meta andromeda update: how to structure campaigns, creative and budgets in 2026

andromeda made the creative do the targeting, so the old account structure now works against you. here is how many ad sets, creatives and budgets an indian lead business should actually run, and how to scale without resetting learning.
in this article8 sections
- 01what did the meta andromeda update change about campaign structure?
- 02meta andromeda creative diversification: how many creatives do you actually need?
- 03should you run an advantage+ leads campaign or a manual lead campaign?
- 04broad targeting in meta ads: broad vs interest vs lookalike in 2026
- 05how to scale facebook ads without resetting the learning phase
- 06cbo vs abo in 2026: when does each still win?
- 07the account structure we would set up for an indian lead-gen advertiser
- 08who wrote this, how and why
our meta ads strategy for 2026 explains what andromeda is and why your creative is now your targeting. this guide goes one level down: how to lay out the account itself.
- andromeda is meta's ad retrieval engine. it narrows tens of millions of ads to a few thousand candidates per person before ranking, and meta reported +6% recall and +8% ads quality on selected segments at launch (meta engineering, dec 2024).
- an ad set stays in learning until it gets about 50 optimisation events in the week after its last significant edit (meta business help). every split you add makes that harder.
- in meta's early testing, leads campaigns with advantage+ on averaged 14% lower cost per lead and 10% lower cost per qualified lead than with it off (meta). that is meta's own figure.
- budget and bid changes may restart learning depending on how big they are, per meta. no official percentage exists.
what did the meta andromeda update change about campaign structure?
the meta andromeda update moved the job of finding buyers from your audience settings to your ads. so structure now exists to feed the system data, not to fence it in. one campaign per goal, as few ad sets as your hard limits allow, and several distinct concepts in each ad set is the default layout for 2026.
meta built andromeda for the far bigger number of ad variations that generative tools now create (meta engineering). its advice followed. ppc land reported that meta's november 2025 festive guide suggested at least three creative formats and advised against running several ad sets aimed at overlapping audiences (ppc land). we saw only that summary.
the old habit was to slice by interest, lookalike and age band. each slice got a thin budget and never left learning.
| structure decision | before andromeda (pre-2025 habit) | after andromeda (2026 default) |
|---|---|---|
| campaigns | one per audience or funnel stage | one per business goal, plus optional retargeting |
| ad sets | one per interest, lookalike or age band | one per hard limit: state, language, project or offer |
| targeting | stacked interests, tight lookalikes | location, age floor, advantage+ audience on |
| ads per ad set | 2 or 3, often near-copies | several genuinely different concepts |
| testing | new ad set per test | new concepts added to the live ad set in batches |
| budget | ad set budgets, adjusted daily | campaign budget, changed in planned steps |
a developer with two projects has two rera numbers and two price points, so two ad sets make sense. one offer in one language rarely needs more than one prospecting ad set.
meta andromeda creative diversification: how many creatives do you actually need?
meta andromeda creative diversification is about distinct concepts, not ad count. meta has not published a fixed number. our rule of thumb is three or four live concepts below ₹50,000 a month, four to six around ₹1 lakh, and six to ten from ₹1.5 lakh, each varying the person, problem, format or language.
what meta has said is directional. social media today reported meta's claim that campaigns using diverse creative across formats and messaging types saw up to 32% better cpa and 9% more incremental reach (social media today). meta published no method. the claim that "one ad set with 25 creatives beats five ad sets with five" is widely repeated, but we could not trace it to meta.
the real limit is budget: each concept needs enough spend for meta to learn who it suits.
| monthly spend | prospecting ad sets | live concepts per ad set (our rule of thumb) | format mix | new concepts per fortnight |
|---|---|---|---|---|
| under ₹50,000 | 1 | 3 to 4 | at least 3 formats across the set | 1 to 2 |
| ₹50,000 to ₹1.5 lakh | 1 | 4 to 6 | video, static, carousel | 2 |
| ₹1.5 lakh to ₹5 lakh | 1 to 2 | 6 to 10 | all three, plus a talking head | 3 to 4 |
| ₹5 lakh to ₹15 lakh | 2 to 4, one per hard limit | 6 to 10 each | all three, per language | 4 to 6 |
| above ₹15 lakh | one per hard limit | 8 to 12 each | all three, per language and offer | 6 or more |
only the three-format floor traces to meta, as reported. the rest is our working rule. count a concept only if it would appeal to a different person for a different reason; a 9:16 recut is not one. our ad creative strategy guide has the checklist we use to tell the two apart.
