Two of my accounts produced a cost per purchase I can attach a volume and a date to: €14.18 across 1,715 orders, and $5.05 across more than 63,000 purchases. Both figures cover the full year 2022, and that date travels with them every time they appear.
Almost nothing else written on this subject carries either detail. You get a table of industry averages, no sample size, no date range, and no way to check any of it.
So this page gives you three of my own accounts instead, including one running right now in 2026 at a number I am not proud of.
Read on if you sell products online and cannot tell whether your acquisition cost is normal or terrible.
The two cost per purchase figures I can actually date
Both come from accounts I ran myself, and both are averages across a full calendar year.
The first is €14.18 across 1,715 orders, from an EU fashion ecommerce client, full year 2022. That result came from Meta and Google together, and the majority of the sales arrived through Google Search and Shopping. Anyone presenting it as a Meta number is misreading it, including me if I get lazy about saying so.
The second is $5.05 across more than 63,000 purchases, from a US electric mobility ecommerce client, also full year 2022. That account ran on Google Search, Performance Max and Display.
Look at the gap between them. One account paid close to three times what the other did in nominal terms, and both were healthy businesses at the time.
Different products, different price points, different margins. That gap is the whole reason a single industry benchmark cannot help you.
Why the volume and the date range matter more than the number
A figure drawn from 40 orders tells you almost nothing. The same figure across 1,715 orders is a fact about that year, and across 63,000 it is close to a law.
Volume is the first thing to look for on any published benchmark, and it is usually the first thing missing.
Date range is the second. Auction prices move, so a 2022 figure is not a 2026 target. Meta’s own investor reporting put its average price per ad up 12% year on year in both the first and the second quarter of 2026.
Apply four years of that to an undated benchmark and the benchmark is already wrong.
My own numbers carry exactly the same weakness, which is why I date them in the same breath. A 2022 result is a record. It is not a target for anybody buying media in 2026.
The 2026 anchor, and the ad set that ruined it
Here is a live account, so nothing rests on my memory of four years ago.
A US flag and banner ecommerce client, Meta sales campaigns, June to August 2026. Across all six sales ad sets it bought 64 orders at a cost per purchase of $67.24, on a 1.58 return on ad spend.
Remove one failing ad set and the same period reads 61 orders at $53.72, on a 1.99 return.
The spread inside those six is far wider than the average suggests. My best ad set bought orders at $29.37 with a 4.84 return. My worst active one paid $93.96 for a 0.98 return.
So the account average was accurate and close to useless. Every decision worth making lived at ad set level, and one blended figure hid all of them.
Cost per purchase against cost per lead
Different denominators, and the difference matters more than most people expect.
A lead is a promise. Purchases are money that has already moved. So a cost per lead of £11.40, which is what I held across 6,070 leads on Google Search for a UK staffing and childcare client, means nothing until you know how many of those leads closed.
Ecommerce removes that step. Cost per acquisition and cost per purchase describe the same measurement when the conversion event is the sale itself.
That is why ecommerce acquisition cost is the cleaner of the two metrics, and why I trust it further. It is also why a broken purchase event does more damage here than a broken form fill does in lead generation.
When a broken pixel invents a cost per purchase
One account in my portfolio reports a cost per purchase that is pure fiction, and its traffic side is perfectly fine.
It is a US automotive equipment ecommerce account. Delivery numbers are real and unremarkable: $8.28 CPM, $0.28 CPC, a 2.92% click-through rate, 43,415 link clicks.
Then the conversion side reports a return on ad spend of 44.37. No ecommerce account produces that. What it produces is a purchase event firing more than once, or firing on the wrong page, or being counted by two systems at the same time.
The lesson is not subtle. Before comparing your figure with anybody’s benchmark, prove that your purchase conversions are counted once each. An account with broken measurement will cheerfully show you a spectacular number while it quietly loses money.
What Meta’s own 2022 claim is actually worth
Meta has published exactly one official figure on this, and it is old.
In a Newsroom announcement dated 10 August 2022, Meta claimed Advantage+ Shopping delivered 12% lower cost per purchase conversion than the same advertisers’ business as usual ads, based on 15 A/B tests.
Read what is absent. No return figure at all. Nobody discloses how many businesses sat inside those 15 tests. And the claim has not been refreshed in four years.
If a page tells you Meta reports a specific percentage uplift in return from that product, the page invented it. The 12% figure is the only official number in existence here, and it is a vendor measuring its own product.
Reading traffic metrics that point the wrong way
The most expensive lesson in that 2026 flag account cost roughly $1,027.
One ad set targeted a neighbouring state. It produced the highest reach of anything in the account, 17,086 people, and the most link clicks, 542. On every traffic metric available it was the winner.
It also produced three purchases at roughly $341 each, on a 0.25 return. That made it the only genuine loser in the account.
The cause had nothing to do with the advert. My client manufactures locally, and the delivery advantage stops at the state line, so buyers one state over had no particular reason to choose them. The binding constraint was fulfilment, not interest.
Reach and clicks cannot see any of that. Revenue can, however, which is why cost per purchase is the only metric I let make a budget decision here.
Setting a cost per purchase target you can defend
Start from margin, never from an industry table.
Across my ecommerce clients, gross margin runs between 20% and 50%. That puts break-even return on ad spend between 2.0x and 5.0x, and it converts directly into a maximum cost per purchase.
Take a real average order value. Shopify reported $114.70 across its full merchant base for Black Friday Cyber Monday 2025, published on 2 December 2025. That is first-party data from the platform itself, which is rare in this field.
At 40% margin, that order contributes $45.88. So $45.88 becomes your break-even, and anything above it loses money on the first order. At 20% margin the ceiling falls to $22.94. Push margin to 50% and it rises to $57.35.
Those three ceilings differ by more than double on identical revenue. Acquisition cost benchmarks published without a margin attached cannot tell you anything you can act on.
What to change this week
Three things.
Check that your purchase event fires once per order, using a test order and the platform’s own event tool. A figure built on a double-counted event is worse than having no figure at all.
Then work out gross margin and multiply it by your average order value. That single number is your break-even, and most advertisers I meet have never once calculated it.
Then break the reporting down by ad set rather than by campaign. My own average of $67.24 concealed a best of $29.37 and a worst of $93.96, and only one of those three was worth acting on.
Cost per lead and cost per acquisition are not the same number covers why the lead generation version of this metric needs a close rate before it means anything, and what happens when the platform edits your ads covers a related measurement problem on the creative side. Both of my dated figures were earned mostly on search, which is where my Google Ads work applies this arithmetic. To have your own numbers checked before you trust them, book a teardown.