MadeToMargin+166% managed sales growth vs +48% in the holdout, at the same efficiency.
MadeToMargin founding account · Baseline edition (data through Oct 9, 2026 · final holdout verdict Dec 1, 2026)
Managed keywords grew sales 166%. The holdout grew 48%.
During back-to-school season, at essentially the same cost efficiency, on a randomized 10% holdout of the same account.
| July (before) | Aug 1 – Sep 26 (managed period) | Change | |
|---|---|---|---|
| Managed keywords, sales/day | $1,902 | $5,065 | +166% |
| Holdout keywords, sales/day | $360 | $531 | +48% |
| Managed ACOS | 28.3% | 30.6% | +2.3 pts |
| Holdout ACOS | 24.5% | 27.3% | +2.8 pts |
Ad spend on the managed side scaled ~2.9× per day through the season while efficiency held within ~2 points. Both arms rode the same seasonal wave and the same product pages. The gap between +48% and +166% is the management.
Interim read on settled attribution data; the holdout freezes through Nov 30 and the final season-long verdict publishes Dec 1, 2026.
Why the holdout matters
Every agency shows before/after charts. Before/after can't separate the agency's work from seasonality, price changes, or luck.
On day one we froze a stratified random 10.1% of the account's keywords (156 entities): no bid changes, no pauses, nothing, all season. The ledger shows zero touches across 10 weeks of daily automation. Whatever the managed side does above the holdout is the service, causally. We are not aware of another Amazon PPC provider that measures itself this way.
What the system did
Profit targets derived from real margins
Family-level ACOS targets computed from landed costs + actual Amazon fees. The account optimizes contribution profit, not a vanity ACOS. Loss-leader families are managed as a strategy, not an accident.
Back-to-school ramp, governed
A pinned keyword cohort stepped +20%/day toward spring-anchored targets with an automatic circuit breaker. Result: 1.56× impressions at 0.98× the target ACOS, scaled into the season without buying it at a loss.
A review-starved hero product turned around
Phonics flagship: conversion 8.5% → 13.6%, sessions +40%, daily bleed cut from $78/day to $13/day while deliberately growing rank (an owner-approved investment mission with a hard $50/day ceiling).
Silent money leaks found and fixed
A branded-keyword feedback loop had ratcheted branded CPC $0.52 → $1.37 over 8 weeks (43 raises); guarded and unwound. 53 harmful negatives removed, freeing 42 converting search terms. 23 forgotten 2021-era console budget rules neutralized before they could stack on live automation.
Deals handled like a calendar, not a surprise
Coupon/deal windows freeze affected bids, quarantine deal-day data so it never teaches the rules, double budgets on deal day (bleed-guarded), and restore everything exactly the next morning.
The safety story
- Hard guardrails: ±20% max bid move per day, 7-day cooldowns, daily change caps, anomaly brake (an unusual day auto-applies nothing and pings a human).
- Every write is ledgered and reversible: one command reverts any day's changes.
- 30-day propose-only trust period: the system recommends in daily reports; nothing is applied until the client arms each capability, one ratchet at a time.
- Daily transparency: a Slack report with a TLDR pulse (yesterday/7/14/30-day, top movers, flags) and an interactive dashboard; weekly recaps; 300+ automated tests behind every release.
37 ASINs, US Sponsored Products + Brands
Trailing-30-day ad-attributed sales ~$190K at ~32% blended ACOS during the seasonal peak (account's blended margin target band). Engine live since late July 2026; standing automation armed by the owner in stages July 30 – Aug 3.
Honest caveats: what we won't claim (we publish these on purpose)
- Interim numbers; the final, season-long holdout comparison lands Dec 1, 2026.
- The managed arm includes keywords the system created (harvesting is part of the service); the Dec 1 analysis will also report a matched pre-existing-keyword cohort.
- Campaign-level actions (budgets, placements) partially benefit holdout keywords too, which biases the measured lift downward, not upward.
- First-party account. The methodology is the point: the same holdout design ships with every client engagement.
Start with a free 48-hour audit.
Read-only access. A one-page findings memo. Yours to keep either way.