Facebook Ads for an Online Store (2027)

What you are actually buying when you run a Facebook ad

Before any of the mechanics, get the shape of the thing right. You are not buying customers. You are buying impressions in an auction, and everything downstream (whether those impressions become clicks, and whether those clicks become orders) is decided by your creative and your offer, not by the buying.

The pool is enormous. DataReportal's Essential Facebook Stats put Facebook's reported advertising reach at 2.28 billion people in January 2025, about 39.4% of all adults aged 18 and over. That is a platform-reported figure, so it may count duplicate accounts, and it is worth noting that DataReportal's most recently published consolidated Facebook ad-reach number is still the January 2025 one. Read it as scale, not precision. The point stands: you are not going to run out of people. You are going to run out of patience and budget while you figure out what to say to them.

That is why the order of this guide matters. Research, then creative, then launch, then measurement. Most beginners invert it: they open Ads Manager first, spend three days on campaign structure, launch one product photo, and then wonder why nothing sold. The campaign structure was never the problem.

Why creative is the lever, and targeting mostly is not

Here is the number that should reorder your week. NCSolutions analysed nearly 450 CPG campaigns and attributed 49% of incremental sales to creative. Brand factors came second at 21%, reach at 14%, targeting at 11%, and recency last at 5%. Creative alone outweighed reach, targeting and recency put together.

This lines up with how Meta's delivery stack now works, which is the second reason to stop obsessing over audience settings. Meta's system is retrieval-driven rather than advertiser-audience-driven. As Search Engine Land describes it, Andromeda decides which ads are even eligible to show a given person based on historical engagement, creative elements and format, and GEM feeds predictions into that retrieval step. Meta reports GEM is "4x more efficient at driving ad performance gains" than its original ranking models.

Translate that into store-owner terms. You are no longer instructing the machine who to find. You are giving it material to work with, and the material is your creative. Variety of format (image, video, carousel, UGC, testimonial) is the main lever a beginner actually controls, because format and creative elements are inputs to eligibility itself. Ten interest checkboxes will not save one boring product photo.

Step 1: Research the angle, and capture it before it disappears

Start where the money already is. Open the Meta Ad Library, search the brands selling something adjacent to your product, and read what is running right now. Not the branding. The angle. Is the winning ad in your category leading with a price, a problem, a face, a before-and-after, a founder talking to camera, or an unboxing? Write down the pattern, not the pixels.

There is one hard constraint here that nobody warns beginners about, and it will bite you. Meta's Transparency Center confirms that all active ads are searchable in the Ad Library, and it documents exactly two long-retention rules. Ads about social issues, elections and politics are stored for seven years regardless of status. Ads delivered in the EU are archived for one year after their last impression. No equivalent retention is documented for ordinary ecommerce ads outside the EU, which is why, in practice, everything you actually care about is gone from the library the moment the advertiser pauses it.

So a "look back at what worked last Black Friday" is not a thing you can do in the Ad Library in January. The ads are not there. This turns research from a task you do when you need it into a habit you keep: screenshot and file as you go, month by month, or work from a searchable archive such as AdPlay.ai that retains the creative after the ad stops running. Either way, decide now, because the ads you did not capture this season are not recoverable next season.

What you are looking for, concretely: the three or four recurring angles in your category, the hook style each one uses in the first two seconds or the first line, and the offers that keep reappearing (bundles, free shipping thresholds, first-order discounts, seasonal sets). Those repeat because they work. Your job in step two is to pick the one your catalog can genuinely deliver.

Step 2: Build an offer your catalog can actually carry

Assume your store platform is already connected to Meta and your products are syncing into Commerce Manager. That connection is a prerequisite, not this guide's subject: get the sales channel or feed wired, confirm the catalog populates, and come back. What matters here is what you point the ads at.

