Audience Architecture — AS BUILT (verified live in Meta)

Date: June 13, 2026

Account: LG Military Ads (1034037341617540)

Status: ✅ Built and live — verified against Meta API. Supersedes the Jun 11 architecture spec.

TL;DR — what we did today

Replaced a 10-month-old stale audience layer with a fresh, verified one — 18 new audiences built, all live, plus two account-level segment swaps. Match rates landed healthy (~45-51%), so the lookalikes are sitting on strong seeds. The earlier "8% match" scare was a populating-snapshot artifact, not a real problem — confirmed once lists finished processing.

Headline numbers (verified live):

Suppression list: 82,900–97,600 matched (~46% of 195,137)
HighValue seed: 27,800–32,700 matched (~49% of 62,299)
Auto seed: 41,900–49,300 matched (~50%)
Combined seed: 50,100–58,900 matched (~49%)
Auto phone file: 10,000–11,800 matched (~51% of 21,375)

Part 1 — What changed (before → after)

Layer	BEFORE (stale)	AFTER (as-built today)
Customer seed	`Custom_Recruiting-Leads_08-05-2025` — 10 months old	6 fresh `GTW_` customer lists (Mar–May + 7-mo)
LAL ladder	1/3/5% off the stale 2025 seed	12 fresh `GTW_` LALs off 4 new seeds
High-value targeting	None	`GTW_HighValue_Seed` + LAL ladder (SAC-compliant value workaround)
Phone identifier	None (email-only)	`gtw_mil_auto_owned_phone` added (phone + conversion values)
Existing-customers segment	`Mil FB Customer Match 4.12.21` (5+ years old)	`GTW_Mil_AllCustomers_7mo_06-13` (~90K matched)
Engaged segment	Not defined	4 page-engagement audiences (~1.5M warm pool) ✅
Warm pool	1.5M sitting unused	`GTW_Warm_PageEngagement_365D` saved audience built

Part 2 — The audiences we built (live, with real match rates)

Customer-list seeds

Audience name (in Meta)	Source file	Raw rows	Matched (live)	Match %
`GTW_Mil_AllCustomers_7mo_06-13` (shows as mil_suppression_all_7mo.csv, label Customers)	`mil_suppression_all_7mo.csv`	195,137	82,900–97,600	~46%
`GTW_HighValue_Seed_06-13`	`mil_high_value_seed_7mo.csv`	62,299	27,800–32,700	~49%
`GTW_Combined_Seed_06-13`	`mil_combined_seed_90d.csv`	112,264	50,100–58,900	~49%
`GTW_Auto_Seed_06-13`	`mil_auto_seed_90d.csv`	91,871	41,900–49,300	~50%
`GTW_Home_Seed_06-13`	`mil_home_seed_90d.csv`	21,777	8,700–10,200	~43%
`GTW_Bundle_06-13`	`mil_bundle_customers_7mo.csv`	2,698	~1,000	(small — track only)
`gtw_mil_auto_owned_phone_06-13`	`mil_auto_owned_phone_06-13.csv`	21,375	10,000–11,800	~51%
`mil_auto_owned_phone_highvalue_06-13`	`mil_auto_owned_phone_highvalue_06-13.csv`	9,796	4,600–5,400	~51%

Lookalike ladders (12, all live)

Ladder	Audiences	Built from
HighValue ⭐	`Lookalike (1/3/5%) - GTW_HighValue_Seed_06-13`	HighValue seed
Auto	`Lookalike (1/3/5%) - GTW_Auto_Seed_06-13`	Auto seed
Home	`Lookalike (1/3/5%) - GTW_Home_Seed_06-13`	Home seed
Combined	`Lookalike (1/3/5%) - GTW_Combined_Seed_06-13`	Combined seed

(Meta no longer shows LAL size at creation — it displays when the LAL is loaded into an ad set, sized to that ad set's US location.)

Saved audience + account segments

GTW_Warm_PageEngagement_365D — saved audience, 1.6–1.9M (union of 4 page-engagement audiences)
Existing customers segment → swapped to GTW_Mil_AllCustomers_7mo_06-13 ✅
Engaged audience segment → set to the 4 page-engagement audiences ✅

Part 3 — The phone-file finding (important)

We added Auto_owned_export (phone + email + conversion values, ~3-month Auto window) expecting it to dramatically outperform the email-only files on match rate. It didn't — and that's good news, not bad:

Phone Auto file matched ~51%
Email Auto seed matched ~50%

They're essentially equal. This proves the email data was never the problem — the scary 8% we saw earlier was just a mid-population snapshot. Both land at a healthy ~50%.

