Meta ads: why the cost per sale has gone up, and what to do next

Julie's messages of 15 and 16 October, which this report answers, alongside the wider questions about the budget, the audiences and the broad campaign.

Meta is charging less to show the ads than it did two weeks ago. The rise comes after that: somewhat fewer people click through, and fewer of those who do go on to buy. Three things line up with it. The audiences running out is not one of them: the odour-conscious group has reached about 1% of the people in it (section 7).

Checked on the live website at noon Pacific on Friday 16 October, and still the same at 12:44 AM on Saturday. Shopify stops selling a product when its stock count reaches zero, and these two reached zero with the last orders that took them. The bottles Julie picked up this week are not in the counts yet.

Where the running ads send people:

Ads Manager shows an average over the date range chosen, so over a long range a change shows up slowly and late. Split by week, it is clear where the rise happened. Top figure: Meta's cost per purchase. Below it: cost per order using Shopify's count, which only credits an ad when it was the last thing the buyer clicked.

Weeks run from launch day, Thursday 24 September; the last column is a day and a half. The retargeting and interest campaigns started on 4 and 5 October. Meta counts a purchase if the buyer clicked through from an ad in the previous seven days or saw one the day before, the same rule for all three campaigns. Shopify's count is stricter: it misses people who saw an ad and came back another way. The truth sits between the two, so both are shown. More of the broad campaign's purchases come from people who saw an ad without clicking (9 of 58, against 1 of 13 and 1 of 12), one reason its Meta count runs further ahead of Shopify's. * In these two columns a large share of the broad campaign's spend (55% and 35%) went to ads without tracking tags, so Shopify could not credit their sales and its figure reads too high; Meta's count is the better guide there.

Cost per sale is three things multiplied together: what Meta charges to show the ad, how many people click, and how many of those buy. From the first full week of October to the second, the broad campaign's cost of showing the ads went down, clicks per view fell by about a fifth, and orders per 100 clicks fell by about a third. Most of the rise is in that last step, after the click.

The whole store shows the same pattern. Visits went up by over 40% from the first full week to the second, and the share of visits ending in an order went down.

• The first full week of October was the strongest so far. Monday 5 and Tuesday 6 October alone brought 24 orders, so the comparison with it looks steeper than the trend.
• The broad campaign changed its main video, and the page it linked to, on 9 October (section 6).
• The broad campaign has rarely had a settled week. Meta treats a new ad or a changed video as a reason to restart its learning, and a large budget change can do the same: a period that "can lead to less stable performance and a higher cost per result", in Meta's words. It settles after about 50 purchases in a week without such a change; the broad campaign makes about 22 to 26 a week by Meta's count. Since 9 October a new ad or a changed video went in on three days and the budget moved by a third or more on four (section 13).
• The last day and a half is the least reliable. About a third of the broad budget went to the Pants image ad, which drew fewer clicks (2.0% against about 4%) and has no tracking tags, so Shopify could not credit its sales. Without it, the broad campaign's clicks and orders per click are close to the week before. The stock-out falls in the same days (section 3).

Why these three: on 9 October the two ads carrying most of the broad campaign's budget changed video, from Skin Won't Heal to Coconut oil in one and to Frank's RedHot in the other. Same campaign and audience. The page each one links to changed at the same time: Skin Won't Heal went to the 11 reasons guide page, Coconut oil goes to the homepage and Frank's RedHot to the two-bottle bundle page. The last row is Skin Won't Heal running in the interest and retargeting campaigns over the same days as the two newer videos.

Skin Won't Heal★ Sells best

"What actually works on yeast"

Names the problem the buyer has

Frank's RedHot

"Stank is like Frank's RedHot for the skin"

One bottle, many uses

Coconut oil

"Coconut oil is a monster"

About an ingredient

The page matters as well as the video. Since 1 October, 5.4% of visits to the guide page added to the cart, 6.5% on the homepage and 2.7% on the two-bottle bundle page. So part of Frank's RedHot's gap is the page it links to. Coconut oil links to the homepage, where shoppers add to cart as readily as on the guide page, and still brings far fewer, which points to the video. Both are reasons to send Skin Won't Heal back to the guide page.

