Lookalike Audiences for Healthcare Meta Ads in India (2026 Playbook)

Every year someone declares lookalike audiences dead, and every year they keep working — just differently. The 2020 playbook of "1% lookalike of everyone who ever gave you their number" no longer works because the seed is too noisy and Meta will happily find you thousands more of the wrong noise. The 2026 playbook is quieter and more surgical. This piece is the way ICG uses lookalikes across 23+ healthcare Meta ad accounts on ₹9.14Cr of monthly optimised spend — what seeds actually earn a lookalike, which percentages still convert in Indian healthcare, when to layer, when to skip lookalikes entirely, and the specific mistakes we most often inherit when we take over an account.

What a lookalike actually does (and does not)

A lookalike is Meta's attempt to find people who match a source list on the behavioural signals it can observe. The lookalike is only as good as the source. A source of 200 verified consulted patients will produce a materially better lookalike than a source of 20,000 raw form fills where 80% never picked up the follow-up call. Lookalikes do not read patient intent, do not know about your city catchment, and do not compensate for weak creative. They give Meta a shape to search against, nothing more.

The seed hierarchy for healthcare in India

Not all sources are equal. Ranked from most to least useful for a lookalike in a healthcare Meta ad account:

• Consulted patients (in-clinic attendance confirmed) — top-tier seed
• Paid patients (procedure completed) — best for premium-price positioning
• High-LTV segments (multi-service or repeat) — for scaling upmarket
• Qualified leads (spoke on phone, matched fit criteria) — solid mid-tier seed
• Website purchase or booking events — good if your booking funnel is clean
• Raw form fills — usually too noisy in 2026
• Website visitors — too broad to seed reliably

The first four are the seeds we still use every day. The last three we mostly do not. If your CRM cannot separate consulted from raw leads, that is the first fix before any lookalike work.

Minimum seed size — the real number

Meta's documented minimum is 100 people in the source. In practice, for a stable Indian healthcare lookalike, we want at least 500 in the seed. Below that, the lookalike drifts week to week and CPL bounces around unpredictably. If you have 220 consulted patients, wait until you have 500 or accept that the lookalike will be volatile. For long-cycle specialties like IVF where consult volume is smaller, we sometimes seed with 300 and rebuild the lookalike monthly to keep it fresh.

Which lookalike percentages still work in 2026

The old ladder — 1%, 2%, 3%, 5% — has narrowed. What we see across the portfolio: 1% and 2% lookalikes still meaningfully outperform generic broad in tier-2 and tier-3 cities. 3% and 5% have largely converged with broad Advantage+ audience in metros, so we rarely run them as standalone anymore. Instead we use them as seeding layers inside broad campaigns — telling Meta "here is the direction to search" without hard-restricting the audience.

Meta Catalyst IQ Audience Size vs Efficiency — broad vs narrow verdict per account, portfolio distribution by audience size" loading="lazy" decoding="async" style="width:100%;height:auto;display:block;">

Lookalike refresh cadence

Rebuild lookalikes monthly for high-cycle specialties (aesthetic, dental, derm) and every 60 days for longer-cycle (IVF, oncology, cardiac). Stale lookalikes are a slow leak — Meta updates the algorithm continuously, your consulted-patient list grows, and the lookalike from three months ago is optimising against a seed you no longer have. Set a calendar reminder and treat it as a hygiene task.

Stacked lookalikes vs single lookalikes

Stacking multiple lookalikes (say 1% of consulted + 1% of paid + 1% of high-LTV) into one ad set works when the source lists have low overlap. When the overlap is high — the same people appear in all three — stacking adds noise without adding scale. Check overlap in Meta's Audience Insights before stacking. Overlap above 30% means treat them as one seed. Below 30%, stacking is safe and usually adds 15–25% audience reach without CPL degradation.

The value-based lookalike — worth the effort

Value-based lookalikes take a source list with an LTV or value column, and Meta weights the lookalike toward the higher-value members of your seed. For specialties with price bands — aesthetic packages, dental aligners, IVF workups with add-ons — value-based lookalikes consistently pull higher AOV leads. The setup requires you to pass a numeric value with each customer record, either via file upload or via CAPI. The lift is real; the operational overhead is moderate. Worth it for accounts spending over ₹5 lakh/month.

Layering lookalikes with interest targeting

Do not layer a 1% lookalike with a five-interest stack. The audience will shrink below Meta's optimisation floor and CPL will climb. If you must combine, use one lookalike + one broad interest layer, no more. In practice, we mostly run lookalikes and interests as separate ad sets in the same campaign and let the auction distribute budget. That gives cleaner learning per ad set and keeps the CPL comparison honest.

Lookalike plus geography — the tier-2 pattern

The single most consistent lookalike pattern in the ICG portfolio is: 1% lookalike of consulted patients + city-level geography + no interest layer. In tier-2 cities where addressable audiences are small, this stack lets Meta find the shape of your ideal patient inside the city catchment without over-fragmenting. It is the default we start with when we take over any new tier-2 account, and it usually beats whatever narrow interest stack was running before.

Mistakes we most often inherit

Five recurring lookalike mistakes we clean up in new accounts:

• Lookalike from a raw form-fill list with 22% consult-conversion — noise in, noise out
• Ten different lookalikes across ten ad sets, all fragmented under learning phase
• Lookalike source last rebuilt 8 months ago
• Value-based lookalike attempted with no value column actually passed
• Lookalike layered with three interests, audience shrunk to 40,000, CPL doubled

Each of these is a Money Wastage Cleanup finding. Individually they cost ₹10–50K/month; together they can quietly consume 20% of a mid-sized account's spend.

