Pulsar · Metrics playbook · No. 14 · Custom audiences vs lookalikes. Real audiences from the test account.
Custom vs Lookalike
One re-reaches people who already know you; the other finds strangers who resemble them. Two opposite jobs — use both.
Companion to No. 01–13 — it sits alongside No. 03 · Lookalike Seed Quality (what makes a seed good) and No. 08 · Almost Bought (retargeting the warm). Leans on No. 02 · Quantity vs Quality (which funnel stage means intent) and No. 04 · Invisible Sales (matched, real-people data feeds both tools).
Contents
01 Two opposite tools · 02 Custom audiences · 03 Lookalikes · 04 The seed is everything · 05 Who to use when · 06 The missing exclusion · 07 The stack
Two opposite tools — constantly confused
A custom audience is people you already have — your visitors, your buyers, your list. A lookalike is new strangers Meta thinks resemble them. One re-reaches the warm; the other prospects the cold. Mixing up their jobs wastes budget.
A custom audience re-finds people who already know you. A lookalike finds strangers who act like them. Never hire one to do the other’s job.
Custom vs Lookalike — the six differences that matter
Same ad account, opposite directions: amber re-reaches people you have; blue goes looking for people you don’t.
| Custom audience · warm | Lookalike · cold | |
|---|---|---|
| Who | People you already have — visitors, buyers, your list | New strangers Meta thinks resemble them |
| Temperature | Warm — they’ve already met you | Cold — they’ve never heard of you |
| Source | Your data — site events, a customer list, engagers | A seed custom audience + Meta’s modelling |
| Job | Retarget the warm · exclude the bought | Prospect at scale |
| Size behaviour | Finite — shrinks as intent decays | Scalable — 1% to 10% of a country |
| Depends on | The quality of your data (No. 04) | The quality of the seed (No. 03) |
Interactive in the HTML version — a toggle dims one column so each tool’s profile reads on its own. The printed state shows both columns at full strength.
Custom audiences — your known people
Built from your data — site events, a customer list, video or page engagers. They have exactly two jobs: retargeting (warm — they’ve already met you, No. 08) and exclusion (stop paying to reach people who already bought). And they are finite: they shrink as intent decays.
Warm, known — and limited. A custom audience can never be bigger than the people who actually touched your funnel, and it shrinks as the window rolls and intent decays. The 52 in Stalled Cart today are not the same 52 next month. That finiteness is exactly why custom audiences can’t grow an account — growing is the lookalike’s job (section 03).
A custom audience is a memory, not a growth engine. It can only ever re-reach people your funnel already met.
Lookalikes — strangers who resemble them
Meta takes a seed custom audience and finds new people with similar behaviour — so you can prospect at scale (cold traffic). Exactly two things define a lookalike, and you control both.
A lookalike is a statistical echo of its seed — it can only be as sharp as the audience it was modelled on.
The seed is everything — built from this account’s real lists
A lookalike is only as good as the custom audience it’s built from. On the test account, Purchasers alone = 20 people — too thin (Meta wants roughly 100+ matched people to model from). Add the 79 Shared-Details near-buyers — people who entered checkout and gave their details but didn’t complete, likely a shipping fee — and the combined High-Intent Seed = 99: right at Meta’s floor, and still all high quality.
Pick a seed — size against the ~100 floor, intent against reality
All four are real custom audiences on the test account. A seed needs two things at once: enough people to model from, and high enough intent to be worth cloning.
Interactive in the HTML version — buttons switch between the four candidate seeds (Purchasers 20 · Shared Details 79 · High-Intent Seed 99 · Stalled Multi-Viewer 1,520), each gauged against the ~100 floor with an intent meter and a verdict. The printed state is the High-Intent Seed: 20 buyers + 79 near-buyers = 99 — just enough, and still high quality.
The rule: seed on your highest-quality converters, not your biggest low-intent list. When converters alone are too few, extend down the intent ladder one rung at a time (buyers → near-buyers) until you clear the floor — and stop there. Every rung further down dilutes what Meta learns to clone.
Who to use when — the funnel-temperature map
Every stage of the funnel has a temperature, and every temperature has exactly one right tool. Match the audience to the temperature — cold gets a lookalike, warm gets a reminder, customers get left alone — or cloned.
The funnel-temperature map
Pick a stage — the funnel highlights it and names the one audience type whose job it is. Stage counts are the account’s real lists from section 02.
Interactive in the HTML version — selecting a funnel stage highlights its band and names the right audience type and job. The printed state highlights the cold band: strangers → lookalike → prospect broad.
| Temperature | Who they are | Audience type | The job |
|---|---|---|---|
| COLD | Never met you | LOOKALIKE | Prospect / grow — broad, seeded on converters (section 04) |
| WARM | Visited, didn’t buy | RETARGETING CUSTOM | Recover — remind them (No. 08 · Almost Bought) |
| HOT | Cart / checkout / shared details | TIGHT RETARGETING CUSTOM | Close — the 52 + 22 + 79 are one objection from buying |
| CUSTOMER | Bought | EXCLUDE + SEED | Exclude from prospecting (section 06) · seed lookalikes · upsell separately |
Match the audience to the temperature. Cold gets a lookalike, warm gets a reminder, customers get left alone — or cloned.
The exclusion you’re probably missing
Always exclude your Purchasers custom audience from prospecting and top-funnel retargeting — you already paid to win them; stop paying to re-sell (No. 03, No. 08). And exclude current customers from their own lookalike’s delivery too — the people most similar to your buyers include your buyers.
You already paid to win them. Stop paying to re-sell to them.
Exclusions are the cheapest optimization there is: pure saved spend, zero downside. No creative to test, no learning phase, no risk — one checkbox that stops money leaking to people whose money you already have. If you change nothing else after this deck, add the Purchasers exclusion.
The stack — put it together
A healthy account runs three at once: a lookalike to find new people, retargeting custom audiences to recover the warm, and a Purchaser exclusion so nothing is wasted — every one fed by clean, matched, real-people data (No. 04 · Invisible Sales).
Lookalike — to grow
Seeded on the High-Intent Seed (99: Purchasers 20 + Shared Details 79), starting at 1%. The only tool in the stack that finds people your funnel has never met (sections 03–04, No. 03).
Retargeting customs — to recover
Stalled Cart 52 · Stalled Checkout 22 · Stalled Multi-Viewer 1,520 — each stage of stall gets its own reminder, tightest where intent was highest (section 05, No. 08).
Purchaser exclusion — to stop the bleed
Purchasers (20) excluded from prospecting, top-funnel retargeting and their own lookalike’s delivery (section 06). Pure saved spend — the checkbox most accounts never tick.
Custom to recover, lookalike to grow, exclusions to stop the bleed.
| Thread | What it decides | Where the full read lives |
|---|---|---|
| Seed quality | GARBAGE IN, GARBAGE OUT what makes a seed worth cloning | No. 03 · Lookalike Seed Quality |
| Retargeting the warm | RECOVER how to re-reach the carts and checkouts that stalled | No. 08 · Almost Bought |
| Matched data | REAL PEOPLE whether the lists are built on identified, matched events | No. 04 · Invisible Sales |
| Funnel stages | INTENT LADDER which stage means intent — the temperature scale of section 05 | No. 02 · Quantity vs Quality |
Where Pulsar fits, in one honest line: Pulsar builds both halves of this deck from your funnel — rule-based custom audiences per stage (the Stalled Cart / Checkout / Multi-Viewer lists are exactly that), a Purchase + Shared-details seed, and one-click Meta lookalikes from a synced seed — so the seed is high-quality and the exclusions are automatic.