G1: Spend Reallocation — 2025 Historical Pattern

Section 1: Where Spend Went (April 2025)

Budget concentration by demographic segment shaped everything that follows. Before reading the ROAS charts, see where money actually landed.

The two largest segments (35-44 male, 23% of spend; 45-54 male, 14%) both delivered below-account-average ROAS in April. The consistently top-performing segment (25-34 male) received equal budget to the worst-performing one.

Claims: MD-D-001, MD-D-002, MD-D-007. Donut evidence: MD-D-002 (budget misallocation — 25-34 male at 14% of spend, 11.1x ROAS vs 35-44 male at 23%, 2.5x ROAS). Efficiency quad: MD-D-002 (bleeders lower-right vs winner upper-center). CAC chart: MD-D-002 (cost quality divergence)

Section 2: ROAS Collapse — March vs April by Segment

Eight of twelve demographic segments declined from March to April. The 45-54 and 35-44 male drops are the loudest — but they rode an account-wide wave, not an isolated targeting failure. Read the full picture before acting on individual segments.

Claims: MD-D-006, MD-C-003 (seasonality confounder). Delta bar: MD-C-001 (8 of 12 declined). CAC chart: MD-D-002 (25-34 male cost quality vs 45-54 male)

Section 3: Pause List — Bleeders with $ Wasted

Two lenses: placement bleeders (small in absolute KES) and demographic bleeders (where the real money is). Placement pauses are quick but won't move the needle. Demographic restrictions require adset-level changes.

The placement-level bleeders total only ~5,054 KES spend — rounding error against the 258K account total. The Facebook Search mobile_app "0.55x ROAS" is based on exactly 1 purchase. Do not act on it as a signal.

Claims: MD-D-003 (placement bleeders), MD-D-005 (Mar vs Apr placement comparison). Bubble chart: MD-D-002 (bleeders vs winner scatter — 35-44 male lower-right quadrant; 25-34 male upper-right)

Section 4: The Scale-Up Candidate — 25-34 Male

25-34 male is the account's most consistently efficient segment across all 12 months. But the April ROAS of 11.1x is the single highest monthly observation of the year — and it is partly an Average Order Value (AOV) anomaly, not a pure audience quality signal.

The roas_if_aov_2000 column shows what ROAS would have been if April's AOV had held at the 12-month median of ~2,000 KES rather than spiking to 3,241 KES. This is the AOV decomposition the analyst did not run — added by the critic (MD-C-001).

Claims: MD-C-001 (AOV spike bar — shows April 3,241 KES vs median 2,000 KES), MD-D-002 (25-34 male winner consistency across all 12 months vs comparators)

Claims: MD-D-002 (winner consistency), MD-C-001 (AOV decomposition), MD-C-003 (AOV anomaly flag)

Section 5: Projected Uplift — Honest Math

The +45.3% headline was built by combining April's peak 25-34 male ROAS (11.1x) with the assumption that doubling spend maintains that ROAS. Both inputs are optimistic. The critic retired this as a decision input. Here is the corrected picture.

The math for the conservative case:

• Redirected spend = 50% of 35-44 male April + 50% of 45-54 male April = (59,948 × 0.5) + (37,068 × 0.5) = 47,508 KES
• Baseline revenue from that spend at blended 2.35x ROAS = 111,655 KES
• Projected revenue at 5.5x ROAS = 261,212 KES
• Net uplift = +150K KES ≈ +16% of April's 956K total

Claims: MD-D-005 (reallocation arithmetic), MD-C-007 (planning fallacy / peak-on-peak)

Section 6: Action Checklist

Actions are ordered by confidence × impact. Placement pauses are low-risk/low-reward. Demographic restructuring is the real lever — but requires the AOV stability check (MD-C-003) before committing full budget.

Claim Provenance

All underlying SQL is reproducible. The parquet files are from Meta Insights API pulls (W3 = placement_insights, W4 = demographic_insights).