G2: Catalog Cash Leak — Hypothesis Tested and Mostly Overturned
⚠ Data window warning (read first): This analysis covers 2025-01 to 2025-12 only. 2026-01 through 2026-04-24 (today) is NOT in the extraction. The "Dec 2025 → Jan 2025" comparison is within 2025; whether the CPM inflation continued, stabilized, or reversed in Q1 2026 is unknown until G3 extraction runs. Treat as historical backdrop, not current-state diagnosis.
Founder context (2025 pattern): The original G2 hypothesis was that a "catalog cash leak" — OOS SKUs, Flash Sale dead sets, and DSers-sourced dropship products — was draining $182–307/week in recoverable ad spend. The Socrates critic tested all seven analyst claims against the data via elenchus. Most collapsed. The honest defensible recovery is $7–10/week from Audience Network exclusion — the one claim that survives.
The bigger story the analyst missed: Monthly purchases grew from 427 (Jan 2025) to 517 (Dec 2025). Revenue was not leaking in 2025 — CPM costs were inflating. This is an auction-cost problem, not a catalog-hygiene problem. Whether 2026 continues the same pattern requires fresh data.
Defensible Weekly Recovery (USD)
Audience Network ROAS (12-mo)
Dec 2025 Purchases
Jan 2025 Purchases (baseline)
Honest overturn: The $182–307/wk headline is struck through. Dec 2025 had MORE purchases (517) than Jan 2025 (427). If catalog leaks were compounding over 12 months, purchase volume would trend down — it trends up. The "leak" framing is not supported by the outcome data.
Section 1: The Killer Fact — Purchases Grew, Not Fell
Before evaluating the catalog hygiene claims, look at what actually happened to purchase volume. The analyst's compounding catalog leak model predicts degrading purchase volume over time. The data shows the opposite.
Jan 2025: 427 purchases. Dec 2025: 517 purchases (+21%). The catalog-leak compound-loss model is falsified by the outcome series. Revenue per purchase (AOV) held approximately flat. What changed is CPM — cost of reaching 1,000 users — which rose from 131 KES to 489 KES (+274%).
Claims: MD-C-005 (purchase volume overturns leak hypothesis), MD-D-007 (ROAS trajectory is CPM-driven, not catalog-driven)
Section 2: CPM Inflation — The Real Culprit
CPM rose uniformly across all three platforms. This is an auction-wide macroeconomic signal, not a placement-specific degradation. The analyst's IG saturation narrative (MD-D-004) required IG CPM to inflate faster than FB CPM. The data shows the opposite: Facebook inflated 3.5x, Instagram 2.9x.
MD-C-002 finding: The frequency field is zero across ALL rows in the placement dataset. The "Instagram audience saturation via frequency fatigue" mechanism claimed in MD-D-004 cannot be observed in this dataset. The IG saturation narrative has no measurable frequency signal to stand on.
Claims: MD-C-002 (IG saturation falsified — FB inflated faster), MD-D-007 (CPM is the systemic driver)
Section 3: Audience Network — The One Claim That Holds
Audience Network is the only finding that survives critic scrutiny intact. Over 12 months: $369 USD spent, 3 real purchases (excluding the Dec noise spike), ROAS 0.076 excluding December. This is a structural exclusion opportunity — not because of catalog hygiene, but because AN rewarded/classic placements are unsuitable for direct-response e-commerce at any catalog state.
Why Dec AN data is noise (MD-C-001): December showed 10 purchases on $16 USD AN spend, ROAS 20.4x. Excluding December, the remaining 11 months show 3 purchases on $343 spend, ROAS 0.076. The December cell is a statistical artifact — likely a single high-value order attributed to AN via view-through. The 0.155x "aggregate" ROAS the analyst leads with is an averaging artifact. Correct read: 0.076x structural ROAS, 10 purchases/yr excluding December noise.
Claims: MD-D-003 (AN bleed confirmed), MD-C-001 (Dec noise correction — 0.076x is the real number)
Section 4: Catalog Structure — Bloat Is Real, Leak Is Not
The catalog has structural problems worth fixing — but they are not producing measurable revenue leakage. The "Flash Sale" set has zero products but zero active creatives. The "All Products" set exposes 99K SKUs with no filter, but purchase volume held. This is a hygiene issue, not a cash drain.
Catalog Set Inventory
Claims: MD-D-001 (All Products unfiltered structure), MD-D-002 (Flash Sale 0 SKUs), MD-D-005 (Shipped From Abroad size), MD-C-003 (DSers identity unverifiable from data), MD-C-006 (Flash Sale has 0 active creatives — no spend impact today)
Section 5: Recovery Claim Waterfall — Honest Math
The analyst estimated $182–307/week in recoverable spend. The critic's decomposition shows each component collapses when tested against data. The only surviving element is Audience Network exclusion.
Waterfall — Claim-by-Claim Verdict
Claims: MD-C-004 (OOS rate is unverifiable benchmark), MD-C-006 (Flash Sale no spend impact), MD-C-005 (IG saturation is macro CPM), MD-D-003 (AN exclusion survives)
Section 6: Loop Verdict — Critic's Synthesis
The analyst proposed three compounding loops driving catalog cash leak. Each loop was tested. The table below shows which loops are real, which are unobservable, and which are falsified.
Loop-by-Loop Critic Verdicts
Claims: All MD-C-* claims — synthesis table. Treemap: area = evidential support for each loop post-critique.
Section 7: Zero-Purchase Placement Dead Zones
Two placement classes had zero purchases across their entire history. These are not catalog-leak related — they are structural mobile-first audience failures on desktop placements. Small absolute spend but clean signal.
Desktop Dead-Zone Placements (0 Purchases, All Time)
Claims: MD-D-006 (desktop dead zones confirmed)
Section 8: Action Priorities — Reframed
Given the critique results, actions are reordered by defensible evidence, not analyst estimate size.