Service 13 · Paid Social
Meta rebuilt its delivery system: ads are now matched to people by what the creative itself signals, not by hand-picked audiences. Winning under Andromeda takes the two things most accounts lack, clean conversion signals and genuine creative variety, and our program engineers both.
Under Meta's Andromeda and GEM ranking systems, weak conversion signals throttle delivery no matter how good the creative is. We run the Pixel and Conversions API together with shared event IDs for deduplication, then push Event Match Quality above 7 out of 10 by passing hashed emails, phone numbers, and click IDs server-side. For many accounts this single fix explains months of 'the algorithm stopped working.'
The optimization event has to be the real business outcome, purchase, qualified lead, never a top-of-funnel proxy. For lead gen we wire CRM feedback loops so Meta optimizes toward leads that actually close; Meta's own testing showed about 10% lower cost per qualified lead with conversion-leads optimization.
Account structure gets consolidated, because fragmentation is the classic legacy failure: dozens of ad sets slicing the same audience, each stuck in learning. The Andromeda-era build is one campaign per objective, broad targeting, Advantage+ on, and budgets sized so ad sets reach roughly 50 conversions a week and exit learning.
Meta reads each ad's visuals, spoken content, text, and tone, fingerprints it as an entity, and matches it to people. Twenty near-identical variations collapse into one entity and cannibalize each other in the auction. The unit of testing is the concept, so we build a matrix: persona by value proposition by format, UGC, demo, expert talking-head, lifestyle, social proof.
The working set is 8 to 15 active ads per ad set: proven leaders, mid-tier performers, and fresh tests, with new concepts making up 20 to 30% of each refresh cycle. Refreshes trigger on fatigue signals, falling first-time impression ratio, rising frequency, rising CPA, never on a calendar. Failed hypotheses get logged so the same dead concept isn't rebuilt in month three.
Measurement stays honest. Platform-reported ROAS is modeled, so we triangulate: Meta's incremental attribution setting, blended MER and CAC, and geo holdouts on larger accounts. Cost per lead is always reported next to qualified rate and closed revenue.
Pixel + CAPI with deduplication verified; Event Match Quality graded and pushed above 7.
Fragmented ad sets merged into one campaign per objective so learning can complete.
Creative planned across persona × proposition × format, genuinely different concepts.
Real conversion events, budgets sized to exit learning, and no structural edits mid-phase.
First-impression ratio, frequency, and CPA trends decide refresh timing, never a calendar.
Incremental attribution, blended MER, and CRM quality loops, reported monthly.
Meta's delivery system now reads the creative to decide who sees it, so this program engineers the two inputs that decide outcomes: signal quality and genuine concept variety. Learning phases get protected, refreshes trigger on fatigue data, and reporting is triangulated instead of taken on the platform's word.
What we do
What you get
What we do
What you get
What we do
What you get
What we do
What you get
What we do
What you get
Weeks 1–3
Every week
Every month
Every quarter
The stack we run for this
Meta Ads Manager
Google Analytics 4
Looker Studio
HubSpotYes, for advertisers who rebuilt their signal stack. Running the Pixel and Conversions API together with Event Match Quality above 7 restores most of the match quality iOS and ad blockers stripped away. Meta's modeled and incremental attribution also measures more honestly than the old last-click inflation, so the results you see are more trustworthy, not less.
Advantage+ is Meta's AI campaign framework, and since Meta merged the manual and Advantage+ setup paths in 2025 it is the default structure rather than a separate toggle. For most spend, yes: Meta reports around 17% lower CPA versus comparable manual campaigns. It only outperforms when fed genuinely diverse creative concepts and clean conversion signals, which is exactly what our program supplies.
Eight to fifteen active ads per ad set, refreshed every one to two weeks, each a genuinely different concept. Twenty crops of the same video accomplish nothing: Meta fingerprints similar files as one entity, so they compete with themselves in the auction instead of testing anything. Vary persona, value proposition, and format.
Enough to buy roughly 50 conversions per week per ad set, which is what the learning phase needs to complete. At the reported average lead cost of about $28, that is a floor near $1,400 a week for lead generation. Optimizing to cheaper upper-funnel events lowers the floor but trades away lead quality, so we set the event first and the budget second.
Usually creative fatigue plus weak signals, not the algorithm punishing you. Check the first-time impression ratio, frequency, Event Match Quality, and how different your active concepts really are before touching budgets. In audits, a stale concept mix and an EMQ under 5 explain most of the sustained CPL creep we see.
Adjacent disciplines