"Watched past 3 seconds" is Meta's 3-second video views per impression, read for 1 to 8 October (Skin Won't Heal in broad) and 10 to 16 October (the newer videos). Spend and sales are split at the hour each video was changed. "Added to cart per $100" is Meta's add-to-cart count for every $100 spent. Skin Won't Heal's time in broad included the strongest days so far (5 and 6 October), which flatters it a little; its results with new people in the odour-conscious group since 9 October show it was not only the timing. Page figures are all visits from every source.

No. An audience wears out when most of the people in it have seen the ads several times; the cost to show the ads then climbs. Here the interest groups have reached a small fraction of their people, the cost to show the ads has gone down, and each person reached has seen an ad about once.

• Odour-conscious is the keeper. It has run Skin Won't Heal throughout and holds at about $27.67 per order on Shopify's count ($22.14 on Meta's).
• Clean beauty has done its job as a test. Since 9 October it has run a reworded version of Skin Won't Heal and the Boob image, with one order in the last seven days, and it accounts for the interest campaign's higher cost in the week of 8 October. Its ads were switched off on Friday morning, so it no longer spends. Its slot is free for a new group.
• Retargeting has the lowest cost per sale, as expected, because it reaches only people who already know STANK. It spends its full budget, about $20 a day this week, and the people it reaches see an ad 1.71 times a week. The warm audience grows as the other campaigns bring new visitors, so there is room for a small step up.

Meta no longer lets anyone target health conditions: eczema, dermatitis, psoriasis, body odour and fungal infections are not in its list. The way around it is to target what people with those problems buy. Each group below was checked in Meta's own list this morning, with its size in Canada.

• Same video in each, so the audience is the only difference. Skin Won't Heal, $10 a day each, judged on cost per order after about $60 to $80 each. A group that clearly beats the broad campaign gets the budget; one that does not is switched off.
• Interests now act as a starting point, not a fence. For campaigns that optimise for purchases, Meta applies its "Advantage+ detailed targeting" automatically, "with no option to opt out", and shows the ads beyond the chosen interests when it expects better results. So the group mostly decides who Meta tries first, and the video decides who buys.
• Next after these: a lookalike of the buyers. Meta builds an audience of people similar to existing customers. It needs a source of "at least 100 people", and works best from 1,000 or more. STANK has had 91 different customers since launch; Meta can only use the ones it can match, so this is likely a week or two away.

The broad campaign is working as broad campaigns should: Meta has found the buyers on its own. 59% of its spend goes to people aged 45 to 64 and 77% to women, and Quebec takes about the same share of the spend (8%) as it gives of the orders (7%). So the targeting itself needs no change. These are the changes that would help.

• Skin Won't Heal back in, as a new ad beside the others, linking to the 11 reasons guide page. In a broad campaign the video does much of the targeting: it decides who stops, who clicks and who buys. Keep Frank's RedHot, Coconut oil and Dysbiosis running alongside, and Meta will move the budget to whichever sells.
• Make these changes together. Skin Won't Heal, leaving out buyers and the budget each restart or can restart Meta's learning. Done in one go, it relearns once; spread over three days, three times. Then leave it for at least three days.
• New videos go into a small test campaign first. A broad audience at about $20 a day, the same settings as the main campaign. Videos that bring sales there join the main campaign together, once a week. Adding a new ad restarts the main campaign's learning, so one planned day a week costs far less than a new ad most days. This is how many agencies run it, and Meta's own advice is to give it a variety of ads to choose from.
• More videos in the Skin Won't Heal vein. Julie on camera, opening on one problem in a large headline. Two or three new ones a week keeps the test campaign fed.
• Leave out people who have already bought. A bottle lasts a while, so the next few weeks of ads are better spent on new people. Retargeting already leaves buyers out; the broad campaign does not yet. Meta keeps this exclusion even with its automatic audience on. In one published test, a broad campaign spent a quarter of its budget on people who already knew the brand.
• Send people to pages that can sell. The guide page and the homepage turn the most visits into carts; the guide page's button and the new reel's page should go to a product that is in stock.
• Every ad carries the tracking tags. The Pants image went out without the tracking line in its website settings, so Shopify could not see its sales. A new ad copied from an existing one keeps the line; one built from scratch needs it added.
• Change the budget in steps, not jumps (section 10).

