In short
Yes, for a lot of stores, but the job is not the one it was five years ago. Precise interest targeting has given way to broad audiences and automated delivery, so the creative and the quality of the data you send back are now doing the work that targeting used to do. Stores that produce a steady stream of new creative and send clean server-side purchase data tend to do well. Stores relying on a narrow audience and two old images tend not to, and read that as the platform rather than the setup.
The lever moved from who you target to what you show them
Delivery systems now find the buyer largely on their own, given a broad enough pool and a clear enough objective. What they cannot do is invent a reason for someone to stop scrolling, and that is the part left to you.
In practice this means the account structure has got simpler and the production requirement has got heavier. Fewer ad sets, broader audiences, and a lot more creative moving through them. A campaign that would once have been built from twenty interest segments and one image is now more likely to be one audience and twenty pieces of creative.
It also means creative is where the testing budget goes. Testing a headline against another headline is a smaller question than testing a completely different idea about why the product matters, and the second is where most of the difference lives.
For a store, the useful mental shift is to stop treating the ad account as a targeting problem and start treating it as a publishing schedule. The account needs new material at a predictable rate, and the rate is set by how fast the audience sees the old material.
Signal quality decides how well the delivery can work
Browser-based measurement lost a share of its coverage when tracking prevention and app-level permissions became normal, and everything downstream of that depends on how much of the gap you have closed.
The server-side conversions integration is the main repair. Sending purchase events from your server, matched with hashed customer details and deduplicated against the browser event using a shared event identifier, restores a large part of what the browser alone no longer reports. On Shopify the official channel can do most of this, and the part worth checking is that deduplication is actually working rather than double counting.
Match quality matters as much as coverage. The more identifying fields you send with each event, and the more consistently they are formatted, the more events can be tied to a person, and the better the delivery system can find similar people. Meta reports on this in the events tooling, and improving it is usually cheaper than raising the budget.
Consent sits over all of it. Where a visitor has declined, events should not be sent, and the honest position is that some coverage is unavailable by design. Build the reporting to expect that rather than treating it as breakage.
It suits some catalogues far more than others
The best fit is a product with visual appeal and a story that can be told quickly to somebody who was not looking for it. Apparel, homeware, beauty, food and drink, pet products, anything with an obvious before and after: these are categories where a scroll can be interrupted by a picture.
The second good fit is a product with a defined problem. If the buyer knows they have the problem but did not know a product existed for it, a short demonstration does what a search ad cannot, because nobody is searching.
The harder cases are commodity products bought on price and specification, replacement parts, and business supplies where the buyer purchases on a schedule from a chosen supplier. Those exist and are best reached at the moment of intent.
Order value and repeat rate change the arithmetic more than category does. A modest average order that reliably repeats can support acquisition costs that a one-off purchase at the same price cannot, which is why the retention programme belongs in this decision.
Judge it on the store, not on the platform's own column
The reported return inside the ad platform includes attribution choices that your accounts do not share, so it will not match your order records and was never going to. Use it to compare campaigns against each other and use your own numbers to decide whether the channel deserves the budget.
The practical scoreboard is three figures. Total marketing spend against total store revenue, which tells you whether the whole programme is efficient. New customers acquired and what they cost, which tells you whether the spend is buying growth. And contribution after product, shipping and fees, which tells you whether any of it made money.
For a real answer on incrementality, hold something back. A geographic split or an on-and-off schedule over several weeks will tell you what the spend added, and that is a different question from what the platform attributed.
Do this before concluding the channel does or does not work. Most of the disagreement about whether Meta ads still work is really a disagreement about which number people are looking at.
Two accounts, same budget, different setup
- Account A audiences
- Six narrow interest sets
- Account A creative per month
- Two images, unchanged
- Account A purchase signal
- Browser pixel only
- Account B audiences
- One broad audience
- Account B creative per month
- Eight concepts, several cuts each
- Account B purchase signal
- Server-side events, deduplicated, good match quality
- Where the difference shows
- Cost per new customer, over a quarter
An illustrative contrast rather than a measured result. It describes where effort is best spent today, not a promise about what either setup would return.
Related questions
Is broad targeting really better than detailed interests?
For most stores with a reasonable budget and clean purchase signals, broader pools give the delivery system more room to find buyers, and the creative does the qualifying. Narrow targeting still has a place for genuinely niche products, small budgets and tightly defined professional audiences.
Do I still need the browser pixel if I send server-side events?
Yes, in most setups. Running both and deduplicating them with a shared event identifier gives better coverage than either alone, because each catches events the other misses. What you must avoid is running both without deduplication, which counts a single order twice.
How long before a new Meta account can be judged?
Give it enough time to move through a full purchase cycle plus a creative rotation, which for most stores means several weeks rather than several days. Judging in the first week measures the delivery system settling rather than whether the channel suits your product.