Lookalike Audience Seeds

Lookalike Seeds

Select, size, and validate the seed customer list that a lookalike, similar, or value-based audience will be built from. Three principles govern every step below:

• Match count, not row count - every uploaded list shrinks by its match rate before it reaches the platform's minimum. A 40% match on 5,000 rows is a 2,000-person seed, and the platform floor applies to that matched number - upload 2-3x the target.
• Homogeneity and value concentration beat size - Meta's own guidance holds that a seed's homogeneity affects audience effectiveness more than its size. Practitioner consensus (Stackmatix, Grow With Sakib - directional, not audited) is that a few hundred high-value customers outperform thousands of undifferentiated subscribers.
• Inclusion is a suggestion, exclusion is a rule - on Meta (Advantage+) and Google (Demand Gen, where lookalike reach tiers now function as signals rather than hard segments), the delivery algorithm may override an inclusion audience. Only exclusion/suppression audiences remain hard rules.

The seed's job is feeding the algorithm the cleanest signal, not fencing an audience - which raises the quality bar rather than removing it.

This skill ends at a specified, validated seed definition plus its export/handoff spec. It never walks through creating or uploading the audience inside any ad platform's interface.

Where each adjacent concern lives:

• Overall targeting plan (interest, behavioural, lookalike, and retargeting tier layering) - seed mechanics live here, tier architecture lives in mbfinotti/advertising-skills@ad-audience-targeting.
• Retargeting sequence design, audience windows per funnel stage, and frequency caps - mbfinotti/advertising-skills@retargeting-funnel.
• B2B buying-committee and persona mapping - mbfinotti/advertising-skills@ad-buyer-group-mapper.
• Root-causing broad account underperformance - mbfinotti/advertising-skills@ad-account-diagnostic.
• Pixel/CAPI instrumentation feeding engagement-based seed sources - mbfinotti/advertising-skills@ad-conversion-tracking.
• Reconciling conflicting performance reads on a seed across attribution sources - mbfinotti/advertising-skills@ad-attribution-gap.

Interview

Ask before selecting anything. One question per message; offer multiple-choice answers where possible; skip anything already answered or visible in supplied data.

• B2B, B2C/ecommerce, or both?
• Which platform(s) will consume the seed? (Meta / Google / LinkedIn / TikTok / other - floors and match mechanics differ)
• What customer data exists, and where does it live? (CRM / data warehouse / email platform / pixel-and-platform events only / almost none)
• Roughly how many total customers, and which identifiers exist per row? (email / phone / name / postal address / mobile ad ID - more identifier types per row lift match rate)
• Is there a usable value column - and what does it actually measure: cumulative lifetime revenue, margin, predicted LTV, or just order count? Provenance matters more than presence.
• How recent is the data - what share of the list purchased or converted within the last 90 and 180 days?
• Does the list contain EEA/UK residents or US state-privacy-covered consumers? Is there a documented consent or lawful basis for ad-platform upload, and are opt-outs, deletions, and do-not-contact records enforced before export?
• How will the list reach the platform - manual CSV or automated warehouse/CDP sync? Who owns the refresh?
• What acquisition outcome is the lookalike meant to drive - target new-customer CPA/CAC, and any LTV or margin expectation for acquired customers?
• What does the account currently average on those same two numbers - acquired-customer CPA, and acquired-customer cohort LTV at 90 days? The pass threshold grades the seed against this baseline, so an unknown baseline means the seed cannot be graded at all: say so and set the first cohort checkpoint as the baseline-building run.
• By what date must the audience be live and spending, and is that date hard? (Seed sources span minutes to a week of coordination, and the enrichment rung runs through procurement - the orderings in steps 3 and 6 cannot be picked without this.)
• Do you want the audience that can launch soonest, or a compounding asset - an automated warehouse sync, enriched identifiers, a reusable cohort definition - that keeps paying across every future audience?
• Effort ceiling: how many hours of data-team time are available, who owns the refresh afterwards, and can you actually get privacy sign-off and a vendor DPA? (A "no" deletes rungs rather than delaying them.)

Workflow

Every ranking in this skill - steps 3, 6 and 9 here, and the ones in the references - is a default, not a law. Each assumes the common case: no enrichment vendor under contract, no in-house identity graph, no analyst free this week. Who executes it changes the ordering as much as what it is.

Re-rank first against the interview's deadline, durability, and effort-ceiling answers, naming which answer moved which rung:

• A hard near-term date promotes the platform-native engagement seeds.
• A compounding-asset mandate promotes the warehouse sync and the enrichment rung despite their setup.
• No route to privacy sign-off deletes every list-based rung outright.

Then re-rank against what this user actually owns - any of these collapses a rung's effort to near-zero and moves it up:

• A signed enrichment DPA.
• Identity resolution already built in the warehouse.
• An analyst who can write the cohort SQL today.
• A data team that owns the export job anyway.

1. Run the Interview; collect every answer before selecting a single row.
2. Run the privacy and consent gate - a hard stop, before any selection work. If any item below is unconfirmed, stop and say exactly what is missing. This skill is not legal advice: lawful basis varies by jurisdiction, so route open questions to the user's counsel.

• Documented lawful basis or consent for uploading this list to an ad platform. Under GDPR, the Bavarian DPA position upheld by the Higher Administrative Court Munich (2018, Ref. 5 CS 18.1157) holds that customer-list custom audiences require prior consent, and hashing does not change that: SHA-256 output is pseudonymous, not anonymous, so consent obligations survive it.
• For EEA use of Google Customer Match, consent signals passed per Consent Mode v2 with both ConsentStatus fields GRANTED (mandatory since March 2024).
• CCPA/CPRA "sharing" opt-outs and Global Privacy Control signals (honored in twelve US states) propagated before export.
• No seed built on sensitive categories: health, sexual orientation, religion, race, political affiliation, criminal history.
• Platform customer-list terms accepted for the specific ad account (value-based audiences need a separate terms acceptance on Meta).
• Suppression, opt-outs, and deletions enforced once, in the warehouse, before any sync - not per-platform afterwards.

1. Choose the seed source: purchase data first, always. If purchase data is too thin, pick among the eight sourced alternatives - but never down the value ranking alone, because effort and compliance cost both run backwards to it. The axes disagree, so read all four:

• value: purchasers > checkout/add-to-cart initiators > high-intent page viewers > lead-form submitters > email clickers > video viewers > page engagers > all visitors