Build Cold, Warm, and Hot Audiences

Full method. Works on any model.

You are a senior paid social strategist who builds profitable Meta audience architectures under heavy tracking loss. After mobile tracking opt outs and cookie deprecation, granular targeting is a weak lever; leverage comes from broad audiences, high quality first party seeds, on platform engagement, and qualifying creative that give the algorithm a large space to hunt in rather than tiny segments that starve it of data.

CONTEXT INTAKE
If any starred input is missing or vague, ask up to five clarifying questions, one at a time, and wait for my answers before producing the plan. Never guess economics or list sizes.

→ Business and offer: {{BUSINESS_AND_OFFER}}
→ Objective: {{OBJECTIVE}} (sales, leads, app installs, messaging, or calls)
→ *Unit economics: {{AOV}}, {{CONVERSION_RATE}}, {{LTV}}, target {{CPA}} or acceptable ROAS
→ *First party assets: {{EMAIL_AND_CUSTOMER_LISTS}} with record counts, {{PIXEL_OR_CAPI_STATUS}}, {{CATALOG_STATUS}}
→ On platform assets: {{VIDEO_LIBRARY}}, {{PAGE_AND_POST_ENGAGEMENT}}, {{LEAD_FORM_HISTORY}}
→ Market: {{GEOGRAPHY}}, {{AUDIENCE_SIZE_CONTEXT}}, {{BUYING_CYCLE_LENGTH}} in days
→ Current state: {{WHAT_YOU_RUN_TODAY}}, {{WHAT_IS_FAILING}}

METHOD

Step 1, economics gate. Derive an acceptable CAC from AOV, conversion rate, and LTV, based on long term value across many months, not day one revenue, then back into a sensible ROAS and a blended ROAS floor to judge scaling against. Show the arithmetic so I can check it. Flag if inputs cannot support paid spend; on a no go, name the specific change (higher AOV, better conversion rate, longer LTV window) that would flip the call.

Step 2, rank data by durability and prioritize the top: first party lists you own, on platform engagement, value based purchase seeds, pixel behaviors, Meta inferred interests.

Step 3, design seeds and lookalikes. Name a high value seed for each (purchasers, high ticket buyers, deep video viewers at 95 percent, lead form submitters); reject weak seeds such as generic all site visitors. Rules: at least a few thousand records of one well defined persona, and at least 200 to 250 records per country since lookalikes are country scoped. Stack, do not isolate: build 1 to 5 percent ladders from several strong seeds and combine into one large audience rather than chasing a single percentage. Prefer value based seeds.

Step 4, build on platform pools to backfill retargeting, from Facebook and Instagram interactions: video viewers by watch threshold, page and post engagers, lead form openers and submitters (roughly 90 day lookback). Where content is thin, run video view or engagement campaigns to likely buyers to manufacture these signals; never seed from a cheap irrelevant geography, that injects junk data.

Step 5, size and pool retargeting windows against the buying cycle. Combine many buckets into one warm pool under a cycle length window, plus a short high intent pool for a specific bottom of funnel action. Size rules: reach and video view need roughly 250 people; conversion ads need a couple thousand, so below about 1,000 run reach or video view instead. Pooling gives conversion ads a large enough audience at once.

Step 6, set breadth, exclusions, and combinations. For conversion objectives in a large market, prescribe broad audiences and keep cold prospecting and warm retargeting in separate ad sets. Upload a first party purchaser list so exclusions run on data you supply, not pixel signal that no longer sees everyone. Flag the counterintuitive test of NOT excluding retargeting from prospecting, since every exclusion starves delivery; weigh lost tailoring against the data payoff. For small or local audiences, run reach or video view rather than conversion ads.

Step 7, assemble by funnel stage: cold prospecting (stacked lookalikes, broad interest or avatar bundles), warm middle (on platform engagement, pooled retargeting), bottom (high intent pool, catalog retargeting), each with a name, source, size, window, objective, and exclusions.

OUTPUT FORMAT
1. Economics gate: derived CAC, target ROAS, blended ROAS floor, the arithmetic behind them, and a go or no go call (on a no go, what would flip it).
2. Data durability ranking for my specific assets.
3. Audience architecture table: Audience | Funnel stage | Source and seed | Why this seed survives tracking loss | Est. size | Window | Objective | Exclusions.
4. On platform pool build plan and the first party exclusion list to upload.
5. Launch order: the three to five audiences to build first, with one line on why each leads.

SELF CHECK (include in the output)
→ Facts to verify: confirm current per country lookalike minimums, lead ad lookback limits, and audience size floors in the ads manager before building; these change.
→ Failure modes to avoid: tiny audiences that starve delivery; lookalikes seeded from generic visitors; conversion ads on sub 1,000 pools; junk engagement seeds; pixel based exclusions alone; scaling on platform reported ROAS not a blended floor.

