The hardest part of Meta advertising in 2026 is not the creative or the bidding. It is that you can no longer fully trust the numbers in your dashboard. Privacy changes, browser restrictions, and people clearing or blocking everything mean the pixel sees less than it used to. Run a store long enough and you will notice Meta reporting fewer sales than your actual checkout, sometimes by a wide margin.
You cannot fix that by wishing the old days back. You fix it by feeding the platform better data and by changing how you read results. Here is the setup I use to stay profitable when tracking is unreliable.
Send the data server-side
The single most important change is moving from a browser-only pixel to server-side tracking through the Conversions API. The browser pixel gets blocked, throttled, and cleared. A server-side event sent from your own backend does not.
I run both together, with proper deduplication so the same purchase is not counted twice. The result is that Meta sees a much higher share of conversions, which does two things. It makes your reporting closer to reality, and, more importantly, it gives the algorithm cleaner signal to optimize on. Better data in means better delivery out, and that shows up directly in cost per acquisition.
If you do one technical thing this quarter, make it this. Everything else is secondary.
Stop micromanaging the structure
When data was perfect, splitting campaigns into many tiny ad sets made sense. You could read each one and prune. With less reliable, more delayed signal, that approach now starves the algorithm. Small ad sets never gather enough conversions to learn, so they thrash.
I run consolidated structures: fewer, broader ad sets with enough budget and enough conversions to actually exit the learning phase. Meta's broad and automated options, the Advantage+ style campaigns, work well precisely because they pool signal instead of fragmenting it. Give the system room and volume, and let it find the people. Your job moves from manual targeting to feeding it good creative and good data.
The instinct to control every audience worked when tracking was precise. Today it just denies the algorithm the volume it needs to learn.
Read ROAS at the account level, not the ad
Here is the mindset shift that keeps me sane. In-platform ROAS per ad is now an indicator, not a verdict. The reporting is incomplete and delayed, so judging a single ad on day two by its dashboard ROAS will lead you to kill winners that simply had not reported their sales yet.
I zoom out. The number I actually trust is the relationship between total ad spend and total revenue across the whole account, compared against what the business needs to be profitable. That is real money in and real money out, and it does not lie the way per-ad attribution can.
MER is the number that does not lie
So the metric I anchor on is MER, the marketing efficiency ratio: total revenue divided by total marketing spend. It ignores attribution entirely, because it only cares about two real figures, what you spent and what you made.
If MER is healthy and growing as you scale spend, the program is working, whatever the pixel claims about any single campaign. If MER drops as you spend more, you are buying unprofitable growth, even if a dashboard somewhere shows a flattering 6x. I check the platform numbers for direction and the MER for truth.
Give the algorithm a few days
Delayed reporting changes the rhythm of decisions. A purchase today might not show in the dashboard for a day or two, especially the server-side ones. So I stopped making changes based on a few hours of data.
I let new campaigns run for several days before judging them, and I avoid the temptation to tweak budgets daily. Every edit resets learning. Patience is a performance lever now, not just a virtue.
A simple, durable setup
Put together, the system I run looks like this:
- Conversions API plus pixel, deduplicated, so Meta sees most of the truth.
- Consolidated, broad campaigns with enough budget to exit learning.
- Strong creative volume, because with targeting handed to the algorithm, the creative is now the main lever you control.
- MER as the scoreboard, with in-platform ROAS as a directional hint.
- Decisions on a multi-day cadence, not a multi-hour one.
The takeaway
You are not going to get perfect attribution back. That era is over. The advertisers who keep winning on Meta have made peace with it: they feed the platform clean server-side data, give the algorithm volume and good creative, and measure success by whether real revenue outpaces real spend.
Imperfect data plus a sound system beats perfect data that no longer exists. Build for the world you are actually in, and the ROAS follows.