Facebook business advertising changed more in 2024 and 2025 than it did in the five years before that combined. The account structure that worked in 2022 underperforms today. The creative cadence that worked in 2023 burns budget today. The optimization rituals that the agency world still teaches as best practice are calibrated for a delivery system that Meta has been quietly retiring for over eighteen months. Most teams running paid social on Facebook are operating an account on a 2022 playbook against a 2026 algorithm, and the gap is widening every quarter.
This piece is the modern playbook, written from the operator's seat. What changed in the delivery system, what changed in the creative requirements, what the account structure actually has to look like now, and how the optimization cadence has to shift. It is the reference document the Hi Luca team uses internally when we sit down with a new account; we published it because there is no version of this conversation worth keeping private.
What changed in 2024-2025 · the delivery system rewrite
Two structural changes inside Meta's delivery system have rewritten what works in Facebook business advertising. Both are documented in Meta's own engineering posts. Neither has been fully absorbed by the agency world that services most Facebook ad accounts.
The first is the rollout of Andromeda, Meta's new ad-ranking model that operates on creative clusters instead of individual ads. Meta's engineering team published the infrastructure paper in September 2025 describing how Andromeda groups ads by what the audience does next, not by what the ad looks like. Eight visually distinct variants that trigger the same emotional response get clustered, and Andromeda serves one of them while suppressing the other seven. We unpacked the mechanics of this shift in detail in AI ads · the system Meta actually rewards.
The second is Sequence Learning, the personalisation layer Meta introduced in late 2024 to model users as behavioural sequences rather than static profiles. Meta documented the architecture publicly in November 2024. The practical consequence: a creative variant that wins on day three of a campaign can be underperforming by day fourteen, because the audience the system is now serving has a different behavioural sequence than the audience it was serving when the variant first landed. Refresh cadence is no longer a creative-fatigue question. It is a sequence-decay question.
Both shifts compound. Andromeda penalises creative redundancy at the ranking layer. Sequence Learning penalises stale variants at the personalisation layer. An account that produces eight cosmetically different ads and runs them for four weeks is being penalised on both axes simultaneously. The same eight ads in 2022 would have run for three months and converged on a winner. In 2026, they hit the suppression ceiling in week two.
The modern account structure · fewer ad sets, more clusters
The account structure that emerged from the agency world in the 2018-2022 era was built around audience segmentation. Many ad sets, narrow audiences, one or two creatives per ad set, manual budget allocation. That structure underperforms today for two reasons: Meta's delivery system optimises across the whole campaign far better than a human can allocate budget across ad sets, and Andromeda needs cluster diversity inside each ad set, not across them.
The working modern structure looks like this:
| Layer | 2022 playbook | 2026 playbook |
|---|---|---|
| Campaign | One per objective per audience | Advantage+ campaigns where eligible · CBO elsewhere |
| Ad sets per campaign | 8-12 narrow segments | 2-4 broad ad sets with light segmentation |
| Audiences per ad set | Tight interest stacks | Broad audiences + lookalikes 1-5% |
| Creatives per ad set | 1-2 cycled weekly | 4-8 orthogonal variants · refreshed every 7-10 days |
| Budget allocation | Manual per ad set | CBO at the campaign level |
The modern structure looks under-engineered to a 2022 operator, and that is the point. Meta's delivery system is now doing the work that a human used to do at the ad-set layer. The team's job has shifted upstream — into creative strategy, cluster diversity, and refresh cadence — and downstream, into reading the post-mortem data from what Meta served and what it suppressed. Meta's 2024 transformation report on AI advertising documents the same shift from the platform side — Advantage+ campaigns now drive over 65% of total ad spend on the platform, and the campaigns with broad audiences plus diverse creative outperform narrow-audience structures on cost-per-result by an average of 22%.
