Saved Audience
A saved audience is a reusable set of targeting criteria, such as location, age, interests, and behaviors, that you define once and apply across multiple ad campaigns.
Key takeaways
- A saved audience stores conditions, not members, and resolves fresh at campaign runtime.
- Its population drifts as the platform reclassifies users, even with the definition unchanged.
- Each narrowing layer multiplies the pool down and can starve delivery optimisation.
- Expansion settings let platforms deliver outside your criteria, making them hints not boundaries.
- Inferred interests and self-reported job data are unverifiable proxies you cannot audit.
In depth
A saved audience is a stored query rather than a stored list. It holds a set of conditions expressed in the platform's own taxonomy, locations, age bands, languages, job attributes and inferred interest categories, joined by narrowing logic. Nothing is resolved until a campaign runs, at which point the platform evaluates the conditions against its current classification of its users. The definition you wrote last quarter can therefore return a different population today, because people move in and out of inferred categories without you changing a single field.
Each narrowing layer you add multiplies the audience down, and the effect compounds faster than most people expect, so three modest conditions can leave a pool too small for the delivery system to optimise within. That is the central trade-off: precision at the definition level competes with the algorithm's need for room to learn. Expansion settings complicate it further, since many platforms will deliver outside your criteria when they predict better results, which means a saved audience is a strong hint rather than a hard boundary.
Used well, a saved audience is the constant in an experiment. Naming and versioning them lets a team hold targeting fixed while creative or offer varies, and makes results comparable across campaigns and across people. Recording what each criterion is a proxy for keeps the definition auditable later. A scorecard quiz supplies the evidence to revise it, since the declared roles, company sizes and answers of high-scoring respondents show which of your assumed criteria actually track fit and which were guesses.
The inferred categories underneath are proxies of unknown accuracy that you cannot inspect or verify, and the platform can retire or redefine them without warning, silently changing what your saved audience means. Professional attributes rest on self-reported profiles that people update rarely, so job-title targeting skews toward whoever last edited their profile. A saved audience also cannot express anything about behaviour on your own properties; the moment your criterion involves what someone did on your site, you need a different audience type.
Example in practice
How to measure it
Watch the estimated audience size alongside delivery. A definition that keeps shrinking between campaigns signals taxonomy drift rather than market change, and a pool small enough to hit high frequency quickly will drive up costs regardless of creative quality. Compare cost per qualified lead against a deliberately broader version of the same definition; if the narrow one does not win, the extra conditions are costing reach without buying quality.
Then check the criteria against outcomes rather than against intuition. Take recently won customers, look up the attributes the saved audience filters on, and count how many they would have satisfied. A low share means the definition is excluding the people you actually sell to. Quiz responses give a faster version of the same read, because respondents declare role and company size before any deal exists.
Common mistakes
Teams commonly stack every plausible criterion into one definition, adding interests, behaviours, job functions and age bands until the audience is technically precise and practically undeliverable. Costs rise, learning stalls, and the campaign never gathers enough conversions to optimise. Start with location, language and one meaningful attribute, then add a layer only after a test shows the broader version underperforms. Precision that starves delivery is not precision, it is a self-inflicted reach problem.
The second failure is treating a saved audience as documentation of who you target and never reconciling it with who actually converts. The definition persists across years, gets copied into new campaigns, and nobody checks whether the interest categories inside it still exist. Schedule a review that compares the criteria against the attributes of recently won customers, and delete any condition nobody can explain the reasoning behind.
Frequently asked questions
When should I use a saved audience?
Use one when you repeatedly target the same well-defined market segment and want to launch campaigns quickly without rebuilding filters. It is ideal for prospecting top-of-funnel audiences with a consistent, reusable definition.
What is the difference between a saved audience and a custom audience?
A saved audience is a stored set of targeting conditions using the platform's own categories, and it resolves to different people each time it runs. A custom audience is a list of specific individuals built from your data or their engagement with you. One describes a type of person to find; the other names people you already know. They are usually used together, one prospecting and one retargeting.
When should I use a saved audience instead of broad targeting?
Use one when a hard constraint genuinely applies: a language you can serve, a country you ship to, a regulated profession. Those conditions protect budget from waste that no algorithm can learn away. For softer preferences such as interests, broad targeting with strong creative often performs as well, because the delivery system infers fit from response faster than a static definition can express it.
How do I keep a saved audience accurate over time?
Review it on a fixed cadence against the attributes of customers you recently won, and check that every interest or behaviour category it references still exists in the platform. Version the definition rather than editing it in place, so performance before and after a change stays comparable. Record the reasoning for each criterion, because an unexplained condition is the first thing to become obsolete.
Why does my saved audience size keep changing?
Because it is a query, not a list. The platform recalculates who matches every time, and its own classification of users shifts as people change behaviour, update profiles, or get moved between inferred categories. Estimated sizes also reflect recent activity, so seasonal patterns move the number. A large sudden drop usually means a category you reference was retired or redefined.
How many saved audiences should a team maintain?
Keep one per genuinely distinct targeting decision, typically per market or per segment you would message differently. Beyond that, duplicates accumulate, get copied into new campaigns, and nobody knows which is canonical. A shared naming convention that encodes market, segment and version does more for consistency than adding further definitions to the library.
Do saved audiences work for B2B targeting?
They work where the platform holds reliable professional attributes, which is uneven. Job title, function and company size come from self-reported profiles that people update rarely, so targeting on them skews toward recently active profile editors. For narrow business segments, a custom audience built from your own list, or a lookalike seeded from it, usually reaches the right people more reliably.