On Meta, hand-built targeting has mostly disappeared. With Advantage+ Shopping Campaigns and Advantage audiences as the default path, the algorithm decides who sees the ad.
It is tempting to conclude that there is less for an operator to do. The opposite is true. The more targeting is automated, the more performance depends on what you feed the algorithm. How the account is consolidated, how accurately conversion signals arrive, and how the reported results are read: those remain.
BGROW treats these three as the real variables in Meta operations. Creative and budget allocation work on top of them.
1. What automation took, and what it left
Meta's automation finds the optimum faster than a person can across audience discovery, bidding, placement and creative combinations. Manual operation does not beat an algorithm that reacts to millions of impressions in real time.
But what the algorithm does is optimise within the conditions it is given. Which objective to optimise for, which conversion counts as success, and how much data it has to learn from are set by whoever designs the account.
Give it the wrong conditions and it will optimise precisely in the wrong direction. Set add-to-cart as the objective and it will find people who add to cart, and the campaign will report well even when nothing sells. A large share of underperforming accounts are not badly run; they are badly configured.
2. Account structure is not a filing system
Splitting campaigns and ad sets makes reporting easier to read. It also tends to cost performance.
Meta puts every ad set through a learning phase, and exiting it requires roughly 50 optimisation events in seven days. Divide one budget across five ad sets and all five can fall short, leaving the whole account in a learning limited state.
So the first thing to look at when designing structure is not a tidy taxonomy but the rate at which conversions accumulate. Convert monthly conversions to a weekly figure, divide by fifty, and that is the ceiling on how many ad sets the account can support. The fewer the conversions, the simpler the structure has to be.
There are real reasons to separate ad sets: a budget that must be guaranteed, a different bidding strategy, or results that become uninterpretable when mixed. Absent those three, consolidating is usually better.
3. The algorithm is only as accurate as the signal it receives
Conversion signals are the raw material of optimisation. When signals go missing or arrive late, the algorithm looks for the wrong people. No amount of creative polish fixes a leak here.
A browser-side pixel alone does not deliver the full picture. Ad blockers, cookie restrictions and Apple's App Tracking Transparency policy all remove part of it. When BGROW takes over an account, the first thing reviewed is this delivery path, not the creative.
Conversions API. Sends events server to server so they survive browser-side loss.
Deduplication. Pixel and Conversions API must share an event ID so one purchase is not counted twice. Get this wrong and ROAS reads better than reality.
Event Match Quality. A score out of ten for how well the customer data sent from your server matches real accounts. Meta recommends 6.0 or higher.
Domain verification and event priority. Aggregated Event Measurement caps prioritised conversion events at eight per domain. For users who opted out of tracking on iOS, ads optimised for events outside that list will not deliver properly.
None of this is visible work. There is nothing to show for it the way a new creative can be shown. Yet with the same budget and the same creative, an account with clean signal and one without perform differently.
4. Reported conversions are a point of view, not a fact
The conversion number in a Meta report is not a measurement. It is the output of a rule, and changing the rule changes the result.
The default attribution setting is 7-day click and 1-day view. Someone who merely saw the ad and bought within a day is counted, even if they were going to buy anyway. Someone who buys on the eighth day is recorded nowhere.
Running several channels compounds this. When Meta, Google and Naver each claim the same order under their own rules, the sum of channel revenue exceeds actual revenue. Every report is correct and only the total is wrong.
So BGROW uses platform ROAS as an operating metric but not as the final verdict. It is read alongside marketing efficiency ratio against real revenue, and cross-checked against whether total sales actually moved when spend went up or down. No single number is trusted on its own; several angles have to point the same way.
5. Why a separate analytics layer
Past this point, an ads manager screen is no longer enough. There are three reasons.
First, a platform report is only accurate inside its own channel. Meta shows Meta's attribution. Seeing across channels requires a reference point outside of them.
Second, ad data alone does not complete a decision. Which products actually sold, whether stock remains, whether the buyer is new or returning: without those, there is no next action. A high ROAS on an out-of-stock product is not usable information.
Third, no one reviews the same screen at the same depth every single day. As accounts multiply, warning signs get missed, and one missed day can spoil a week of budget.
A tool is needed not to make numbers look better but to run the checks a person cannot repeat reliably.
6. How BGROW uses DATAWAVE
BGROW built this layer in-house as DATAWAVE rather than buying it. Fashion commerce does not operate in the units a general dashboard assumes; season, drop, product and option do not map onto generic categories.
Ads and revenue sit on one screen. Spend and conversions by channel are read next to store orders, product-level revenue and inventory. Attribution can only be interpreted when the gap between platform reports and real revenue is visible.
Anomalies are detected automatically. Budget pacing, rising frequency and sudden drops in conversion rate are given thresholds and trigger alerts. Finding a problem a day earlier wastes a day less of spend.
Creative results accumulate at the element level. A creative is recorded not as one artefact but as a combination of angle, format, model and copy, so the hit rate of the next production improves.
Metrics are defined per brand. A brand that needs new customers and one that needs repeat purchases have no reason to share a dashboard.
After putting this into practice, the time taken to collect, analyse and act on advertising data fell from about an hour to twenty minutes. That time goes into looking at problems outside the data, not into producing more reports.
7. Tools do not make the judgement
A better dashboard does not raise performance by itself. It lets you see faster and more accurately; what to do about it is still decided by people.
Two people can read the same screen and reach different conclusions. When conversions fall, whether to cut budget, swap creative, fix the product page or rethink the assortment is not something the numbers announce. Experience running fashion brands fills that gap.
In short, three variables remain controllable on Meta: consolidating the account so it can learn, making sure conversion signals arrive intact, and re-validating reported numbers against real revenue. Creative and budget only work properly once those three stand.
The more the algorithm takes on, the more the remaining judgement decides the outcome.