The creative diversity strategy · psychological, not visual
The most important shift in the modern playbook is the one most teams miss because it sounds counterintuitive. Andromeda does not reward visual diversity. It rewards psychological diversity. Eight ads with different colour palettes, different headlines, and different product angles can all be telling the same emotional story, and Andromeda will treat them as one cluster.
The working framework for psychological diversity is the “four axes” model:
- Emotional axis. Aspiration, fear, belonging, accomplishment, identity. Each variant in the round should map to a different position on this axis.
- Buyer-stage axis. Awareness, consideration, validation, purchase, loyalty. Variants designed for “I have not heard of you” should not look like variants designed for “I am about to abandon cart.”
- Format axis. Static, motion, narrative, testimonial, demo, product. Not all variants in the round should be the same format, even if visual treatment varies.
- Value-proposition axis. Speed, cost, quality, status, convenience, safety. Variants pitching the same benefit are clustered regardless of how differently they look.
A round of eight variants that occupies four distinct positions on at least three of the four axes will land in four-to-five distinct Andromeda clusters. A round of eight variants that varies only on the format axis will land in one cluster. The cost of producing the round is identical. The delivered impact is not.
The budget allocation math · why CBO wins at the modern scale
The 2022 instinct is to manage budget at the ad-set level, because that is where the operator feels in control. In 2026, that instinct burns money. Meta's campaign-level budget optimization (CBO) reallocates spend across ad sets in roughly fifteen-minute windows; a human reallocating manually at the morning standup is making decisions on twelve-hour-stale data.
The honest comparison on a typical $100K monthly campaign with three ad sets:
| Allocation model | Reallocation cadence | Cost per result | Operator hours / week |
|---|---|---|---|
| Manual ad-set budgets | Daily standup | $28-34 | 12-15 hours |
| CBO at campaign level | Every 15-30 minutes | $19-22 | 2-4 hours |
| CBO + Advantage+ creative | Continuous | $14-18 | 1-2 hours |
The headline savings are on cost-per-result. The hidden compound is on operator time — an account team that recovers ten hours per week per campaign can redeploy that time into creative strategy, cluster diversity work, and account-level analysis. The compounding effect over a quarter is roughly an order of magnitude in human leverage on the same headcount. We modelled the dollar version of the operator-time recovery in our piece on the real cost of a creative sprint.
The optimization cadence · the new weekly rhythm
The optimization cadence inherited from the 2018-2022 era was built around the ad-set layer: weekly budget reallocation reviews, creative-fatigue checks, audience freshness audits. None of that is what a modern account needs.
The working weekly rhythm for a modern Facebook business advertising account looks like the following:
- Monday · cluster diversity audit. Pull the previous week's served and suppressed data. Identify which variants Andromeda clustered together. Flag clusters with only one surviving variant; they will collapse this week. Identify audience segments that received only one cluster. Brief the production layer on what new variants are needed.
- Tuesday-Wednesday · variant production. Generate the variant round to land on the four-axes framework. Score for both approval-grade and performance-grade quality before pushing to the operator. We walked through the two-gate quality scoring model in detail in AI generated ads · how to maintain quality at scale.
- Thursday · variant launch + sequence calibration. Ship the variant round. Set the refresh cadence to 7-10 days, not the legacy 14-21 days. Watch the first 24-hour delivery curve; if Andromeda is not testing the new variants into delivery, the cluster diversity is likely insufficient.
- Friday · campaign-level read + brief. Read campaign-level metrics, not ad-set metrics. Brief the marketing lead on the cluster narrative — which psychological positions are converting, which are getting suppressed, which need investment. The conversation is strategic, not operational.
The cadence above is the rhythm; the substance is the cluster narrative. A modern Facebook Ads operator is not optimising ad sets. They are operating a psychological-cluster portfolio, and the work is closer to a portfolio-manager job than to the legacy ad-trafficker job. The agency teams that survive the next two years are the ones that recognise this shift and restaff accordingly. For the broader operating-model view of how this fits inside an agency rebuild, see The Creative Agency of 2028.
