational. Retail is just where the click-shaped distortion is loudest.
The marketing team is not optimizing wrong. The metric stack is. Cost Per Click was the best a retail brand could do when the funnel was visible and the click was the unit of accountability. The funnel is no longer visible and the click is no longer the unit. The retailers that rebuild their KPI stack first will spend the next two cycles compounding a structural cost advantage the rest of the market is still arguing about in ROAS.
Sources & further reading
- About Performance Max campaigns — Google Ads Help. How PMax synthesizes signals across Search, YouTube, Demand Gen, Discover, and Display in a single campaign.
- About Smart Bidding — Google Ads Help. The auction-time bidding system the platform uses to value each impression.
- Value-based bidding best practices — Google Ads Help. The technical primitives behind feeding the platform a per-conversion value instead of a flat goal.
- About conversion value rules — Google Ads Help. How conversion values flow into Smart Bidding optimization at the campaign level.
- About Demand Gen campaigns — Google Ads Help. How Demand Gen surfaces product images on YouTube, Discover, and Gmail before query intent forms.
- About data-driven attribution — Google Ads Help. The model that replaces last-click and uses account-level data to credit conversions.
- Embrace customer lifetime value for growth — Think with Google. The case for moving retail measurement from cart-level metrics to retained lifetime margin.