Paid search ·Explanation

Paid search ROAS and CPA: judge cost against value

Read paid search ROAS and CPA honestly: scope conversions to real sales, set a break-even line from your margin, and find the page that leaks.

Marco’s paid search ROAS looks healthy, and his campaign report shows nine hundred conversions this month. Marlowe’s finance lead sees revenue that barely moved. Both numbers are true, and the gap between them is the whole subject here: a “conversion” that counts a newsletter sign-up next to a wool overcoat sale, a return figure built on a number that was never a sale, and a cost per acquisition read against no ceiling at all. Marco runs paid search at Marlowe, a mid-size clothing brand with its own shops and an online store. A good paid search ROAS is one above the line your own margin sets, not an industry average, and this is how to read whether a paid click actually pays back, in the order that stops the flattering number from lying to you.

Marlowe is a fictional worked example. Its figures illustrate the reading method; they are not customer results or independent benchmarks.

What does it mean for a paid click to pay back?

A paid click pays back when the sale it drives is worth more than the click cost, judged against a ceiling you set from your own margin, and counted only when the conversion is the action that means money. Everything else is traffic you are paying full price to feed. That is the answer, and paid search conversions and return only mean something once all three conditions hold at the same time. Search advertising is the single largest line in digital ad spend, just over 40% of the US total (eMarketer, 2024), so the margin between “looks profitable” and “is profitable” is the biggest number most accounts never read.

paid search return := whether a paid click earns more, in real sales value, than it cost to buy, once conversions are scoped to the action that means money and cost per acquisition is compared with contribution profit per sale; the separate click-cost ceiling also depends on conversion rate.

The measure of a search program is not how many conversions it logged. It is whether the ones that were actually sales cost less than the margin they earned. A campaign can post a proud conversion count and a return figure above target, then turn out to have counted sign-ups, and paid two dollars a click for a product whose margin only ever supported one.

The profitability chain, in one pass

Profitability is not one number, it is a chain, and it breaks at a link, not in the middle of the air. Read it in one pass, in this order:

Link The question What breaks it
Convert Did the click become an outcome at all? A strong click-through rate feeding a weak conversion rate
Scope Is the outcome a sale, or a sign-up? Counting every configured action as if it were revenue
Cost What did that outcome cost? Cost per acquisition read against nothing
Ceiling Is that cost below what the margin allows? A cost per click that looks cheap but sits over the break-even line
Pay back Did the sales value beat the spend? Return judged without a target, or on inflated conversion value
Page If it did not, where did it leak? A slow or off-message landing page nobody tested

Each link only reads true if the one before it did. Scope the conversion wrong at link two and every number downstream, cost per acquisition, return, the break-even verdict, is quietly computed on a fiction. So start there.

Scope the conversion, or profitability reads zero

In a raw Google Ads export, conversion figures come back empty until you scope them to a specific conversion action, and once you do, the action you pick decides everything. Marlowe’s account tracks three: newsletter subscriptions, account sign-ups, and completed purchases. Left unscoped, the profitability view came back blank. Scoped to subscriptions, the report showed roughly eighty-nine thousand “conversions”, a conversion rate that looked spectacular and a cost per acquisition of pennies, none of which paid for a single coat. Scoped to purchases, the honest number: $319,000 of conversion value from completed purchases. Same account, same month, two stories that share no digits.

This is the most common way a paid search account lies to the person reading it, and it is not the platform’s fault. Micro-conversions like a sign-up are real signals worth tracking, but they are not revenue, and blending them into one “conversions” total means the profitability math runs on a mix of sales and email addresses. The fix is boring and total: decide which action means money, scope to it, and read cost per acquisition and return against that action alone. Everything in the rest of this piece assumes you have.

Did the click convert, and what did it cost?

High click-through rate with a low conversion rate is the most expensive failure mode in the channel. You paid full freight for every one of those clicks, and they left without buying. WordStream‘s benchmark reports put median search click-through in the mid-single digits and cross-industry conversion around 7%, but the average is a poor judge: conversion rate is heavily category-dependent, so read it against your own vertical, not the cross-industry line. When Marlowe’s rain-shell campaign posts a click-through rate well above its category and a conversion rate well below, the diagnosis is not the ad. The ad did its job. What broke is downstream, on the page or in the offer.

Cost per acquisition is the next read, and on its own it is meaningless. Compare acquisition cost with contribution profit per sale, and click cost with that profit multiplied by conversion rate. Marlowe’s wool overcoats earn $40 of margin, and 5% of the clicks convert, so the most Marco can pay for a click and still break even is $40 times 0.05, which is $2.00. A cost per click of $1.80 on that line is a winning click; a “cheap” $2.40 click is a losing one, and the sticker price never told him which. The same logic sets the ceiling on cost per acquisition: keep it below the margin the average sale earns, and read it per campaign, because WordStream’s benchmarks show cost per acquisition swinging several-fold across industries and, inside one account, across product lines. Marlowe’s account came in near $7.90 per purchase overall, comfortably under the overcoat margin, but that blended figure hides the line that runs over it.

