Audience Testing Framework: Systematic Approach to Finding Winners

Audience discovery begins by separating new-customer acquisition from re-engagement. A recent visitor or past buyer is not an equivalent benchmark for a person who has never interacted with the brand. Compare strategies within the same acquisition objective, then evaluate their customer contribution and limits.

Separate warm campaigns from the prospecting framework

Lookalike or other suggested prospecting audiences can be hypotheses, but a source of buyers is not the same as delivering only to new buyers. Record the accepted configuration and its limits.

An exact prospecting allocation, not expected win rates

This allocation is a cash planning example, not a recommendation supported by observed win rates. Warm re-engagement has a separate budget. The $500 exploratory cell may be too small for a purchase comparison; calculate feasible evidence before launch.

Write a targeting integrity record

• Record inclusion and exclusion lists, geographic controls, accepted age settings, objective, optimization event and actual expansion/suggestion behavior. Distinguish a firm control from an audience suggestion.
• Verify that the same eligible customer-status definition is used in all acquisition strategies. Reconcile lag in customer-list updates and match failures.
• Check whether audiences overlap. If the experiment mechanism supports mutually exclusive randomized eligible groups, document that assignment. Separate ordinary ad sets or identical budgets do not themselves randomize people.
• Record placement, offer, creative and simultaneous enrollment conditions. Inspect delivered audience summaries and spending; do not assume a campaign shares learning in a way that removes confounding.
• If expansion makes intended cells indistinguishable, interpret the result as a comparison of available delivery strategies rather than a test of isolated named interests.

Meta’s SDK schema exposes targeting, custom-audience exclusions and automation fields. That is evidence of field definitions, not universal eligibility or an available UI setting. Use the audience overlap diagnosis when interpreting actual configuration and delivery.

Replace a mixed metric score with contribution

A has lower acquisition cost ($25 versus about $33.33), but B has higher contribution under these assumptions. Include refunds, servicing, fulfillment and qualification in real variable costs. B’s higher average order value does not establish higher lifetime value; compare equally mature future retention and contribution cohorts before extending the conclusion.

A reviewable test brief

Illustrative brief: compare two available prospecting delivery strategies for new retained customers in the same region, with one approved message and a fixed outcome window. Prespecify the allocation mechanism, primary contribution measure, cash cap, exclusion refresh and uncertainty method. For a binary retained-customer indicator use an appropriate sample plan; contribution per assigned person needs a variance-aware analysis. No threefold acceleration or universal broad-targeting lift follows from this framework.

Prepare the same audience hypotheses and ad message as NovaStorm campaign drafts for review. Record exclusions and experiment assignment in the tools that actually implement them. Draft preparation is not automatic overlap removal or a causal measurement service.

Source: Meta targeting SDK schema. The allocation and cohort arithmetic are explicitly hypothetical.