The era of micro-targeting lookalikes and interest stacks is officially over. Meta's Andromeda ad-delivery algorithm is now powered by deep multi-modal machine learning. Today, your creative is your targeting.
Broad Audiences & Dynamic Creative Testing (DCT)
When you restrict your Meta campaigns to narrow interest groups, you artificially bid up your CPMs and restrict the algorithm from finding high-intent buyers.
Under our creative-first framework, we run Broad Demographics (Open Targeting) paired with 3:2:2 Dynamic Creative Testing (3 creative assets, 2 primary text variants, 2 headlines). Meta's algorithm matches specific visual hooks to the psychological profiles of users most likely to purchase.
Structure the test so the algorithm can learn. Run one CBO campaign per product line, place the testing ad set inside it, and keep targeting broad with only age, geography, and placement guardrails set. Meta's own guidance is to plan around fifty conversion events per ad set per week for stable delivery. Below that volume, consolidate products or optimize temporarily for a higher-funnel event such as AddToCart while data matures, rather than fragmenting spend across a dozen ad sets that each learn nothing.
Judge creative on a ladder of signals, cheapest first. Hook rate comes first — the share of viewers still watching at three seconds. Then outbound click-through rate, then cost per landing page view, and only then cost per purchase. As a working rule, let each creative spend roughly three to five times your target cost per acquisition before you call it a loser; two hundred impressions buys a guess dressed up as a verdict. Winners graduate into the scaling campaign untouched. Editing a winning ad to improve it mid-flight resets the very delivery you are trying to keep.
The Creative Engine: What to Produce Every Week
Andromeda rewards diversity of concept. Five near-identical unboxing videos are one creative to the delivery system, while a founder explaining the problem, a customer filming a rough phone review, a screen-recorded demo, and a static side-by-side comparison are four. When accounts stall in 2026, this is usually the cause: the team ships variations, and the system runs out of new signals to match against buyers.
Hold a weekly production rhythm. Three to five new concepts a week is a workable cadence for most active eCommerce accounts, mixing UGC-style talking heads, product demos, objection-handling clips, and statics for retargeting pools. The first two seconds carry the concept, so show the product, the problem, or the result immediately, and burn captions into every video, because a large share of feed viewing happens at low volume. Keep a simple testing log — concept, hook, angle, launch date, verdict — so nobody re-shoots an idea that already failed in March.
Concepts should come from evidence, not brainstorms. Mine your product reviews and support tickets for the exact objections buyers raise, read the comments under your own ads, and study what competitors keep re-running in the Ads Library, because an ad that survives six months in a rival account is a market signal. One proven angle, expressed by a different face in a different setting, counts as a new concept to the algorithm and to the viewer.
Watch for fatigue before the dashboard confirms it. Rising frequency in a prospecting pool, a click-through rate decaying week on week, and comments shifting from questions to complaints are the early warnings. Refresh the concept, not the caption. This production-and-testing loop is the core of our Meta Ads management work, not a monthly creative batch.
First-Party Defences: Conversions API Without the Double-Counting
The browser pixel alone is a leaking bucket. Between iOS opt-outs, ad blockers, and browsers expiring cookies, a common estimate is that fifteen to thirty percent of purchase events never reach Meta from the browser. The algorithm then optimizes toward the customers it can see and quietly undervalues the rest. Conversions API sends the same events from your server, and the two paths together give the delivery system the complete picture.
Three parts are non-negotiable. First, deduplication: browser and server events must carry the same event name and event ID, or Meta counts purchases twice and your reported ROAS inflates into fiction. Second, match quality: pass hashed email and phone plus the fbc and fbp click identifiers, then check Event Match Quality in Events Manager and aim for a score around six or higher on Purchase. Third, verification: run Test Events after every theme change, app install, or tracking edit, because CAPI integrations break silently and usually surface three weeks later as a mysterious performance slide. Most major commerce platforms offer a native gateway integration that handles this without custom development; use it before commissioning a bespoke server build.