Lookalikes get better when Meta CAPI is clean

Every point of EMQ (Event Match Quality) improvement in your CAPI setup makes your lookalikes better, because Meta has better signal about who your source actually is. If your CAPI is running at EMQ 4.5 and you fix it to 7.5, expect a measurable improvement in lookalike performance over the next 30 days — Meta is rebuilding the shape with cleaner data. This is why our CAPI and lookalike work is always sequenced together in the first 30 days of a new engagement.

When to skip lookalikes entirely

Two situations where we do not use lookalikes at all. First, when the account is under ₹1.5 lakh/month spend — the audience layer adds complexity without proportional lift and simple broad + interest beats it. Second, when the seed is genuinely bad and cannot be cleaned in the near term — a lookalike built on noise will underperform even generic broad. In both cases, get the account healthy on broad + creative first, then layer lookalikes in month two or three.

Lookalike patterns by specialty in the ICG portfolio

Not every specialty rewards lookalikes the same way. Patterns we see across 23+ healthcare accounts:

• IVF — lookalikes of consulted patients outperform broad in most tier-2 cities; in metros the gap narrows to under 10%
• Dermatology — lookalikes of paid customers work well for premium clinics; primary derm often runs equally well on broad
• Dental primary — lookalikes provide modest lift; geo + broad is often equally efficient
• Dental aligners — value-based lookalikes of completed cases materially outperform, worth the effort
• Hair transplant — 1% lookalikes of paid customers are near-mandatory for premium positioning
• Aesthetic — value-based lookalikes segment high-value from single-service customers cleanly
• Hospital cardiac — lookalikes rarely outperform interest-based targeting; the audience shape is too heterogeneous
• Oncology — lookalikes of second-opinion enquirers work; lookalikes of general leads do not

The seed hygiene checklist we run before every lookalike build

Before building any lookalike, the seed list gets a hygiene pass. Deduplicate by phone and email; strip test entries; strip employees and their household contacts; strip patients who have unsubscribed under DPDP consent withdrawal; strip patients whose consult conversion is documented as "not a fit" by the sales team. On typical CRMs this removes 8–18% of records — small change, disproportionate impact on lookalike quality. A dirty seed teaches Meta the wrong pattern, and the wrong pattern is expensive to unlearn once it is trained.

Lookalikes and DPDP consent — do not skip this

Uploading a customer list to Meta to build a lookalike is a data-sharing act under DPDP. Patients on that list must have consented to marketing use of their data, or the upload is non-compliant. Practically: your consent capture at the lead stage must include a specific marketing-data-sharing consent, and patients who withdrew that consent must be removed from the seed before upload. Beacon's consent registry drives this automatically inside ICG-managed accounts. If your CRM does not track consent state, that is the first fix before any lookalike upload.

Powered by Meta Catalyst IQ — the decision engine behind every Meta ad ICG runs

ICG built Meta Catalyst IQ because most Indian healthcare brands running Meta ads waste 30–50% of budget without knowing it. It's the diagnosis + decision layer above Ads Manager — Hygiene Factors 12-point checklist, Naming Intelligence (surfaces conflicts costing ₹50K–₹2L/account/month), Creative Scoring Matrix (Core Performer / Scalable / Getting Started / Review), 2-Day Comparative, SLC Framework, Money Wastage column in ₹.

• Master Dashboard — 23+ accounts, ₹9.1Cr+ spend/mo optimised, ₹1,581 blended CPL vs ~₹3,200 market benchmark.
• Diagnose → Optimise → Grow — daily hygiene checks, weekly creative scoring, monthly money wastage cleanup.
• CPQL Engine — cost per qualified lead (not just cost per lead) at ad-set level. Try the interactive CPQL calculator.
• Portfolio benchmarks — IVF ₹632, derm ₹520–1,180, dental ₹620–1,800, aesthetic ₹400–900, hospital cardiac ₹3,200.

Included free with every ICG Meta ads or Performance Marketing engagement (Starter ₹20,000/-/month tier and above). Not sold standalone. Book a free 48-hour Meta ad diagnostic or WhatsApp us.

Frequently asked questions

What is the minimum useful seed size for a healthcare lookalike?

Meta's documented floor is 100 people, but stable performance requires at least 500. Below that, week-to-week volatility eats the benefit.

Should we use 1%, 2%, or 3% lookalike?

1% and 2% still outperform in tier-2 and tier-3 cities. In metros, 3% and higher have converged with broad Advantage+ audience and are better used as seeding layers than standalone.

How often should we rebuild a lookalike?

Monthly for high-cycle specialties (aesthetic, dental, derm), every 60 days for longer-cycle (IVF, oncology, cardiac). Stale lookalikes are a slow leak.

Do value-based lookalikes work in India?

Yes, especially for specialties with price bands — aesthetic packages, dental aligners, IVF workups. Requires you to pass a numeric value column with each customer record.

Can we stack multiple lookalikes in one ad set?

Yes, if seed overlap is below 30%. Above 30%, stacking adds noise without adding reach. Check overlap in Audience Insights before stacking.

Why has my lookalike stopped working since last quarter?

Three usual causes: stale seed (rebuild), poor EMQ score on your CAPI (fix signal quality), or fragmentation into too many ad sets (consolidate).

Related reading

• Broad vs narrow targeting for healthcare Meta ads
• Meta CAPI and server-side events for healthcare
• EMQ score optimisation guide
• The 5-zone creative testing framework
• DPDP consent form fields for healthcare Meta lead ads

Authority reading: Meta's guidance on building lookalike audiences.