Paid search ROAS: did it pay back, and is the leak on the page?

Paid search ROAS is conversion value divided by spend, and it means nothing without a target and a caveat. Scoped to purchases, Marlowe’s paid search returned about $1.25 in sales value for every dollar spent last month, against Marco’s 4-to-1 target: a thin result the unscoped report had dressed up as healthy. Judge it per campaign, because the blended number is an average of winners and bleeders. And judge it honestly: last-click return credits the paid click with the whole sale even when the shopper already knew the brand, which is why Google’s own Search Ads Pause studies (2011) matter, they found most paid clicks are incremental rather than a substitute for organic, but “most” is not “all”, so a return figure is a strong signal, not a courtroom verdict.

Return on ad spend comes in two flavors, and knowing which you are reading matters. Revenue on ad spend divides conversion value by spend; profit on ad spend nets out your product margin first, so it reads truest for low-margin lines, where a healthy revenue-based return can still lose money once cost of goods is counted. Connect your commerce data alongside search in one model and the profit view opens up; on search data alone, read return as revenue-based and let the break-even ceiling catch the low-margin lines a revenue figure flatters.

When the chain says the click converts poorly and the return sits below target, the leak is almost always the page. Message match is the first suspect: the ad promised packable wool overcoats and the landing page opens on the whole outerwear catalog, so the shopper who clicked with intent lands in a browse. Speed is the second, and it is brutal: as a mobile page’s load time goes from one second to three, the probability of a bounce rises 32% (Think with Google). Checkout friction is the third. Median landing-page conversion rates cluster in the low single digits across most industries in Unbounce‘s Conversion Benchmark Report, so there is real room to move, but here is the discipline that separates paid search from guesswork: a page tweak is a hypothesis, not a fix, until an A/B test proves it moved conversion rate. Maya can tell you which page leaks. Whether a new headline plugs it is a question only a test answers, and that on-site work is read from your web analytics, covered in the web analytics sibling.

Two more reads: bids and automated targets

Two profitability-adjacent reads round out the picture, and both live in how the auction and the platform’s bidding work. You read them in your ad platform’s own interface, then bring the sale-scoped conversion-and-return verdict back to the governed model.

  • What you actually pay versus your maximum bid. In the auction you pay only enough to beat the ad below you, divided by your quality score, so a strong quality score means your real cost per click sits well under your ceiling, and the headroom is a lever. Your ad platform shows the average cost per click you actually paid, which you read against the max bid you set, and the gap between them is that headroom.
  • Automated-bidding targets. Target cost-per-acquisition and target return bidding hand the per-click decision to the platform’s algorithm, and the useful read is whether those targets are being met and where automation blurs which query earned the sale. Google Ads and Microsoft Advertising both show the target you set beside what the strategy delivered, so you can read whether the algorithm is hitting the number you asked for.

How Maya turns scattered data into answers

Most reporting stops at description: here is your conversion count, here is your return. Maya climbs the value pyramid past that, to diagnosis (which link in the chain leaks), to decision (the specific move), to outcome (did the fix work). The questions below are written to land on a decision, not a dashboard.

The answer is not a chat box bolted onto a mess. It is three layers, built in order:

  1. A marketing data warehouse unifies your raw platform exports, Google Ads next to GA4, your e-commerce orders, SEO and paid social, into one cloud database, reconciled so numbers across sources become comparable.
  2. A governed semantic layer defines every metric once, so conversion rate, cost per acquisition and return mean the same thing on every question and the model cannot invent a definition to fill a gap. Governed means the definitions are enforced, not that you are restricted.
  3. A conversational layer, a Model Context Protocol (MCP) server, lets an assistant like Claude, ChatGPT or a Slack bot query that governed layer in plain language. Your prompts steer what gets asked; the semantic layer enforces what the numbers mean. That separation is why the answers stay grounded instead of drifting into a fabricated metric.

For the profitability chain specifically, that stack does four things the ad platform will not:

  • Conversion scoping handled for you. The governed layer applies the sale-defining conversion action for you, so the numbers come back scoped to purchases without you remembering to set the filter, and the sign-ups never quietly inflate your return.
  • Bring the correct break-even line. Compare purchase CPA with contribution profit per sale, and CPC with that profit multiplied by conversion rate. Do not assume an analysis applies those thresholds unless its inputs and output establish it.
  • Return read next to your real orders. Because search sits in the same model as your Shopify orders, the return the ad platform claims reads next to the revenue that actually landed, and with your commerce margins connected it opens the profit-on-margin view that a revenue-only number can never reach on its own.
  • Clusters you define, ten levels deep. Marco does not want a flat export of every landing page his ads point at. He wants to read conversion rate and return for the whole shop, then a category, then one line, as one hierarchy.