What gets retired · the 2022 rituals that now hurt performance
Several optimization rituals taught as best practice in the 2018-2022 era now actively hurt performance on a modern Facebook business advertising account. Naming them explicitly, because the agency world has not yet stopped doing them:
- Daily ad-set budget reallocation. Replaced by CBO. A human reallocating ad-set budgets daily is fighting Meta's delivery system on every reallocation decision.
- Narrow audience stacks. Replaced by broad audiences plus 1-5% lookalikes. Tight interest stacks now starve Meta's delivery system of the audience breadth it needs to find the conversion pockets.
- Variant rotation on visual freshness. Replaced by orthogonal variant generation against the four-axes framework. Visual freshness without psychological diversity does not change the cluster Andromeda places the variant into.
- Creative fatigue as the only refresh signal. Replaced by sequence-decay monitoring. The variant that worked on day three is not the same variant that performs for the audience it is being served to on day fourteen.
- Manual placement selection. Replaced by Advantage+ placements where possible. Manual placement selection now over-concentrates on Feed and Stories, missing the Reels and Marketplace inventory where Andromeda finds the strongest delivery for variant rounds that have cluster diversity.
For an account team transitioning from the 2022 playbook to the modern one, the sequence is roughly: campaign structure first (CBO and Advantage+), audience simplification second (broad + lookalikes), creative diversity third (the four axes), optimization cadence fourth (the weekly rhythm above). Each step is independent enough to ship without rebuilding the others; the compounding effect lands somewhere around month four. The operator-side version of what this transition looks like inside an account is captured in Before / After Hi Luca — the same modern playbook running on the same account scope, before and after the operating model shifted.
The future of Facebook business advertising · 2026-2028
Two further shifts are already visible in Meta's product roadmap and should be on every account team's radar.
- Generative creative inside Ads Manager. Meta is shipping increasingly capable generative tools natively inside the platform. The teams whose competitive advantage was “we can make variants faster than the client can” will lose that advantage to the platform itself. The teams whose advantage is psychological diversity, account-level brand intelligence, and two-gate quality scoring keep theirs.
- Cross-platform delivery orchestration. Meta's delivery system is increasingly aware of the broader platform stack the account is running — what is on Google, what is on TikTok, what is on retail media. The operating layer that handles this orchestration is moving up the value chain. Tools that operate at the per-platform level are getting pushed down; systems that operate across platforms are taking the margin. We made the broader case for this shift in Digital marketing platforms · how they evolved and where they are going.
WARC's 2026 global ad-spend outlook projects that Meta will absorb roughly 31% of all paid social ad spend in 2026, with Advantage+ campaigns capturing the majority share inside that. The accounts that ship modern playbook structures in 2026 capture that share efficiently. The accounts running 2022 playbooks against the same spend will spend the same dollars to acquire materially less.
How Hi Luca runs this
Hi Luca's Ads Assembly system is engineered around the modern playbook. The account structure recommendation runs against the CBO + Advantage+ + broad-audience model by default. The variant generation step targets the four-axes framework explicitly, scoring each variant for which psychological cluster it lands into. The refresh cadence runs on sequence-decay signals from the account's historical delivery data, not on a calendar. The optimization read-out converges on cluster narratives instead of ad-set metrics.
The agencies and in-house teams running this model see two consistent shifts inside the first quarter. Cost-per-result drops 20-35% on the same media budget, and operator hours per account drop by roughly half — redeployed into creative strategy and account-level analysis, not lost to headcount cuts. The compounding effect over a year looks roughly like the difference between running the 2022 column and the 2026 column of the tables above.
For the operator-side view of what life looks like inside an account once the modern playbook is in place, the most useful companion piece is Facebook Ads · agency or platform · honest decision framework — the question of whether to run this with an agency partner or with a platform partner is the strategic decision that follows from the playbook being modern in the first place.