If you learned Meta Ads between 2016 and 2020, most of what made you good at it is now actively counterproductive. Detailed interest stacking, tight ad set segmentation, manual bid caps, one ad per ad set — these were the right answers for a platform that no longer exists.
Here is what actually drives performance on Meta today, based on what we run for clients across e-commerce, real estate, healthcare, and education in India.
Change 1: Consolidation Beats Segmentation
The old instinct was to split budget across many narrowly-defined ad sets so you could see which audience performed. The problem is that Meta's optimisation needs roughly 50 conversions per ad set per week to exit the learning phase and stabilise. Splitting ₹1,00,000 across eight ad sets means eight ad sets permanently stuck in learning, each making decisions on noise.
The current structure that works is deliberately boring:
- One campaign per objective, not per product line.
- One or two ad sets with broad or Advantage+ targeting.
- Six to twelve creatives inside each ad set, genuinely different from each other.
- Campaign budget at the campaign level, letting Meta distribute it.
You lose granular audience reporting. You gain performance. Most advertisers who resist consolidation are protecting a reporting habit rather than a result.
Change 2: Signal Quality Is the Whole Game
Browser-based pixel tracking now misses a substantial share of conversions due to iOS restrictions, ad blockers, and browser tracking prevention. Every missed conversion is a lesson the algorithm never learns.
The Conversions API — server-side event reporting — is no longer an advanced setup for large accounts. It is the baseline. Accounts running CAPI alongside the pixel with good event deduplication typically recover a meaningful share of otherwise-invisible conversions, and the optimisation improvement compounds from there.
Two things matter as much as installing it:
- 1Optimise for the right event. If your business sells over the phone, a form submission is a weak proxy. Send back a "qualified lead" event from your CRM once someone actually answers and fits. The algorithm will then go find more of those people instead of more form-fillers.
- 2Pass strong customer parameters. Hashed email, phone, city, and external ID materially improve match quality — and in India, phone number is usually the strongest identifier you have.
Change 3: Creative Is the Targeting
With broad targeting, the creative itself decides who sees the ad. A video that opens with a shot of a construction site self-selects an audience just as effectively as any interest layer used to. This is the mental shift that separates advertisers doing well from advertisers complaining that broad targeting doesn't work.
What performs right now
- Native-looking vertical video. Ads that look like an ad get scrolled past; ads that look like content get watched.
- The first three seconds carrying the entire hook. Not a logo, not a slow pan — the problem or the payoff, immediately.
- Captions burned into the video. A large share of feed viewing is muted.
- User-generated and testimonial formats, which consistently outperform polished brand films on cost per lead.
- Static images that look screenshot-like — a WhatsApp conversation, a review, a before-and-after — which cut through precisely because they are visually plain.
- Language matching. For Tier 2 and Tier 3 audiences in India, Hindi or regional-language creative regularly halves cost per lead versus the English version.
Stop asking which audience is working. Ask which creative is working, because on Meta in 2025 they are the same question.
— Our standard answer in performance reviews
Testing discipline
Test concepts, not colours. Changing a button shade produces no learning. Changing the core promise, the format, or the person on screen produces learning you can reuse. Give each new batch at least three to four days and enough spend to reach statistical relevance before killing anything.
Change 4: Advantage+ Is Genuinely Good Now
Advantage+ Shopping campaigns for e-commerce and Advantage+ lead campaigns have moved from unreliable to default-recommended for most accounts with sufficient volume. They hand almost everything to automation — audience, placement, budget distribution, and increasingly creative combination.
Where they still fail: accounts with fewer than about 30 conversions a week, businesses with narrow eligibility requirements the model has no way to know about, and any account where the conversion event is misconfigured. Automation applied to a bad signal simply produces bad results faster.
Change 5: Measure Outside the Platform
Meta's reported ROAS is a modelled number that flatters itself. It is useful for relative comparisons between your own campaigns, and misleading as an absolute truth.
The numbers we hold clients accountable to:
- Blended cost per qualified lead — total ad spend divided by leads the sales team confirmed as real.
- Cost per closed deal, tracked back to source in the CRM.
- Incremental revenue, ideally checked with occasional geo holdout tests.
- Contribution margin after ad spend, which is the only figure that tells you whether to scale.
A campaign showing 6x ROAS in Ads Manager and 1.8x in the CRM is not a good campaign. It is a reporting artefact, and scaling it is how ad budgets get burned.
A Structure That Works
- 1Prospecting: one Advantage+ or broad campaign, 70–80% of budget, 8–12 creatives, optimised for your strongest down-funnel event.
- 2Retargeting: one lean campaign for site visitors and engagers in the last 14–30 days, 15–20% of budget, with offer-led and objection-handling creative.
- 3Testing: a small always-on campaign at 5–10% of budget running new concepts, with winners promoted into prospecting.
Refresh creative before performance drops, not after. In high-frequency Indian markets, most winning creatives start to fatigue in three to six weeks. If your best ad has been running unchanged since December, your next month is already going to be worse.
The Summary
Consolidate campaigns. Fix your signal with the Conversions API and a down-funnel event. Ship more creative than feels necessary, in the language your customer actually thinks in. Let the algorithm target. Then measure the result somewhere other than Ads Manager.
Everything else is a detail.