The broad campaign went from $100 to $175 a day at 4:54 AM Pacific on Thursday and to $250 at 7:00 AM on Friday, two and a half times in about a day. Raising the budget makes sense with stock to move. The timing works against it: the two best sellers cannot be bought, and the video that sold best is not in the campaign, so the extra money is buying clicks that are less likely to turn into orders.

• Why steps. Meta's guidance is to "avoid frequent budget changes": a large one can restart the learning, while a small one is unlikely to. Raising by about 20% every two to three days is the common practice among agencies for growing a campaign without that restart. The move to $175 is a large change too, which is why it goes in with the other broad changes, once.
• The number to steer by. Since 5 October the average order has been $53.62 before shipping and tax. On the roughly 30% margin Julie gave on 2 October, a first order covers about $16 of advertising. The broad campaign is above that at the moment ($42.63 per purchase over the last seven days on Meta's count), so for now it is a bet on reorders and on the new shipper at half the price. The simplest single number is the whole store's cost per order in section 4, all ad spend divided by all orders: $20.64 in the first full week of October and $29.62 the week after. Stepping up while it stays under about $25 keeps growth and cost in balance.
• Google Ads is being set up and starts at the lower end, with results checked before the budget goes up.

Dysbiosis

Julie to camera, 45 seconds, in the broad campaign. Live since 7:12 AM Pacific on Friday.

Spent so far

$16.30

Clicked to the site

4.55% · a strong start, on 418 views

Watched past 3 seconds

32.6%

Orders

none yet · too early to judge

It is the right kind of video: Julie explaining why creams have not worked, which is the Skin Won't Heal formula. People are clicking it. Four things would give it a fairer run:

• Change where it links. Its page has had 37 visits since 1 October and no add-to-carts, and its one buy button goes to the two-bottle bundle. The 11 reasons guide page, where most ad buyers land, is the better home once the stock is updated.
• A headline on screen in the first second. Skin Won't Heal opens on "What actually works on yeast" in large type. This one opens on subtitles. Its caption's own first line would do it: "The reason your creams aren't working?"
• One line of the caption. Near the end it says "Tap the link in our bio"; in an ad the Shop now button does that job, so "Tap Shop now" fits better.
• Give it three to four days and about $50 to $75 before judging it. Meta needs that long to find its buyers, and switching it off and on again would restart the campaign's learning.

It is "reel campaign - Copy", beside "reel campaign" in the campaign list. Meta's log shows both were made on 25 September, before the retargeting and interest campaigns started on 4 and 5 October, and that the copy was made 28 minutes after the original from Julie's login. Neither has spent anything since 27 September.

Paused campaigns cost nothing, so it can stay where it is. Two other things can look like duplicates and are on purpose: Skin Won't Heal runs in more than one campaign, and the retargeting ads use the same posts as ads in the broad campaign, so the likes and comments build up in one place. They do not compete with each other: retargeting only reaches people who already know STANK, and the interest groups leave those people out.

From Meta's activity log. Most of what was switched off on Friday morning had spent little; the two that had spent the most were the Pants image in broad and the reworded Skin Won't Heal in clean beauty. What is running now: Skin Won't Heal in the odour-conscious group and in retargeting; Frank's RedHot, Coconut oil and Dysbiosis in the broad campaign.

Meta's own help pages, read on 16 October 2026, and the published practice the recommendations lean on.

• Meta, About the learning phase
• Meta, Significant edits and learning phase
• Meta, About the attribution setting
• Meta, Use detailed targeting
• Meta, Audience controls and suggestions in Advantage+ audience
• Meta, Create a lookalike audience
• Meta, About managing ad volume
• Meta for Business, Performance marketing on Meta
• Jon Loomer, 5 Meta ads tests that transformed my perspective on targeting (23 Oct 2024)
• Jon Loomer, Detailed targeting is mostly a suggestion (11 Feb 2026)
• Flighted, How to scale Meta ads without decreasing ROAS in 2026
• Channable, Meta creative testing