[ Phase 05 — Iterate ]

Compounding lift comes from killing losers fast.

Most agencies sit on tests for 90 days because killing them looks like failure. Ours conclude in 14 days. Wins are promoted to V2/V3 (where the lift compounds). Losses are documented and inform the next round of hypotheses. The result is 4–8% efficiency lift every month — compounding.

Ongoing — 14-day test windows Tests launch & conclude on a 14-day rhythm Promoted winners + killed losers + next-cycle hypotheses Owner · Lead operator + practice specialists

Why iteration is the product

The first cycle of audit-plan-execute-measure produces an initial lift. The lift after that — quarter over quarter, year over year — comes from disciplined iteration. Tests with kill dates. Winners V2'd. Losers killed without ego. Hypotheses generated continuously. The discipline IS the product.

[ Artefacts ]

What this step produces.

I01 · 14-day test windows Live in experiment register

Every test runs for max 14 days unless statistical significance dictates otherwise. Wins promoted to control. Losers killed. Inconclusive tests revised or abandoned. No 90-day zombie tests.

I02 · Winner V2/V3 register Notion · live

Every winning test gets a V2 brief within 7 days. V2 typically lifts another 8–12% beyond V1. V3 follows V2. Winners that no longer compound are retired. Compounding lift is logged monthly.

I03 · Loss postmortem .md · per loss

Every killed test gets a 1-paragraph postmortem: hypothesis, why it lost, what we learned, how it informs the next hypothesis. Archived for cross-engagement learnings.

I04 · Quarterly compound report .pdf · quarterly

Cumulative monthly lift since engagement start. Channel-by-channel breakdown of compounding wins. Total efficiency improvement vs baseline. Tied to business outcomes — net revenue, payback period, contribution margin.

I05 · Cross-engagement learnings library Internal · referenced in QBRs

Anonymised learnings from across our portfolio applied to your engagement. Hypotheses we've already tested elsewhere skip your test queue. Gives every client the benefit of the whole portfolio's discipline.

[ Timeline ]

How the days run.

Day 0

Test launch

Hypothesis, primary metric, sample size, kill date, success criteria — all locked before launch. Logged in experiment register. Stakeholders notified in Slack.

Days 1–14

Test runs

Daily monitoring for anomalies. Sample size and significance tracked. No mid-test changes unless break-it-fix-it conditions hit.

Day 14

Conclude or kill

Test concludes with a clear verdict: win, loss, or inconclusive. If win, promote to control + brief V2 within 7 days. If loss, postmortem within 48 hours. If inconclusive, revise or abandon — never extend mindlessly.

Days 14–21

V2 brief or postmortem

Winners get V2 hypothesis briefed and queued. Losers get postmortem written and archived. Cycle continues.

Quarterly

Compound review

Cumulative monthly lift reviewed in QBR. Channel-by-channel. Tied to net revenue and payback. Channel mix recalibrated based on which channels compound fastest.

[ Who does what ]

Named owners, not "the team".

RoleLeadsContributesConsumes
Lead operatorTest register stewardship + winner V2Loss postmortems, compound reportsPractice specialists' test design
Practice specialistsChannel-specific test designHypothesis generation, brief writingLead operator's prioritisation
Senior reviewerCross-engagement learnings libraryQuality control on V2 briefsAll clients' loss postmortems
Founder / leadershipStrategic priority callsSign-off on bigger swingsQuarterly compound report
[ Rules ]

Non-negotiables for this step.

  • Tests conclude in 14 days. Period.
  • Wins get V2'd within 7 days. No exceptions.
  • Losses get postmortems within 48 hours.
  • No 90-day zombie tests.
  • Hypotheses pulled from cross-engagement library before being generated from scratch.
  • Compounding lift tracked monthly. If not compounding, reroot.
DTC · Beauty · $11M GMV
The 14-day rhythm changed how we thought about testing. Killing tests at day 14 felt aggressive — until we saw the compounding effect three months in. We stopped sitting on losers.

47 tests in Q1. 28 wins promoted, 14 killed, 5 inconclusive. Cumulative paid efficiency +52%.