for marut drones, creative and forms were state-specific across 13 states: diversification by language and context, not colour. ₹25 lakh of spend produced 23,420 leads and ₹8.7 crore in revenue, peaking at 36x roas (marut drones case study).
should you run an advantage+ leads campaign or a manual lead campaign?
start with an advantage+ leads campaign if your offer is broad, your lead form is short and you send quality signals back. go manual when you need hard age or gender limits, separate ad set budgets, or a test meta's automation would blur. meta reports lower lead costs with advantage+ on, but those are its own tests.
meta's page says that when a leads campaign runs with advantage+ on, which is the default for the leads objective in ads manager, advantage+ audience, placements and campaign budget are also switched on. manual controls remain available (meta). search engine land reported in february 2025 that changing the recommended settings can switch advantage+ off for that campaign (search engine land).
jon loomer has questioned whether lead quality holds up, since advantage+ audience doesn't let you set age or gender as a hard control (jon loomer). we have no independent dataset of our own on this split, so test it in your account.
| your situation | advantage+ leads | manual lead campaign |
|---|---|---|
| one offer, one language, broad buyer | start here | only as a test |
| buyer must be 25+ or one gender (a legal or product limit) | risky: age and gender are suggestions | better: set them as controls |
| you need fixed spend per state or project | use ad set spend limits sparingly | better: ad set budgets |
| you send qualified-lead stages back via conversions api | strong fit | works, with less automation |
| under about 50 leads a week in total | fits: pools everything in one place | avoid splitting further |
either way, optimise for a quality signal, not raw form fills. if buyers prefer to chat first, a click-to-whatsapp version of the same campaign is worth testing; our whatsapp marketing guide covers the follow-up side.
broad targeting in meta ads: broad vs interest vs lookalike in 2026
broad targeting in meta ads is the default for prospecting in 2026: set location, age floor and language as hard controls, then let advantage+ audience and your creative find people. interests and lookalikes still help, but mostly as suggestions that guide the system's first steps, not as walls it cannot cross.
meta's advantage+ audience page says that when you add audience suggestions, such as interests, custom audiences or lookalikes, the system prioritises people matching them before searching more widely (meta). location and minimum age stay firm.
| option | how meta treats it in 2026 | when it still earns its place |
|---|---|---|
| broad (location, age floor, advantage+ audience, no suggestions) | system finds people from the creative and your conversion data | default for prospecting with a working pixel or conversions api |
| interests as suggestions | a starting hint; delivery can go beyond it | new accounts with little conversion data |
| original audience with interests | hard limits, no expansion | narrow b2b buyers, or a clean test against broad |
| lookalike of buyers | a suggestion inside advantage+ audience | seeding from a crm list of real buyers, not form fills |
| custom audience (site visitors, leads) | retargeting or exclusion | retargeting, and excluding existing customers |
for one real estate developer, average time-to-lead was 18 days, which is why we keep a small retargeting campaign beside broad prospecting. for a d2c client, ai retargeting delivered 4.5x roas. retargeting earns its own campaign; interest slicing rarely does.
how to scale facebook ads without resetting the learning phase
to scale facebook ads safely, raise budgets in planned steps rather than jumps, batch creative changes, and never change the optimisation event on a winning ad set. meta says audience, creative and optimisation-event changes can restart learning, and budget or bid changes may too, depending on their size. big shifts belong in a new ad set.
meta's "last significant edit" definition covers pausing an ad set and changing its optimisation event, audience or creative. bid and budget changes may count, depending on their magnitude (meta business help). meta says results are less stable during learning (meta business help). the common "20% every few days" rule comes from practitioners, not meta.
| change | can it restart learning, per meta? | safer alternative |
|---|---|---|
| small budget increase | depends on size | raise in steps a few days apart (our habit: about 20%) |
| doubling or tripling budget overnight | likely, by meta's magnitude rule | step up over a week, or add a parallel campaign |
| new optimisation event (lead to qualified lead) | yes | launch a new ad set or campaign with the new event |
| audience edit | yes | duplicate into a new ad set; leave the winner alone |
| adding or editing creative | yes, creative changes count | add new concepts in one weekly batch, not daily |
| bid strategy or cost cap change | yes, or by size for bid amounts | test the new bid in a duplicate ad set |
| pausing an ad set | yes, a pause can restart learning | cut budget in small steps rather than pausing |
the other route is horizontal: a new ad set or campaign for a new state or language, with its own creative, as each marut drones state had.