Pick one hero product, not your whole range. A beginner campaign that advertises forty SKUs is a beginner campaign that learns nothing, because no single creative gets enough exposure to tell you anything. Pick the product with the best margin and the clearest story, and build the offer around it.

Then do the arithmetic before you write a word of copy. If your product sells for $60 and costs $20 landed, your gross margin is $40, so your break-even cost per purchase is $40 and your break-even ROAS is 1.5x. That means the campaign has to bring in $1.50 for every $1.00 spent just to stand still, before you have paid yourself, the packer, or the payment processor. Write that number on a sticky note. It is the only benchmark that can honestly tell you to switch an ad off, and it is far more useful than any published average.

Do the same sum for a bundle, and watch it surprise you. If bundling two units at $100 with a $40 landed cost lifts margin per order to $60, your break-even ROAS rises to about 1.67x: a bigger basket at a thinner margin rate (60% against 66.7% on the single unit). That is still usually the better trade, because you collect $60 of gross margin per order instead of $40. But note which way the number moved. The ROAS you have to clear went up, not down. This is exactly why a bundle campaign that "looks worse" on ROAS can be making you more money, and why "raise the average order value" is the most underrated ad tactic there is: it changes the maths of every campaign you will ever run, without touching Ads Manager.

Step 3: Generate creative variety, deliberately

Now build the material. Given the NCSolutions finding and how Meta's retrieval stack works, the goal is not one perfect ad. It is a spread of genuinely different creatives that give the system distinct things to try.

Different means different angle, not different crop. Five versions of the same product shot on five background colours is one creative wearing five hats. Instead, build across the angles you found in step one:

• A straight product showcase, clean, with the product large and the offer legible.
• A problem and solution cut that opens on the annoyance your product removes.
• A founder or UGC piece, someone real holding the thing and talking about it.
• A testimonial or social proof version built on a genuine customer line.
• A carousel that walks a range or a use-case sequence.

That is five inputs, not five decorations. Real ad examples show what this looks like in practice: Skinlycious runs a founder-to-camera piece, a before-and-after, and a plain discount showcase concurrently, because each one gets retrieved for different people in different moments. Getha does the same thing across a quieter range, pairing a new-product announcement with a straight showcase.

The format mix in the wild backs this up. Across 185,496 archived consumer-goods ads, 52% run as video and 32% as a static image, with the rest split across carousel, catalog and multi-image formats (AdPlay.ai archive, 2026). Video leads, but a third of the market is still static, which is the useful part: the brands doing this well are not picking one format and defending it. They are feeding the system several.

Write the copy to the angle, not to the product spec sheet. The first line does the work: it is what gets read before someone decides to keep scrolling. Lead with the outcome or the tension, put the price where it is unmissable if price is your angle, and keep the call to action boringly literal ("Shop the set", "See the shades"). Clever costs clicks.

Step 4: Edit for the feed, not for your desktop

The edit is where most store creative quietly dies. Three rules cover almost all of it.

Cut for silence. Assume it will be watched muted, which means any claim carried only by a voiceover is a claim nobody heard. Burn captions in. If the video makes sense muted, it makes sense.

Cut for the first two seconds. The hook is not the logo. Open on the product in use, the problem, or the face. The brand can arrive later, once you have earned the attention to show it.

Cut to fit the placements you actually want. Vertical for Stories and Reels, square or 4:5 for the feed. One asset stretched across every placement will look wrong somewhere, and "wrong" reads as "ad" faster than anything else.

Then check the thing everyone forgets: legibility at thumbnail size. Look at your creative on a phone, at arm's length, for one second. If the offer is not readable in that second, the offer does not exist.