Read the chain by landing-page cluster

That last one earns a picture. In Maya’s UI, Marco groups his landing pages into a landing-page cluster from simple text-matching rules on the URL, no code, nested up to ten levels: level one by page type, level two by product category, level three by specific line. Each level previews live as he tweaks the rule and refreshes on its own as new data loads, so nothing goes stale or needs rebuilding.

flowchart TD
  Shop["Shop pages · level 1"] --> Shirts["Shirts · level 2"]
  Shop --> Outerwear["Outerwear · level 2"]
  Shirts --> Linen["Linen · level 3"]
  Outerwear --> Wool["Wool coats · level 3"]
  Outerwear --> Rain["Rain shells · level 3"]

Now he reads the profitability chain at any altitude:

Landing-page cluster Conversion rate Cost per acquisition Return on ad spend What Marco reads
Shop › Shirts › Linen rising well under ceiling 5.1x scaling well, keep feeding it
Shop › Outerwear › Wool coats steady under ceiling 3.8x mature, holding its target
Shop › Outerwear › Rain shells slipping over ceiling 1.2x strong clicks, page not converting: test it

The rain-shell page is the leak: clicks arrive and do not buy, cost per acquisition has crossed the ceiling, and return sits far below target while linen scales. That is the diagnosis a flat export buries.

Without Maya: an analyst exports Google Ads, joins the conversion-action breakdown to GA4 and the order data, builds a pivot, re-checks why half the conversion columns came back empty, guesses at a break-even line in a side cell, and a day later you learn one product line has been paying above its margin all month. With Maya: you ask, “Which product lines are converting below their break-even cost per acquisition this month?” Maya scopes the conversion action, reads cost per acquisition against the ceiling, ranks the lines that clear it and the ones that do not, and hands you the leak with the number attached. Minutes, not a day.

The questions a marketer actually asks:

  • “Is my paid search actually profitable once I only count real sales?” The assistant scopes to the purchase action and reads return against your target.
  • “Which campaigns are paying more than a click is worth?” The assistant reads cost per click against the break-even ceiling, per campaign.
  • “Where are clicks converting worst?” The assistant reads conversion rate by landing page and points at the page to test.

The division of labor is clean: Maya does the assembly and the guardrails, unifying your sources and enforcing one definition for every metric. You supply the cost and margin context needed for profitability, check which inputs the analysis uses, and test whether a page change works.

The bottom line

Stop reading paid search on the conversion count. Read the chain: did it convert, is the conversion a sale, what did it cost, is that below the ceiling your margin sets, did it pay back, and if not, which page leaked. Scope to the action that means money before you trust a single downstream number, because the wrong scope makes every cost and return figure a fiction. Set the break-even line from your own margin and read every click cost and paid search ROAS figure against it. And treat the page as a hypothesis: Maya finds the leak, a test proves the patch.

Let the tooling scope the conversion, hold the break-even math and read return next to your real orders, so what reaches you is a verdict, not a pile of exports. A paid click shows interest, not a decision to buy. So the question is not whether your conversion count went up. It is whether the conversions you counted were sales, and whether those sales cost less than they earned.

Glossary

  • Conversion: a tracked outcome from a click. Some are sales, some are micro-conversions like a newsletter sign-up. Only the sale-defining action belongs in a profitability read.
  • Conversion action scoping: choosing which tracked action counts as a conversion. Unscoped, profitability figures come back empty; scoped to the wrong action, they mislead.
  • Conversion rate: conversions divided by clicks. Did the click convert? Heavily category-dependent; WordStream’s benchmark reports put the cross-industry search median around 7%.
  • Conversion value: the revenue attributed to conversions. Read only on the scoped sale action, or it counts value that was never money.
  • Cost per acquisition: cost divided by conversions. For purchase-scoped conversions, compare it with contribution profit per sale, read per campaign.
  • Break-even click cost: profit margin per conversion multiplied by conversion rate. The most you can pay per click and still break even. A cost per click above this line is a losing click.
  • Return on ad spend: conversion value divided by ad spend. Judge it against a per-campaign target, and read it as revenue-based unless you have netted out margin.
  • Profit on ad spend: return computed on profit margin rather than revenue. Connect your commerce cost-of-goods and it reads alongside search; on search data alone, return is revenue-based.
  • Landing-page cluster: a group of landing pages sharing one theme (a product line, a category), defined by text-matching rules on the URL in Maya’s UI and nested up to ten levels, read together rather than page by page.
  • Message match: how closely the landing page delivers what the ad promised. A mismatch converts clicked intent into a bounce.

Where this fits: this is step 7 of the paid search insight roadmap, conversions, cost per acquisition and return, the read you run once the traffic is clean and correctly scoped.



Ask this about your own account

Connect your sources, then ask Maya the question this piece just answered, on your own numbers.

Start with your store