So what's the phone file actually worth? Two things it has that the email files don't:

Real conversion values (93.5% populated, mean $94.81) → a cleaner high-value seed for the Auto Max Val rescue

Recency (Feb–Jun Auto cut) → fresher signal

Use it as the sharper Auto + HighValue seed, not as a match-rate fix. The match rate was already fine.

Part 4 — File → Audience → Campaign assignment map

This is where each audience gets deployed. (Ad-set loading + budget = Lauren/Anthony's lane — this is the target-state map for when they wire it in.)

Campaign	Ad set	Audience to load	Exclude
Auto Max Val (0.91x — bleeder)	Broad Value	`Lookalike 1/3/5% - GTW_HighValue_Seed` ⭐ (rescue)	Suppression
Auto ALC (1.73x — headroom)	Broad 2.O	`Lookalike 1/3/5% - GTW_Auto_Seed` (or phone-owned)	Suppression + HighValue 1%
Auto BAU (1.76x — engine)	Broad (Adv+)	Advantage+ broad	Suppression
Auto BAU	Warm Layer	`GTW_Warm_PageEngagement_365D`	Suppression
Home Max Val	Broad Value	`Lookalike 1/3/5% - GTW_HighValue_Seed` (combined)	Suppression
Home ALC	Broad	`Lookalike 1/3/5% - GTW_Home_Seed`	Suppression + HighValue 1%
Home BAU	Broad (Adv+)	Advantage+ broad	Suppression
Home BAU	Warm Layer	`GTW_Warm_PageEngagement_365D`	Suppression
ALL prospecting	(every ad set)	—	`GTW_Mil_AllCustomers_7mo` (suppression)

Account-level (already applied, automatic across all Sales campaigns):

Existing customers = GTW_Mil_AllCustomers_7mo → Meta optimizes toward net-new
Engaged = 4 page-engagement audiences → warm pool feeds optimization

Part 5 — Cleanup list (do when convenient — not urgent)

Duplicate/failed uploads from the troubleshooting session. Safe to delete; leave the canonical ones:

Delete	Why
`2gtw_mil_suppression_all_7mo.csv` (INACTIVE, 1,000)	Failed re-upload
`mil_suppression_all_7mo.csv` — Qualified leads label (99,200–116,700)	Duplicate; wrong label. Keep the Customers-labeled one instead

Retire AFTER the new LALs are validated in ad sets (don't delete yet — campaigns still use them):

Custom_Recruiting-Leads_08-05-2025 (stale seed) + its 1%/3%/5% LALs
The ~28 orphaned 1,000-size placeholder LALs (Worldwide/US engagement)

Part 6 — What results to expect, and when

The audiences sharpen WHO Meta reaches. They are not an instant switch. Realistic timeline:

Timeframe	What happens
Immediate	Existing-customers + Engaged segments apply automatically across all Sales campaigns. Meta begins steering toward net-new.
Days 1–3	Once LALs are loaded into ad sets, they enter Meta's learning phase. Directional signal only — don't judge yet.
Days 4–7	Learning phase stabilizes. CPL on LAL-targeted ad sets should begin to settle.
Weeks 1–2	Clean read. Expect: prospecting CPL improvement from fresher/sharper seeds, less waste from suppression, the HighValue LAL giving Auto Max Val a cleaner audience than its broken value signal.

Honest expectation-setting: the audience layer is one lever. Two things matter more for hitting Anthony's June number:

Capital deployment — Auto BAU "Broad" runs 1.76x ROAS, Auto Max Val "Broad" runs 0.91x (loses money). Feeding the winners is the bigger, faster lever. (Lauren/Anthony's call — separate memo.)

The fbc/CAPI value-signal fix — Max Val can't optimize properly until that's resolved. The HighValue LAL is the interim audience-side workaround, not the fix.

The audience work makes every prospecting dollar work harder. It compounds with the above; it doesn't replace them.

Appendix — quick reference


Account: 1034037341617540 (LG Military Ads)

GTW audiences live (verified 6/13):
  Seeds (6):     AllCustomers_7mo, HighValue, Combined, Auto, Home, Bundle
  Phone (2):     auto_owned_phone, auto_owned_phone_highvalue
  LALs (12):     {HighValue, Auto, Home, Combined} × {1%, 3%, 5%}
  Saved (1):     GTW_Warm_PageEngagement_365D (1.6-1.9M)
  Segments (2):  Existing customers + Engaged → updated

Match rates: all ~45-51% (healthy). 8% scare = populating snapshot, resolved.

Source files: /Users/mo/Downloads/For FB Audiences/output/
Prep script:  deliverables/audit_2026-06-04/prep_audience_csvs.py

← Back