Step 5: Launch, with the four things that changed recently

Meta's prerequisites for Advantage+ catalog ads (the catalog-driven format that used to be widely called dynamic product ads) are short and specific, per Meta's developer documentation: a catalog set up in Meta Business Suite or Commerce Manager, a Facebook Page, and an ad account with registered payment information. An Instagram account is optional and only relevant for cross-platform delivery. The Pixel is not strictly required for a basic catalog ad to run. It is required in practice, because retargeting a dynamic product audience needs product-level events (ViewContent, AddToCart, Purchase) flowing from your site, and Meta's catalog-ads page does not spell out the exact event requirements, so check its dynamic product audience documentation before you rely on it. In practice, wire the Pixel and confirm the events are landing in Events Manager.

Four things about launching have changed recently enough that most of what you will read online is wrong.

Advantage+ Shopping Campaigns no longer exist as something you create. Meta announced deprecation of the legacy ASC and AAC APIs on 8 October 2025, with Marketing API v24.0 blocking new creation and full deprecation across all API versions at v25.0 in Q1 2026. A campaign now enters an Advantage+ state automatically when Advantage+ budget, Advantage+ audience and Advantage+ placements are all switched on. There is no "Advantage+ Shopping" button to hunt for. Meta reports a 22% average ROAS improvement for Advantage+ sales campaigns. The underlying dollar figures Meta has published ($4.52 ROAS against a $3.71 US average) date from the legacy Advantage+ Shopping Campaign product and are US-scoped, so treat them as directional history rather than a number your store should expect. Either way it is Meta's own reported data, not an independent study.

Dynamic Media is on by default for catalog ads. Rollout began 1 September 2025, with 100% enforcement from 20 October 2025 for new Advantage+ catalog ads created through the Marketing API, covering single image, carousel and collection formats. If your catalog carries product videos, Meta may serve one instead of the image you had in mind. The documented opt-out is an API field, media_type_automation set to OPT_OUT, which makes it your feed or integration partner's lever rather than a checkbox in Ads Manager. Working in the UI, your practical move is to audit what media your feed is actually carrying.

The Conversions API no longer needs a developer, for web events. Meta added a one-click Conversions API setup in Events Manager, announced 15 April 2026 and live globally by late April 2026, available to any advertiser with an existing Pixel and web data source. It mirrors your existing Pixel events and parameters with built-in deduplication. App and offline events still need a real integration. Meta also shipped an AI-assisted Pixel upgrade that auto-attaches product names, availability and business details to events. Meta reports a 17.8% lower average cost per result for advertisers running the Conversions API for web events, which is Meta-reported and directional, not independent.

The Meta Pixel has not been renamed or retired. Meta's developer documentation still calls it the Meta Pixel. "Dataset" is the container in Events Manager that the Pixel connects through, not a replacement name. If a guide tells you the Pixel is dead, close the tab.

Then submit and wait. Meta's Transparency Center says ad review runs automatically before ads start and is typically completed within 24 hours, and that ads remain subject to review and re-review at all times. Read that second clause twice. Approval is not permanent, and a re-review can pull a live campaign down weeks in. Keep a spare approved creative paused in the account so you are never one disapproval away from zero delivery.

Step 6: Measure against your margin, not a benchmark

Now the honest part. You will want a number to compare yourself against, and the available numbers are weaker than they look.

The best openable data is WordStream's 2025 Facebook benchmarks: 554 US Traffic campaigns and 726 US Leads campaigns between 1 April 2024 and 30 June 2025, reported in USD, with the "averages" actually being medians. All-industry Traffic sits at $0.70 CPC and 1.71% CTR. Leads sits at $27.66 cost per lead, 7.72% conversion rate and $1.92 CPC.

Now watch that dissolve when you split it by retail vertical. In the same dataset, Shopping, Collectibles and Gifts runs a 4.13% CTR at $0.34 CPC, while Apparel and Fashion runs 1.29% at $0.86. Same period, same methodology, roughly 2.5x apart on both metrics. There is no single ecommerce benchmark, and anyone quoting you one is averaging across categories that behave nothing alike. Also note the scope: US-only, medians, USD. If you are outside the US, these are shape, not price.

