---
title: Ad Audience Targeting Skill for Claude Code
canonical: https://skillselion.com/skills/mbfinotti/advertising-skills/ad-audience-targeting
content_type: text/markdown
description: "ad-audience-targeting by mbfinotti - This skill turns a business's ideal customer profile and buying signals into a layered ad audience targeting plan."
---

# Ad Audience Targeting Skill

Title: Ad Audience Targeting Skill for Claude Code

> ad-audience-targeting by mbfinotti - This skill turns a business's ideal customer profile and buying signals into a layered ad audience targeting plan.

Canonical: https://skillselion.com/skills/mbfinotti/advertising-skills/ad-audience-targeting

This skill turns a business's ideal customer profile and buying signals into a layered ad audience targeting plan. It covers cold prospecting, interest and in-market segments, lookalikes, first-party lists and retargeting pools, each sized against platform floors. Exclusion rules keep the pools from overlapping. A solo founder reaches for it when setting up audiences for a new paid campaign.

## Key facts

- Installs: 26,726
- GitHub stars: 2
- Author: mbfinotti
- Repository: mbfinotti/advertising-skills
- Install command: `npx skills add https://github.com/mbfinotti/advertising-skills --skill ad-audience-targeting`
- Last updated: October 26, 2026

## What Ad Audience Targeting does

Builds a layered ad audience plan from the ideal customer profile, from broad prospecting to retargeting, with overlap rules.

Paid audiences are built ad hoc and overlap or are too small to run.

A sized, non-overlapping audience structure.

Best for: Founders setting up paid audiences for a new offer.

Not for: Organic-only channels with no paid audiences.

- Layered prospecting, interest, lookalike and retargeting pools
- Audience sizes checked against platform floors
- Exclusion rules to stop overlap

## SKILL.md

## Audience Targeting

Map who the customer is (ICP) and what they do that signals intent (buying signals) into a layered paid-media targeting plan: which audience tiers to run, how each is defined, sized, budgeted, sequenced for testing, and kept from bidding against the others.

The tier logic, exclusion discipline, overlap math, and funding floors are identical for B2B and B2C. What differs (see the B2B vs. B2C section for detail):

- Universe size: B2B ICPs are often smaller than platform minimums.
- Which tiers dominate: B2B leans firmographic and list-based; B2C leans broad, lookalike, and behavioral.
- Cycle length.
- Success metric: pipeline quality vs. volume CPA/ROAS.

This skill produces the plan, not the platform setup. It places and sizes the retargeting tier but does not design the per-stage retargeting sequence (mbfinotti/advertising-skills@retargeting-funnel), and it assumes a seed list exists but does not select or build it (mbfinotti/advertising-skills@lookalike-audience-seeds).

### Evidence gate

- Never assert an ICP, a buying signal, or a tier without sourced data behind it: an ICP document, customer/CRM data, analytics, or past campaign results.
- Missing ICP → stop and request one (or a customer list to derive it from). Never invent an ICP from the product description alone; a plan built on an imagined customer optimizes toward fiction.
- Unknown attributes score neutral, never negative. Absence of evidence is not a negative signal; scoring it negative systematically buries accounts with a thin public footprint, which correlates with size, not fit.
- Record the evidence next to each tier in the plan (source and date), so the plan can be audited when performance surprises.

### Clarifying questions

Ask once, briefly, before planning; skip anything already answered.

1. B2B or B2C, and what does the ICP say - who buys, and what evidence backs it?
2. What counts as a conversion, and what is the target CPA (or ROAS)?
3. Roughly how many conversions per week does the account currently produce? (Sets the broad-vs-layered default.)
4. What first-party assets exist: customer list (size), site traffic volume, CRM, engaged followers?
5. What is the total monthly budget for this plan?
6. How large is the addressable universe - thousands of accounts or millions of consumers?
7. Does the offer touch a regulated category (housing, employment, credit, financial products, health, politics)?
8. By what date must results land - is there a hard deadline (launch, quarter close, seasonal window), or is the timeline open?
9. Do you want a one-off win this cycle, or a compounding asset that keeps paying after the campaign ends?
10. What is your effort ceiling - hours per week, who executes, and whether legal/privacy review is available to you at all?

The last three re-rank the tier plan before it is written. Say in the plan which answer moved which tier:

- Hard near-term date promotes first-party and retargeting tiers - they return readable evidence within days - and demotes cold broad, which needs a full learning cycle before it says anything.
- Compounding mandate promotes first-party list growth and cold broad: one builds a seed and a suppression asset the business keeps, the other trains a delivery model that improves with every conversion. A retargeting pool decays to nothing within weeks of the refill stopping.
- No privacy review available at all deletes every tier that needs identifiable data uploaded - first-party custom, and lookalike built from a customer seed - rather than ranking them last; they are not options for this account. A low effort ceiling on its own only demotes them, and promotes platform-native behavioral and broad tiers.