Coverage matters as much as plumbing. Send the full journey — ViewContent, AddToCart, InitiateCheckout, and Purchase — each with value and currency attached, because Andromeda bids against predicted value, not raw conversion counts. A Purchase-only feed leaves the system blind to the many visitors who showed intent and left, and those mid-funnel events are what prospecting optimization falls back on while purchase volume is still thin.
Measuring True Profitability: Blended Marketing Efficiency Ratio (MER)
In-platform ROAS numbers reported inside Meta Ads Manager are frequently distorted by modeled attribution and post-view attribution windows.
We manage client spend through Blended MER (Total Store Revenue / Total Ad Spend) and Contribution Margin. By tracking how paid social acquisition directly impacts bank cash flow, we scale spend aggressively without falling into false attribution traps.
Scaling Rules: When to Push, When to Hold
MER tells you whether scaling is safe. It does not set the speed. Increase budgets in steps of roughly twenty to thirty percent every forty-eight to seventy-two hours while MER holds inside your target band, or graduate a proven creative into a separate scaling campaign at a meaningfully higher budget while the test cell keeps hunting. Pushing vertical budget steps and horizontal expansion in the same week is how accounts ruin a good month.
Expect a wobble after every step. Delivery re-learns for a day or two, costs jitter, and the temptation is to reverse the change by Thursday. Resist it. Judge a scaling step on a rolling three-day MER window rather than the first bad morning, and keep a written log of every budget change with its date. When performance moves, you want to know exactly which lever moved first.
Set the floor from your margins before spending, not after. If contribution margin after product cost, shipping, and returns runs around thirty percent, a blended MER below roughly 3.3 means paid growth is consuming the business while Ads Manager celebrates. Run the arithmetic with your own numbers — our ROAS and profit margin calculator exists for exactly this — and split new-customer MER from blended MER once repeat revenue becomes material, because prospecting may run leaner when lifetime value repays it inside your payback window.
The Mistakes That Stall Andromeda Accounts
- Account structures from 2021. A dozen narrow ad sets fragment budget and data until each one starves below the volume the delivery system needs to learn.
- Editing ads mid-learning. Significant edits reset delivery optimization, and changing budget, audience, and creative in the same week makes results unreadable.
- Verdicts on pocket change. Judging a creative before it has spent a few multiples of your target acquisition cost manufactures confident decisions from noise.
- CAPI installed, never verified. Deduplication failures and silent breakages corrupt the exact data that scaling decisions rest on.
- Great ads, weak destination. Slow mobile pages, unclear pricing, and a checkout that fights the buyer waste clicks you paid a premium to win. Post-click is half the system.
- Creative refreshed only when performance dies. When ROAS visibly collapses, replacements should already have a week of test data behind them.
Frequently Asked Questions
How much budget does a creative test actually need?
Enough for each concept to spend several multiples of your target cost per acquisition inside an ad set receiving steady delivery. For many stores that means protecting a testing budget in the region of ten to twenty percent of total spend, kept separate from the scaling campaign so experiments and winners never fight over the same money.
Broad targeting or Advantage+ — which should I run?
Both, with different jobs. Advantage+ sales campaigns are an efficient default for established catalogues with strong event history, while a manually structured broad campaign gives cleaner creative testing and budget control. Many accounts run Advantage+ as the scaling engine and a broad CBO as the laboratory; the split matters less than keeping those two roles separate.
Why does Ads Manager ROAS disagree with my store revenue?
Attribution windows, view-through credit, modeled conversions, and cross-device journeys all push platform numbers away from bank reality. Reconcile weekly against blended MER and actual store revenue, and treat in-platform ROAS as a relative signal between ads, never as financial truth.
How often should I launch new creative?
Weekly, in small batches, before fatigue forces your hand. Three to five fresh concepts a week keeps the test cell fed and maintains a bench of proven winners ready when a scaling ad decays — which it will, usually with little warning.