“Build me custom audiences” is incomplete. The useful questions are: which seeds are healthy, what overlaps, what is stale, and what should be exclusion.
Portfolio thinking
- Hot / warm / exclusion tiers with different frequency expectations
- Seed size and last-used staleness as health reasons
- Overlap so two “unique” audiences do not bid against each other
- Custom audiences list from synced Meta data where available
Evidence or silence
Weather-linked audience ideas need enough distinct sample days. Stock-aware ideas need inventory truth. If the sample is thin, the honest output is “not enough signal”, not a confident million-person fantasy.
Questions operators actually ask
Should I rebuild lookalikes monthly? Only if seed quality and volume justify it. Stale large audiences with high overlap are expensive comfort blankets.
Exclusion hygiene? Purchasers and recent converters belong in exclusion logic more often than people admit. Template libraries exist so you stop rebuilding the same exclusions under new names.
Questions people actually ask
Does Ralph auto-publish dozens of live custom audiences without flags?
Audience Studio refresh, V2 layers, and writes sit behind flags and routes. Treat live mutation as gated, not always-on for every tenant.
What are [R] audiences?
Ralph-managed template names (for example ATC 7d, IC 7d, exclusions) so you can see what the system owns versus hand-built clutter.
Do you invent audience sizes?
Fabricated size estimates get stripped in composed proposals. Thin weather samples do not ship as strong truth.
Are Google RSA launches stuffed with Meta audiences automatically?
In the Google Ad Studio mutate path, audience is observation-only. It is not silently injected into RSA publish.
How to build and audit a Meta audience portfolio for Shopify
Audiences are a portfolio, and portfolios need auditing rather than expanding. Here is the build, then the maintenance loop that keeps it honest.
- Check the plumbing before you build anything. In Events Manager, confirm your pixel and Conversions API are both firing Purchase, AddToCart and ViewContent, and check the event match quality score. Audiences built on a leaky signal are small, stale and expensive, and no amount of clever segmentation fixes that.
- Build the small set that actually earns its keep. Most stores need six, not sixty: purchasers (180 days), added to cart but did not purchase (30 days), viewed product but did not add (30 days), all site visitors (30 and 180 days), video viewers at 50 percent or more, and Instagram or Facebook engagers (365 days). Build these in Audiences → Create audience → Custom audience and name them so the retention window is in the name.
- Seed lookalikes from value, not from volume. A lookalike is only as good as its seed. Use purchasers, and where you have enough data use a value-based seed from your highest-value customers rather than everyone who ever bought. Start at 1 percent for the seed test. Broad seeds produce broad lookalikes, which is just prospecting with extra steps.
- Run the overlap tool before you trust your segmentation. In Audiences, tick two to five audiences, then Actions → Show audience overlap. Anything above roughly 30 percent overlap is two ad sets bidding for the same people with your money on both sides. Either merge them or make the exclusions explicit.
- Set exclusions deliberately on every prospecting ad set. Recent purchasers almost always belong in the exclusion on prospecting. Existing site visitors usually do too, or prospecting quietly becomes retargeting with a worse creative. Write the exclusion rule once per campaign type and apply it consistently rather than deciding ad set by ad set.
- Audit for staleness on a fixed schedule. Once a month, sort audiences by size and by last used. A large audience nobody has used in three months is not an asset. A shrinking audience means the underlying signal is degrading, usually tracking rather than demand. Delete what you will never use so the list stays readable.
- Judge changes on incremental results, not audience size. Bigger audiences look reassuring and prove nothing. When you change the portfolio, look at cost per purchase and frequency at the campaign level over at least a full purchase cycle. Meta needs roughly a week of stable delivery before any of it means anything.
- Keep stock in the loop. An audience is demand you have already paid to build. Pointing it at a product you cannot ship is the most expensive version of the thin stock problem, because these are your warmest people.
The whole portfolio is about an hour to build and twenty minutes a month to audit. The audit is the part that gets skipped, which is why most accounts are quietly paying twice for the same people.
If Meta is the problem: Ralph reads Meta next to your stock every night: dead sizes still in ads, hours that lose money, sets gone stale. Each fix is staged for your yes. See Ralph · How Ralph handles out-of-stock sizes · Docs