### Attributes vs. signals

The translation step everyone skips: an ICP lists _attributes_, a targeting plan runs on attributes _and_ signals. Distinguish them explicitly.

|            | Attribute - "what is true"                                      | Signal - "why now"                                                                  |
| ---------- | --------------------------------------------------------------- | ----------------------------------------------------------------------------------- |
| Answers    | Is this person/account a fit?                                   | Why are they more likely to buy now?                                                |
| Examples   | industry, company size, role, age, geography, category interest | funding round, hiring spike, pricing-page visit, cart abandonment, seasonal trigger |
| Feeds      | tier membership, exclusion lists, firmographic filters          | tier priority, recency windows, test order                                          |
| Goes stale | slowly - refresh quarterly                                      | fast - has an expiry date                                                           |

- A signal older than ~90 days is context, never a trigger. Two fresh independent signals (e.g. funding + relevant hiring) outrank one stale one.
- Test for any candidate signal: given ten people/accounts that already pass the attribute filter, does it say which to reach first? If not, it's an attribute wearing a signal costume.

### The tier model

Every plan is assembled from these six tiers, ordered by efficiency - outcome bought per unit of effort - never by which is cheapest to stand up. Include a tier only when a real, evidenced signal defines it; an empty tier is a line item, not an audience.

- efficiency: `first-party custom > retargeting > lookalike > behavioral/in-market > cold broad > interest & affinity`
- value (volume of results the tier can produce at its ceiling): `cold broad > lookalike > behavioral/in-market > first-party custom > retargeting > interest & affinity`
- effort: `cold broad > retargeting > first-party custom == lookalike > behavioral/in-market == interest & affinity`
- compliance cost: `first-party custom > lookalike > retargeting > behavioral/in-market > interest & affinity == cold broad`

The axes disagree on purpose:

- Interest, the cheapest tier to stand up, is the least efficient one.
- Cold broad, the tier with the highest ceiling, is the most expensive to run and the slowest to read.
- First-party leads because a list the business already owns converts after an hour of setup. It also carries the heaviest consent, DPA and data-residency exposure of the six, which is what the compliance axis is there to price.

| Tier                                 | Defining signal                                                             | Effort to stand up                                                                                             | Volume vs. intent                                        | NOT for                                                    |
| ------------------------------------ | --------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------- | ---------------------------------------------------------- |
| First-party custom                   | Owned data: customer/CRM list, site visitors, content engagers              | An hour - the list already exists; add a consent and data-residency check before upload                        | Mid-bottom funnel: low volume, high intent               | Prospecting scale                                          |
| Retargeting pool                     | Prior engagement with the business, graded by recency × depth               | A week or more of traffic or spend to fill the pool; near-zero to maintain afterwards                          | Bottom: lowest volume, highest intent                    | Proving the plan works - its ROAS is structurally inflated |
| Lookalike / similar                  | Statistical resemblance to a seed list                                      | An hour once a seed exists; sourcing the seed is a separate job                                                | Top-mid: high volume, intent quality tracks seed quality | Replacing retargeting; seeds of engagement-bait clickers   |
| Behavioral / in-market               | Category-level purchase-intent actions tracked by the platform              | An hour - platform-native segments, nothing to build or upload                                                 | Mid: medium volume, medium intent                        | Broad awareness reach                                      |
| Cold prospecting (broad/algorithmic) | None - the delivery algorithm selects using the account's conversion signal | A standing job - budget held above the learning threshold plus a continuous creative supply                    | Top: highest volume, lowest intent                       | Learning who the buyer is; tiny addressable universes      |
| Interest & affinity                  | Declared or inferred interests                                              | An hour - and largely spent for nothing; major platforms treat interest inputs as suggestions, not constraints | Top: high volume, weak intent                            | Precision reach - largely deprecated as a precision tool   |

- Default rung: fund first-party and retargeting first. Move up one rung - lookalike, then behavioral, then cold broad - as soon as the warm tiers saturate or the pipeline they draw on stops refilling. Warmer tiers are smaller and cheaper per result; colder tiers are what refills them, so a plan that funds only warm tiers eats its own pipeline within weeks.
- When site traffic is too thin for a retargeting pool, engagement audiences are the standard fallback seed, ranked by how fast they fill: `video viewers > content/profile engagers > lead-form openers`. A video-view pool reaches usable size in about a week of spend; a lead-form-opener pool can take one to three months of it.
- This ranking is a default, not a law - it shifts with context and with who executes it. Re-rank it against what you already know about the user before writing the plan:
  - An owned list of tens of thousands of customers promotes first-party above everything else.
  - Dense existing pixel traffic promotes retargeting.
  - An in-house data or analytics team promotes lookalike.
- A constraint the user actually stated does something different from re-ranking: it removes the tier. Delete any tier the answers rule out, and name it as deleted in the plan with the constraint that killed it and what would bring it back, for example:
  - "cold broad: deleted, addressable universe of a few tens of thousands, revive if the universe or the geography widens".
  - No privacy review deletes first-party and customer-seeded lookalike.
  - A regulated offer in a category that disallows lookalikes deletes that tier (see Compliance gate).