Which brings you back to your sticky note. Your break-even cost per purchase, computed from your own margin, is the number that decides things. A $4 cost per purchase is thrilling on a $60 product and a disaster on a $6 one. Judge every campaign against your maths, and use published benchmarks only to check you are in a sane order of magnitude.

Then watch the calendar, because the auction is seasonal. Gupta Media's Social Media CPM Tracker, built on tens of millions of impressions using 14-day trailing averages, put blended Meta CPM (Facebook and Instagram) at $8.19 across full-year 2025. The annual swing is more instructive than any single month: in 2024, CPM averaged $13.42 in the Thanksgiving and Black Friday week (ISO week 48) against $6.05 in January 2024. Roughly double, for the same impression.

For a first-time advertiser that is a scheduling instruction. Learn in the cheap months. Scale in the expensive ones, but only with creative that has already proven itself, because Q4 is the worst possible time to discover your hook does not work. If you are planning backwards from the holiday peak, work out when to start rather than arriving in November with an untested ad.

What the first two weeks should actually look like

Beginners tend to expect a verdict on day two and then panic-edit their way out of ever getting one. Here is a saner shape for the first fortnight.

Days 1 and 2 are review and delivery, not results. Ads clear Meta's automated review, typically inside 24 hours, then start spending. Resist editing anything. Significant edits can send an ad back through review, and Meta's guidance is that they can restart learning, though the exact mechanics are hard to pin to a current neutral source. Treat "do not fiddle on day one" as the safe default rather than a documented rule.

Days 3 to 7 are your first read, and you read exactly two things: is anything getting clicked, and is anything getting bought. If a creative is drawing clicks but no purchases, the problem is almost never the ad. It is the price, the shipping cost at checkout, or the mobile page. Fix that before you touch the creative, because a better ad pushing more people at a broken checkout just costs more.

Days 8 to 14 are the cull. Turn off the creatives with no purchases and no clicks, leave the one or two that are moving, and put your next batch of variations against the winning angle rather than inventing a new one from scratch. That is the loop: you are not searching for a perfect ad, you are narrowing toward an angle and then making more of it.

The learning phase, and how much of it to believe

You will hit a "Learning" label on your ad set and immediately find a hundred articles quoting the same rule: roughly 50 optimization events per ad set per week to exit the learning phase. That figure is Meta's own published guidance in its Business Help Center, not independent measurement, so treat it as a directional planning heuristic and read Meta's "About the learning phase" article before you budget against it. Our own learning phase guide walks the same ground in more detail.

The practical takeaway survives the uncertainty anyway: an ad set optimizing for purchases needs a meaningful volume of purchases to learn from, and a store splitting a small budget across six ad sets gives each one too little to learn from. Fewer ad sets, more budget each, more creative inside them. That advice holds whatever the exact number turns out to be.

A troubleshooting table for the first month

Putting the loop together

Here is the whole thing in one paragraph, because it is smaller than it looks. Research the two or three angles already running in your category, and capture them, because outside the EU they vanish when the advertiser pauses them. Pick one hero product and compute the break-even ROAS your margin allows. Build four or five genuinely different creatives, not four crops of one. Cut them for silence, for the first two seconds, and for the placement. Confirm the three prerequisites (catalog, Page, funded ad account), switch on the Advantage+ levers, turn on the Conversions API for web events, and submit. Read the results against your sticky note, not against a US median. Then take what won and make three more of it.

The reason this order works is the reason the NCSolutions split matters: creative did 49% of the work and targeting did 11%, so the hours you spend are best spent on the thing that moves the number. Meta's retrieval stack has only pushed further in that direction, because format and creative elements now feed eligibility itself.

None of this requires a big budget. It requires you to run the loop more than once. The first round tells you almost nothing except which creative is least bad. The third round is where a store starts to have an opinion worth scaling, and if you have timed it right, you get there in a cheap month with a proven ad in hand before the auction gets expensive.