G3: CPM Inflation Root Cause — Wait vs Act vs Platform-Fix
Founder: April is your worst month. CPM has risen 2.68x since Jan-2025 and is NOT seasonal — Apr-2026 CPM (351 KES) exceeds Apr-2025 CPM (331 KES) by 6% at double the spend. The structural floor has risen. Waiting will not help.
Decision: ACT. Three actions, 30 days, estimated 220K KES/month savings + ROAS recovery path.
Jan-2025 CPM (KES baseline)
Apr-2026 CPM (KES — 2.68x)
Apr-2026 CPA (KES — 5.9x rise)
% of Orders Visible to Meta Pixel
Meta/Shopify Attribution Ratio (Apr-2026)
Section 1: CPM Trend — Structural, Not Seasonal
The CPM inflation is structural. Apr-2026 CPM exceeds Apr-2025 CPM at the same seasonal position. Q4 holiday pressure explains the Dec-2025 peak (3.72x) but not the persistent floor in Jan-Apr 2026 (2.57-2.89x).
Peak CPM (KES, Dec-2025)
Peak Multiple vs Jan-2025 (x)
Apr-2026 CPM (KES)
Apr-2026 Multiple (x)
Delta vs Kenya Benchmark (KES)
CPM Monthly Trend with Regime Events (Jan-2025 to Apr-2026)
Regime change, not drift. Jan-Mar 2025 showed +26.9%, +40.3%, +42.0% consecutive monthly CPM jumps — a step function, not gradual drift. This timing aligns with iOS 18.x ATT enforcement tightening. The account experienced a discrete shock, not a slow ramp.
Jan-12-2026 discrete event: Meta deprecated 7-day-view and 28-day-view attribution windows on Jan-12-2026. NairoMarket's Meta/Shopify ratio dropped from 0.94 (Jan-2026) to 0.81 (Feb-2026) — the sharpest single-month ratio drop in the dataset — consistent with this platform change (MD-R-G3-004).
Mar-2026 platform event: Meta's outcome-based optimisation rollout raised CPMs 15-40% industry-wide in March 2026. NairoMarket's Mar-2026 CPM rose +14% MoM — inside this band. This is a platform event, not Kenyan advertiser competition.
Section 2: H3 — iOS Signal Loss (Primary Driver, 40-45% Attribution)
Meta's pixel is active and fires regularly. But iOS App Tracking Transparency blocks client-side event matching at the OS level. Meta's conversion optimization sees only 73% of actual Shopify orders.
Jun-2025 crossover. The ratio crossed below 1.0 in Jun-2025 and has never recovered. Apr-2026: Meta reports 358 purchases; Shopify processed 490. The 132-order gap is invisible to Meta's optimization algorithm.
Taleb's challenge: This ratio decline is also consistent with Shopify growing non-Meta channels (organic, WhatsApp, TikTok). UTM source audit would distinguish. CAPI is the correct intervention regardless — it either fixes iOS ATT blocking or improves signal quality for whatever portion IS Meta-attributed.
Section 3: H4 — Dead Retargeting Pools (Secondary Driver, 25-30%)
All 18 WEBSITE retargeting audiences show size = 20 users (below Meta's 1,000-user delivery minimum for ad delivery). The account is running 100% on cold broad + Advantage+ targeting.
Retargeting Audience Pool Status (Apr-2026 Snapshot)
Why the pools are dead: With 490 orders/month, a 180-day purchaser audience should accumulate ~2,940 users — far above the 1,000-user threshold. The pools show size=20 because iOS ATT is blocking the purchase events from reaching Meta's Custom Audience pipeline. CAPI server-side events bypass this blockage and would simultaneously fix attribution reporting AND rebuild the pools.
Section 4: H5 — Placement Concentration (Budget Broadening)
87% of all purchases come from Facebook Feed. The account is spending across 20+ placements, most generating zero conversions for months at a time. ROAS median = 0.0 from May-2025 onward (more than half of placement-month rows show zero purchases).
Immediate action available today: Audience Network has generated near-zero purchases across all months in the dataset. Pausing it reduces wasted CPM spend immediately. Facebook Search (CPA=245 KES) is the most efficient placement — consider increasing its budget allocation.