cbo vs abo in 2026: when does each still win?
in the cbo vs abo choice, advantage campaign budget (cbo) wins for most 2026 accounts because it moves money to the ad set getting results without you editing each one. ad set budgets (abo) still win when you must guarantee spend per state or project, or run a clean test between two ad sets.
meta describes advantage campaign budget, formerly campaign budget optimisation, as one campaign-level budget that shifts across ad sets, with optional minimum and maximum spend limits per ad set (meta). its help centre says to judge results at campaign level, not per ad set (meta business help). jon loomer notes that meta tries to spend an ad set minimum but can't guarantee it (jon loomer).
| situation | advantage campaign budget (cbo) | ad set budgets (abo) |
|---|---|---|
| one goal, 1 to 4 ad sets, shared kpi | wins | unneeded control |
| advantage+ leads campaign | part of meta's advantage+ setup | can switch advantage+ off |
| fixed spend per state, project or dealer contract | possible with limits, but leaky | wins |
| clean test: two audiences or two offers | biased: money flows to the early leader | wins |
| retargeting with small audiences | small ad sets get starved | wins |
| scaling a proven campaign | wins: one budget to raise | more edits, more resets |
for a luxury retail client, a single meta campaign drove ₹2.1 crore in sales.
the account structure we would set up for an indian lead-gen advertiser
for an indian lead-gen advertiser, we would run one prospecting campaign on advantage campaign budget, one ad set per hard limit, six to ten distinct concepts per ad set, a small retargeting campaign on its own budget, and qualified-lead stages sent back through the conversions api. then scale in steps.
- campaign 1, prospecting: leads objective, advantage+ on, campaign budget, optimising for the deepest stage that still clears about 50 events a week.
- ad sets: one per hard limit, such as state, language, project or rera registration. nothing split by interest.
- targeting: location, age floor and language as controls; advantage+ audience on; customers and existing leads excluded.
- creative: six to ten concepts per ad set across video, static and carousel, refreshed in one weekly batch.
- campaign 2, retargeting: ad set budget, video viewers and form openers, objection-handling creative.
- lead capture: short instant forms or a landing page, routed to a crm within minutes.
- feedback: conversion leads stages sent back, so meta learns from buyers, not form fills.
illustrative example: two projects, ₹4 lakh a month
this is an illustrative example with made-up numbers, not client data. a pune developer has two projects and ₹4 lakh a month, about ₹92,000 a week. prospecting gets ₹80,000 a week; retargeting gets the rest.
| setup | leads a week at ₹500 | qualified at 30% | exits learning on qualified leads? |
|---|---|---|---|
| one ad set, both projects | 160 | 48 | borderline, just under 50 |
| two ad sets, 60:40 split | 96 and 64 | 29 and 19 | no, neither |
| two ad sets, optimising for leads | 96 and 64 | 29 and 19, tracked not optimised | yes, on leads |
the projects have different rera numbers and prices, so they need separate ad sets. optimise for leads first, feed quality stages back, and move to qualified leads as volume grows. for lodha, 45 flats were sold with 100% attributed to digital, at 3.8x the industry average conversion rate. structure only pays when sales follow-up keeps pace.
who wrote this, how and why
puneetraj chavan wrote this. he is the founder of rapchik media and has spent 8+ years running performance marketing and automation for 30+ brands across india and the us. sources were checked on 24 october 2026. platform-reported figures, including meta's, were not independently audited.
client results come from rapchik case studies, and the worked example uses illustrative numbers. we wrote this because most andromeda advice stops at "use broad targeting", and the expensive mistakes we see are structural.
- meta: engineering: andromeda (dec 2024) · advantage+ leads campaigns · advantage+ audience · advantage+ campaign budget
- meta business help: about the learning phase · last significant edit · about learning limited · about campaign budget optimisation
- trade press: search engine land on advantage+ leads · social media today on meta's creative tips · ppc land on andromeda · jon loomer on advantage+ leads · jon loomer on campaign budget
want your account restructured for andromeda? see our performance marketing services or talk to us.





