A lookalike audience is a modelled segment of users who resemble your first-party seed audience — purchasers, high-value customers, app converters — based on shared behavioural and demographic signals. In CTV advertising, lookalikes allow brands to extend reach beyond their known customer base to prospects who match the same profile, without requiring the prospect to already be in the brand's CRM. For India CTV, lookalike modelling is the primary mechanism for brands whose first-party audience is too small to run as a direct targeting segment at meaningful CTV scale.
How lookalike audiences are built for CTV
The lookalike construction process has three steps:
- Seed audience upload: The brand uploads a hashed identifier list (SHA-256 email, phone, or device ID) representing the seed — typically existing customers or converters. The DSP or publisher matches these identifiers against its own user graph to find the matched seed within its platform.
- Signal extraction: The matched seed users' signals within the platform's graph are profiled — content consumption patterns, device type, geographic data, engagement patterns, inferred demographic attributes. The model learns what distinguishes seed users from the general population.
- Lookalike expansion: The model scores the full platform user base and creates an expansion audience of users with the highest similarity score to the seed. The brand controls the expansion size — a narrow lookalike (top 1%) is more precise, a broad lookalike (top 10%) is larger but less similar.
Lookalike modelling in India CTV
Each platform handles lookalike modelling differently for India CTV:
- DV360 Similar Audiences: Built from Google's identity graph including Search, Maps, YouTube, and Play data. For CTV, the model can identify users who match the seed's profile and are active on CTV-accessible inventory. The DV360 lookalike for CTV has higher signal quality for metro India consumers (who are more active on Google services) than for tier-2/3 users.
- The Trade Desk Predictive Clearing: TTD builds lookalikes from its own identity graph supplemented by third-party data. India signal quality for CTV lookalikes is improving but thinner than DV360 in the India market due to Google's stronger data position locally.
- JioHotstar publisher lookalikes: For direct-sold campaigns, JioHotstar can build lookalike segments within its registered user base (80M+ users) based on content consumption, subscription tier, and demographic signals. These publisher-side lookalikes tend to have stronger India CTV signal than DSP-side models because they are built on actual CTV viewing data. Available as a direct deal, not open programmatic.
- Publisher lookalikes via data partnerships: SonyLIV and Zee5 offer similar publisher-side lookalike capabilities for direct campaigns. The practicality depends on the seed match rate into the publisher's user base.
When lookalikes work — and when they don't
Lookalike audiences perform well when:
- The seed audience is large enough (500+ matched users minimum; 5,000+ for reliable modelling) and behaviourally distinct from the general population
- The seed represents a clear conversion signal (purchasers, subscribers) rather than a broad behavioural signal (visitors, app installs)
- The campaign goal is reach and awareness among new audiences, not conversion optimisation of existing prospects
Lookalikes underperform when:
- The seed is too small (under 500 matched users) — the model has insufficient signal to distinguish seed characteristics
- The seed is not representative — using "everyone who visited the website" rather than "customers who purchased" creates a noisy seed with low signal quality
- The brand category is too broad — in FMCG, a purchaser lookalike may effectively model "everyone" because the product has mass penetration. Lookalikes work best for categories with distinctive purchaser profiles.
- The campaign is trying to reach very specific micro-audiences — lookalike expansion is an approximation and will include users who don't match the true target
Quality signals for India CTV lookalikes
Evaluate the quality of a lookalike audience before scaling spend:
- Seed match rate: If fewer than 30% of your uploaded seed matches in the DSP/publisher graph, the model is working from a thin foundation. Improve match rates by using phone numbers (higher India coverage than email) or device IDs alongside email.
- Audience overlap check: Run an overlap analysis between the lookalike and your known customer audience — high overlap (>20%) suggests the model is not expanding meaningfully beyond the seed.
- Hold-out test: Run the lookalike against a matched unexposed holdout group. If conversion rates in the exposed lookalike group are not statistically higher than the holdout, the model is not identifying genuinely similar users.
- VCR and engagement signals: A lookalike audience that delivers lower VCR than run-of-publisher benchmarks may indicate poor audience quality — users who don't engage with the brand's content even when reached.