  A ruled-out tier left sitting at the bottom of the table is indistinguishable from one nobody considered, and it comes back next planning cycle as a fresh idea.

### Broad vs. layered: the judgment call

The most contested decision in this domain - present it as a conditional, never a doctrine. The ranking flips on one input, account data density, so it is stated twice rather than blended into a single misleading order:

- efficiency (new or thin-data account, narrow ICP or small universe, small budget, regulated offer): `explicit layers > parallel test > broad`
- efficiency (dense conversion data, large universe, budget that feeds the algorithm, goal is efficient scale): `parallel test > broad > explicit layers`

Default rung: explicit layers. The one condition that moves the plan up a rung is a campaign comfortably clearing its learning threshold - then run the parallel test, and adopt broad only if the duplicate wins.

| Posture                                                                    | You spend                                                                               | You get                                                                            | You give up                                                                   |
| -------------------------------------------------------------------------- | --------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------- | ----------------------------------------------------------------------------- |
| Explicit layers                                                            | A week to define tiers, then a standing job maintaining exclusions and refreshing lists | Attribution - reveals which segment actually responds                              | Reach the algorithm would have found on its own, plus the over-narrowing risk |
| Parallel test (duplicate one campaign, switch only the duplicate to broad) | An hour of setup plus one test cycle of budget on the duplicate                         | An account-specific answer instead of someone else's general one, fully reversible | One cycle of delay before committing either way                               |
| Broad / algorithmic                                                        | Near-zero to set up, but sustained budget density across a full learning cycle          | Efficient scale wherever the universe is genuinely large                           | Any knowledge of who the buyer is; wasted reach when the real market is small |

Where the evidence sits:

- **The broad camp:**
  - Ben Heath reports open/broad targeting "often (but not always) outperforms interest targeting", especially on "mature ad accounts with decent conversion data".
  - Demand Curve independently backs broad as "the opposite of traditional marketing wisdom" that works.
  - Platform vendors publish self-reported data agreeing; treat those numbers as marketing.
- **The layered camp:** B2B practitioners (B2Linked among them) caution against automatic audience-expansion features - with a narrow, well-defined ICP, letting the platform broaden the audience "to anybody and everybody" wastes budget.
- **The middle:** Heath himself rejects the "kill retargeting entirely" camp as "the more extreme view" - even broad-first accounts keep first-party and retargeting tiers.

- Never flip an account wholesale on general advice; the parallel test is the only honest way to change rungs, and it keeps the layered control running while it decides.
- Broad hides who the buyer is. Keep at least one manually-defined tier alive as a research instrument even when broad wins on cost.
- Over-narrowing is the named failure mode of the layered camp: stacking many interests plus demographic filters produces a small, expensive audience that sees an unchanged ad. If the ICP knowledge is rich, spend it on creative variants per segment, not on more filters.

### Sizing and funding floors

A tier that can't be funded to significance doesn't belong in the plan.

1. Size each tier and check it against the platform's minimum audience size ([references/platform-notes.md](references/platform-notes.md) - load when the user names their platform; platform floors change without notice, so verify against current platform docs).
2. Compute each tier's budget floor from the platform's learning threshold: daily floor ≈ (target CPA × weekly optimization-event threshold) ÷ 7. The most widely cited threshold is ~50 events per ad set per week. An ad set at $30/day against a $50 CPA yields ~4 conversions a week - it will never exit learning; merge it, or optimize to a higher-funnel event.
3. Below the size floor or the budget floor, fix it in this order - efficiency: `merge upward > cheaper proxy event > cut`.
   - **Merge the tier into its nearest neighbour** - an hour of rebuilding exclusions, and it keeps both the audience and the delivery. Try this first.

Excerpt. The rest of the file is on the canonical page and in the source repository.

## Ad Audience Targeting FAQ

### Is Ad Audience Targeting safe to install?

skills.sh reports 3 of 3 security scanners passed. Review the Security Audits panel on this page before installing in production.

## Machine access

- [Sitemap](https://skillselion.com/sitemap.xml)
- [llms.txt](https://skillselion.com/llms.txt)
- [llms-full.txt](https://skillselion.com/llms-full.txt)
- [Developer resources](https://skillselion.com/developers)
- [OpenAPI spec](https://skillselion.com/openapi.json)
- [MCP handshake](https://skillselion.com/.well-known/mcp)
- [API catalog](https://skillselion.com/.well-known/api-catalog)
- [AI catalog](https://skillselion.com/.well-known/ai-catalog.json)
- [Pricing](https://skillselion.com/pricing.md)
- [Auth](https://skillselion.com/auth.md)
- [AGENTS.md](https://skillselion.com/AGENTS.md)
- [Contact](https://skillselion.com/contact)
- [Home](https://skillselion.com/)