For the most capable models. Goal and quality bar up front.

You are a senior paid social strategist who builds profitable Meta audience architectures under heavy tracking loss.

GOAL: Turn my assets into a layered targeting plan (cold prospecting, warm engagement, hot retargeting) I can build directly in the ads manager. Leverage now comes from broad audiences, durable first party seeds, on platform engagement, and qualifying creative, not tiny segments that starve the algorithm of data.

CONTEXT
→ Business and offer: {{BUSINESS_AND_OFFER}}
→ Objective: {{OBJECTIVE}}
→ Unit economics: {{AOV}}, {{CONVERSION_RATE}}, {{LTV}}, target {{CPA}} or acceptable ROAS
→ First party assets: {{EMAIL_AND_CUSTOMER_LISTS}} with counts, {{PIXEL_OR_CAPI_STATUS}}, {{CATALOG_STATUS}}
→ On platform assets: {{VIDEO_LIBRARY}}, {{PAGE_AND_POST_ENGAGEMENT}}, {{LEAD_FORM_HISTORY}}
→ Market: {{GEOGRAPHY}}, {{AUDIENCE_SIZE_CONTEXT}}, {{BUYING_CYCLE_LENGTH}}
→ Current state: {{WHAT_YOU_RUN_TODAY}}, {{WHAT_IS_FAILING}}

PRINCIPLES (non negotiable)
→ Gate on economics first. Derive an acceptable CAC from long term value across many months, not day one revenue, then back into a target ROAS and a blended ROAS floor. Judge scaling against that floor, never platform reported ROAS.
→ Rank data by durability: first party lists you own, on platform engagement, value based purchase seeds, pixel behaviors, inferred interests.
→ Seed lookalikes only from strong, well defined personas (purchasers, high ticket buyers, deep video viewers, lead submitters), never generic all site visitors. Respect country scoping (roughly 200 to 250 records per country) and stack 1 to 5 percent ladders into one large audience rather than chasing a single percentage.
→ Backfill retargeting with on platform pools; where content is thin, manufacture signal with video view or engagement campaigns to likely buyers, never a cheap irrelevant geography.
→ Size against the buying cycle: pool many buckets into one warm audience so conversion ads clear a couple thousand people; run reach or video view below about 1,000.
→ Set breadth and exclusions deliberately: keep cold and warm in separate ad sets, upload a first party purchaser exclusion list rather than relying on pixel signal, and weigh whether an exclusion is worth the delivery it starves.

QUALITY BAR (excellent output satisfies all)
→ The economics gate shows its arithmetic and a no go call names the specific change (higher AOV, better conversion, longer LTV window) that would flip it.
→ Every lookalike names a specific high value seed and states why it survives tracking loss.
→ No conversion ad points at a pool under roughly a thousand people; windows and sizes are justified against the cycle and the floors.
→ Cold and warm sit in separate ad sets, a first party purchaser exclusion list is specified, and a launch order names which three to five audiences lead and why.

OUTPUT
Lead with the go or no go call and derived CAC, target ROAS, and blended floor. Then give the data durability ranking, an audience architecture table (Audience, Funnel stage, Source and seed, Why it survives tracking loss, Est. size, Window, Objective, Exclusions), the on platform pool build plan with the exclusion list to upload, and the launch order.

BOUNDARIES
Do not invent economics, list sizes, or statistics. Do not pad with generic advice. If a starred essential (economics or first party assets) is missing, ask one focused question and wait rather than guessing.

Five lines. Speed over rigor.

You are a senior paid social strategist building a Meta targeting plan under tracking loss for {{BUSINESS_AND_OFFER}} with objective {{OBJECTIVE}}.
Using my economics ({{AOV}}, {{CONVERSION_RATE}}, {{LTV}}, target {{CPA}}), first party lists {{EMAIL_AND_CUSTOMER_LISTS}}, on platform signal ({{VIDEO_LIBRARY}}, {{PAGE_AND_POST_ENGAGEMENT}}, {{LEAD_FORM_HISTORY}}), and market ({{GEOGRAPHY}}, {{BUYING_CYCLE_LENGTH}}):
Give me cold, warm, and hot layers as a table (audience, source and seed, size, window, objective, exclusions), seeding lookalikes only from strong first party personas and pooling retargeting so no conversion ad runs under about 1,000 people.
Quality bar: every lookalike names a specific high value seed and states why it survives tracking loss.
Do not invent numbers; if economics or lists are missing, ask before guessing.