What is CPA?

A campaign spends A$6,000 in a month and the platform attributes 120 conversions to it. CPA is 6,000 ÷ 120 = A$50. If those conversions are leads and one lead in five becomes a customer, the cost per customer is A$250, not A$50 — the ratio only ever describes the event it counted.

Cost per acquisition divides what a campaign spent by the number of conversions attributed to it. The arithmetic is trivial. Every difficulty in the metric lives in the word "conversion".

A conversion is whatever the account was configured to count. In one account it is a completed purchase. In another it is a form submission, a phone call over thirty seconds, an add-to-cart, an app install, a newsletter signup, or a page view on a thank-you URL. All of these produce a number called CPA, and none of those numbers can be compared with each other. When two accounts report wildly different CPAs, the first thing to check is not the bidding — it is what each of them counted.

The same applies to comparing yourself against a benchmark. WordStream's 2026 study reports an average cost per lead of US$66.69 in Google Ads, from 13,474 US search campaigns measured between April 2025 and March 2026. That figure is a cost per lead: a form or a call, not a customer. Holding a purchase-based CPA up against it tells you nothing except that purchases are harder to get than forms.

CPA looks like one metric but behaves like two, because it decomposes cleanly:

CPA = cost per click ÷ conversion rate

A campaign paying A$2.00 a click at a 4% conversion rate has a CPA of A$50. So does a campaign paying A$1.00 a click at a 2% conversion rate. Identical CPA, completely different problems, and completely different fixes.

This decomposition is the most useful thing about the metric, because it says where to look. A CPA that rose while the click price held flat is a conversion-rate problem: the landing page, the offer, the form length, the checkout, or the tracking. A CPA that rose while conversion rate held is an auction problem: more competition, a worse quality score, a shift in the traffic mix toward more expensive keywords.

Most accounts have more headroom in the conversion rate than in the click price. The auction is a market with other bidders in it; the landing page is entirely yours. Halving the cost per click and doubling the conversion rate produce exactly the same CPA, and only one of those is under your control.

CPA varies enormously across platforms, and the variation is mostly not about auction prices.

Search advertising reaches people who typed the problem into a search box. Feed advertising reaches people scrolling something else. The search lead is further along and closer to buying; the feed lead is cheaper and earlier. WordStream's figures show the shape: US$66.69 average cost per lead on Google search against US$27.66 on Facebook leads campaigns. The Facebook lead costs less than half as much and, in most businesses, closes at a materially lower rate.

Within one platform the same logic applies to network and format. Store Growers' round-up puts Google Shopping at US$38.87 per conversion, Search at US$45.27 for ecommerce, and Display at US$65.80 — the network with by far the cheapest clicks producing the most expensive conversions, because a display impression interrupts and a search query asks.

None of this makes one platform better. It makes the CPA figures non-comparable without the close rate that follows them.

The useful question is not "what is a good CPA" but "what is the most this conversion can cost before it stops being worth having". That ceiling comes out of the business, not the ad platform:

Start with contribution margin per customer — revenue less cost of goods, shipping, payment fees and anything else that scales with the order. Multiply by the rate at which the counted conversion becomes a customer. That product is the most a conversion can cost at break-even.

A business whose leads close at 20% into contracts contributing A$1,200 can pay up to A$240 per lead before losing money, which makes a US$66 cost per lead comfortable. A business whose leads close at 5% into A$400 of contribution can pay A$20, which makes the same US$66 lead a slow loss. Same benchmark, opposite verdicts.

As with break-even ROAS, this ceiling ignores fixed costs. An account running exactly at its break-even CPA is paying for its advertising and contributing nothing to the rest of the business.

It flatters campaigns that harvest demand. Branded search and retargeting convert people who had largely decided already, so their CPA is always the lowest in the account. An account judged purely on CPA will keep concentrating there, and the demand being harvested has to come from somewhere.

It ignores what the customer is worth. Two campaigns at an identical CPA, one bringing in customers who order four times a year and one bringing in single-purchase discount hunters, look the same in every ad platform ever built.

It is unstable at low volume. A campaign with six conversions in a month has a CPA that swings by double digits on one extra conversion. Below roughly thirty conversions in the window, the figure is closer to noise than signal, which is also why target-CPA bidding struggles on thin accounts.

It hides the volume trade-off. CPA almost always rises as spend rises, because the cheapest conversions get bought first. A CPA target held rigidly as budget grows quietly caps the account, and the capping shows up as "limited by budget" long after the decision that caused it.

Check the tracking first. A duplicated conversion tag halves reported CPA and a broken one doubles it, and both look exactly like a performance change. Confirm the count against whatever system actually holds the record — the CRM, the order table — before touching a campaign.

Then work the conversion rate, because it is yours. Page speed, the length of the form, the number of fields that are genuinely required, whether the offer on the page matches the offer in the ad, and whether the page answers the question the ad raised. These are usually worth more than a bidding change.

Then subtract. Search terms with spend and no conversions, placements that never convert, audiences that overlap something better, geographies you cannot serve. Exclusion raises the account's conversion rate without reducing its reach.

Then, and only then, adjust bids or targets — against your own ceiling, not against somebody else's average.

Cost per acquisition is what the ad platform charged for a conversion; customer acquisition cost is what the business paid to win a customer. Three things sit in the gap. The platform sees media spend and nothing else, so agency fees, creative production, tooling and sales salaries are all missing from its figure — in a business with a sales team that is routinely half the true cost. The denominators differ too: cost per acquisition counts whatever event the account configured, which is frequently a lead rather than a customer. And the platform counts conversions it believes it caused, while the customer table counts customers who exist. Read cost per acquisition to manage campaigns week to week, because it moves fast and points at what you can change. Read customer acquisition cost to decide whether a channel deserves to exist, because it is the only one of the two that can be held against what a customer is worth.

• Name the conversion before you compare anything. A CPA counting add-to-cart and a CPA counting purchase are different metrics wearing the same three letters, and the gap between two accounts is more often a definition than a performance difference.
• Work the conversion rate before the click price. CPA is the click price divided by the conversion rate, and landing-page work usually has more headroom than bidding does — halving the click cost and doubling the conversion rate move CPA by exactly the same amount.
• Exclude the traffic that never converts. Search terms, placements, audiences and geographies that have spent without converting raise CPA for everything else in the campaign.
• Check CPA against contribution margin per conversion, not against a published average. A US$66 cost per lead is a bargain at a A$4,000 contract value and a catastrophe at a A$90 one.