Meta's Andromeda Algorithm: Your Creative Is Now Your Audience

If your Meta Ads accounts are seeing rising CPMs despite stable spend, or your detailed interest targeting isn't performing the way it used to, you are not imagining things. Meta's delivery system has undergone a fundamental architectural shift — one that the platform has been quietly rolling out since late 2025 and which is now fully embedded across most campaign objectives and placements. The system is called Project Andromeda, and it has changed the basic logic of how Meta decides who sees your ads.

The core change: Meta's algorithm no longer starts with your audience selection. It starts with your creative. Andromeda evaluates your ad content first — the imagery, the copy, the hook, the format — and uses those signals to predict which users are most likely to respond. Manual interest targeting still exists in the interface, but under Andromeda, it functions more as a guardrail than a targeting engine. The creative itself is doing the audience finding. If your creative is weak, stale, or misaligned with your actual buyer, your delivery suffers — regardless of how well-built your interest stack is.

Accounts that haven't adapted are experiencing the consequences: CPMs creeping upward, learning phases that never fully exit, and Advantage+ campaigns that underperform on paper despite broad reach. The practitioners who understand what Andromeda actually optimises for are outperforming on cost efficiency right now. Here is what they know — and what you need to act on.

How Andromeda Actually Works

Traditional Meta ad delivery worked from the outside in: you defined an audience through demographics, interests, and behaviours, and Meta served your ad to the people in that pool who were most likely to take action. The quality of your creative influenced performance within that defined pool, but the pool itself was the starting point.

Andromeda inverts that logic. When you upload an ad, the system analyses the creative — its visual composition, the language patterns in the copy, the emotional tone, the product or service being featured — and cross-references those signals against Meta's behavioural graph to identify users whose patterns suggest receptivity to that specific creative type. Your audience selection then acts as a filter on top of that predicted cohort, not as the cohort itself.

The practical consequence is significant. A creative that speaks clearly and specifically to a defined buyer — their problem, their language, their context — will find that buyer more efficiently than a generic ad pointed at a tightly-constructed interest audience. Conversely, a creative that is vague, visually generic, or formatted identically to other ads in your account gives Andromeda weak signal and produces inefficient delivery. The algorithm cannot work out who this ad is for, so it casts a wider, more expensive net.

What the Data Shows

The shift is visible in performance benchmarks. Tinuiti's Q1 2026 Digital Ads Benchmark Report shows Instagram CPMs declining 3% year-over-year — the first year-over-year CPM decrease since 2023 — driven almost entirely by Reels share growth (up from 19% to 33% of Instagram impressions). But that aggregate figure masks the disparity between accounts: those running diverse, regularly refreshed creative formats are seeing CPMs hold or fall, while accounts with repetitive or static creative libraries are experiencing meaningful CPM inflation even as overall market CPMs soften.

The implication is that Andromeda is redistributing the cost advantage toward advertisers with strong creative operations and away from those relying on audience engineering to carry performance. In the old model, a mediocre creative with a great audience stack could produce acceptable results. Under Andromeda, that path is closing. The creative has to earn its delivery.

The Three Levers That Matter Now

Adapting to Andromeda comes down to three operational changes. These are not speculative best practices — they are the variables the algorithm is explicitly optimising around.

Creative volume and diversity. The benchmark used by practitioners who are winning under Andromeda is 15 to 20 active ads per campaign with meaningfully different hooks and formats — not the same visual resized for Stories and Feed, but genuinely distinct creative approaches. A direct product demo, a testimonial-style video, a problem-agitate-solve copy format, a UGC-style creative, a bold static with a single clear claim. Each creative gives Andromeda a different signal to work with, which means the algorithm can find different cohorts within your broad target, reducing overlap and keeping delivery efficient. Accounts running three to five ads in rotation under Andromeda are essentially asking the algorithm to find all their buyers with one or two signals — it cannot do that efficiently.

Campaign structure simplification. Andromeda performs better with consolidation. The evidence points strongly toward Advantage+ Shopping as the primary campaign type for e-commerce — fewer campaigns, broader audience parameters, letting the algorithm distribute budget across placements and creatives based on real-time performance signals. Manual campaign structures with heavy audience segmentation fight against how Andromeda wants to operate; they constrain the learning surface and reduce the data density per campaign. For lead gen accounts, Advantage+ Audience — where Meta expands your core targeting based on creative signal — is outperforming manually constrained detailed audiences in most verticals.

Signal quality. Andromeda's learning loop depends on clean conversion data. Pixel and Conversions API running simultaneously is table stakes — CAPI deduplication handles the overlap, and the combined signal volume is meaningfully better than either source alone. The metric to monitor is Event Match Quality in Events Manager: above 7 is healthy, below 6 means you are feeding the algorithm noisy data and it will take longer to find efficient delivery. The most common EMQ killers are hashed email mismatches between your CRM and Meta's identity graph, and missing customer information fields (phone, city, country) on your server events.

What to Stop Doing

Several practices that made sense in the pre-Andromeda world actively work against you now and are worth explicitly deprioritising.

Stop building elaborate interest stacks for prospecting. If you are spending time constructing detailed targeting combinations — layering interests, behaviours, and lookalikes to arrive at a precisely defined audience — you are working on a lever that Andromeda has largely taken out of your hands. The algorithm will override or expand your interest parameters based on creative signal anyway. Your time is better spent on the creative brief than the audience builder.

Stop running the same creative for more than four to six weeks without variation. Under the old model, a high-performing creative could run for months with gradual efficiency decay. Under Andromeda, once the algorithm has found and converted the most receptive users for a specific creative signal, CPMs for that ad rise quickly. The creative's useful life is shorter now, which means creative production cadence has to increase — or you accept the CPM penalty.

Stop treating campaign structure optimisation as your primary performance lever. If your account has ten campaigns each targeting a slightly different audience segment, the instinct when performance dips is to fiddle with budgets, audiences, and placements. Under Andromeda, that fidgeting costs you learning phase resets without addressing the actual issue. The primary lever now is creative. Build the habit of auditing creative before you audit structure.

The Strategic Shift for Agencies

For agency teams, Andromeda accelerates a transition that was already underway: the value of Meta Ads management is shifting from audience architecture toward creative strategy. Clients who come to agencies expecting sophisticated interest targeting to be the edge are operating on a model that is becoming less true every month. The agencies outperforming right now are those that have either built in-house creative capability or built strong systems for briefing and iterating on client creative at the speed Andromeda requires.

This has practical implications for how you scope and price client engagements. If your Meta retainer doesn't include a clear creative review and rotation cadence — with explicit ownership of who produces new ads and at what frequency — you are setting up for CPM drift that neither you nor your client will be able to explain without understanding Andromeda's mechanics. Getting ahead of that conversation now, before performance dips force it, is the professional approach.

The underlying message from Andromeda is the same one Meta has been signalling for two years: the platform wants to do the audience finding. What it needs from advertisers is better creative fuel. The accounts that have accepted that trade — and have built their operations around producing diverse, high-quality creative at volume — are the ones where Andromeda's learning loop compounds into a genuine cost advantage. The accounts still trying to beat the algorithm with audience engineering are fighting a battle the platform has already won.

If your Meta accounts are seeing CPM pressure or learning phase instability and you want a fresh set of eyes on the creative and structure — reach out to NovaReach here. We work across e-commerce and lead gen verticals and can identify exactly where Andromeda is working against your current setup.