Section 5: External Benchmarks — NairoMarket vs Kenya Market
NairoMarket CPM is 2.4x the Kenya e-commerce benchmark. Kenya e-commerce vertical typical CPM is 145 KES (pix-vu, Apr-2026). NairoMarket Apr-2026 = 351 KES. This is 1.5x above even the upper band for all Kenya industries (240 KES). The inflation is account-specific, not market-wide — Kenya total digital ad spend contracted 48.3% YoY in Q4-2025 (CAK data). Fewer local KES are chasing inventory, not more.
Section 6: H1 — Creative Frequency (Confirmed Partial, Upgraded from Indeterminate)
W3-bis ad-level reach/frequency data (273 rows, 97% non-zero reach) empirically closes the frequency gap that the critic flagged. 5 ads exceeded Meta's 3.5 alarm threshold. Account-wide weighted average frequency hit 4.077 in Mar-2026 — above alarm at the account level.
Top Ads by Max Frequency — Ads Approaching or Crossing 3.5 Alarm Threshold
Mar-2026 alarm: account-wide frequency = 4.077. Meta's threshold for creative fatigue alarm is 3.5. The account exceeded this at the aggregate level in Jan–Mar 2026, directly overlapping the Mar-2026 CPM step (+14% MoM). This upgrades H1 from INDETERMINATE to CONFIRMED (partial) — frequency saturation is a contributing CPM driver in Q1-2026, driven by the same audience pool shrinkage that H4 identified. H1 and H4 are mechanistically linked.
Section 7: Hypothesis Verdicts (Updated)
H1-H5 Verdicts + Action Map (Updated Apr-2026)
Section 8: Founder Decision — ACT (not wait, not platform-fix-only)
Wait vs Act vs Platform-Fix Decision Matrix
30-Day Action Plan
30-Day CPM Recovery Action Plan
Estimated total monthly savings (Days 30+): 200-220K KES (~USD 1,700) from placement pruning + spend reduction. ROAS recovery projection: from current 1.11 toward 1.5-1.8 within 60 days assuming CAPI attribution recovery of 15-25%.
Confidence caveat (Taleb): The CAPI intervention is antifragile — it improves signal quality regardless of whether H3 (iOS) or channel diversification explains the ratio decline. The 30-day verification step distinguishes the two scenarios.
Section 9: Critic Overturn Summary (Taleb)
| Analyst Claim | Original | Critic Adjustment | Post W3-bis + Tavily |
|---|---|---|---|
| H1 REJECTED (creative fatigue) | REJECTED | INDETERMINATE (frequency gap) | CONFIRMED partial — 5 ads >3.5, account avg 4.077 in Mar-2026 (W3-bis) |
| H2 INDETERMINATE (auction competition) | INDETERMINATE | No change | SPLIT: H2a Kenya-advertiser REJECTED (CAK -48.3% YoY); H2b platform-mechanic CONFIRMED (Advantage+ 3x + Mar-2026 algo rollout) |
| H3 CONFIRMED at High confidence | High | Med-High — organic channel growth confounder | Unchanged — Jan-12-2026 attribution window deprecation corroborates discrete event (MD-R-G3-004) |
| H4 independent cause | High | Med — likely consequence of H3, not independent | Corroborated — reach efficiency fell 63% from Mar-2025 to Mar-2026 (W3-bis) |
| CPM is smooth trend | Implied smooth | Piecewise step function — Jan-Mar 2025 discrete regime change | Unchanged — two additional discrete events confirmed: Jan-12-2026 + Mar-2026 algo rollout |
| Spend reduction as first action | Priority 3 | Reordered — placements first, budget reduction only AFTER CAPI live | Unchanged |
| Core structural verdict (ACT) | ACT | CONFIRMED — antifragile interventions | Unchanged |
Overturn rate (post all inputs): H1 upgraded, H2 split-resolved, H4 corroborated. Core recommendation unchanged.
Data: W3 placement_insights (5,128 rows), W4 demographic_insights (6,664 rows), W8 audiences (45), Shopify orders (8,993), W3-bis ad_reach_frequency (273 rows, 97% non-zero reach). Period: Jan-2025 to Apr-2026. External benchmarks: MD-R-G3-001..006 (Tavily, 2026-04-24). Run: data/agent_runs/2026-04-24_1045_analyst_g3_cpm_inflation_rootcause