Skip to content
Chandler Nguyen
← All worked examples

GreenLeaf

Fictional company · synthetic data

Phase 4 of 6

Improve the campaign

Completion fell. What should the team change, and what is still unknown?

Campaign phases

Separate synthetic September cohorts show a decline but do not identify its cause; they are not results of launching this proposed plan. Compare each group after its full seven-day period to decide what to investigate.

The work · Dated signal, diagnosis and decision trail

What the team knew, and when

The comparison is retrospective. Compare the groups after both seven-day observation periods have ended.

Quiz completion fell from 45% to 28% between the two September groups after their seven-day observation periods.

Paid mobile · two September groups, each observed for seven days

Week 1 ·
45%3,780 / 8,400

Week 1: 3,780 of 8,400 paid-mobile starts completed.

Seven-day observation window. Cohort is a retained-journey count.
Week 2 ·
28%2,352 / 8,400

Week 2: 2,352 of 8,400 paid-mobile starts completed.

Seven-day observation window. Both fictional cohorts were built using the same rules.
Change
-17 percentage points
Fixed mix
35.33%descriptive, not causal

Paid-mobile groups cover September 1–14, 2026. Both seven-day observation periods ended before the comparison on September 21, 23:59:59 UTC; the figures were recorded September 22.

Fixed-mix check

Across 12 mobile channel × audience × creative strata, Week 2 is 35.33% using Week 1 weights, versus 28.0% observed. The gap shows composition contributes to the observed/fixed-mix difference; all 12 stratum rates also fell from Week 1. This is a descriptive decomposition, not a causal estimate.

Source: Checked campaign data; Week 1 weights for the 12 channel × audience × creative mobile strata.

September 21, 23:59:59 UTC
First comparison after Week 2’s seven-day observation period.Both seven-day observation periods had ended; the figures were recorded on September 22.
Decision
Investigate audience mix and the mobile journey before moving budget.Cause remains unresolved; no early alert is claimed.

From signal to bounded next review

Newly authored decision illustration using the existing synthetic cohort evidence; not a recorded AI response or human session.

Checked signal
After both seven-day windows, paid-mobile completion was 45% (3,780/8,400) for Sep 1–7 and 28% (2,352/8,400) for Sep 8–14: down 17 percentage points. First comparable at Sep 21, 23:59:59 UTC.
Proposed recommendation
Investigate audience mix and reproduce the mobile journey and event capture before moving budget. The counts show a change, not its cause.

Human disposition

Defer budget and creative changes. Accept investigation as the next step, not as a confirmed diagnosis. A human review disposition is illustrated here; none is recorded in the source pack.

Bounded next evidence

For the next review, reconcile audience mix, compare the cohorts’ page/build release history, and test one consented mobile journey against quiz_start and quiz_complete capture and deduplication. Reassess after those checks; no alert or targeting change is claimed.

Why this choice

The decline is observed; the cause is not. A change in audience mix and a defect in the mobile journey can both lower completion, and this dataset cannot separate them.

Next step

The diagnostic question, comparable cohorts and evidence requests go to reporting.

Inspect further: Evidence that would change the recommendation

The team should name this before looking at another chart.

  • Version-assignment records showing a different quiz build in Week 2.
  • Event capture that missed completions on one mobile browser.
  • A creative or audience change large enough to explain the mix shift.
  • A reproduced journey that fails before the completion event.
Full reference: Monitoring and decision log

Record what changed, where the figures came from and when both seven-day periods ended. The monitoring steps below are a proposal; they did not happen in this example.

  • Proposed early monitor: launch day through hour 48, check delivery, consented event capture and duplicate journeys at the start and end of each shift. Owner: growth lead with analytics support. Stop and investigate if events fail reconciliation; this example does not claim an alert fired.
  • September 1–7 and September 8–14: compare each paid-mobile cohort after its seven-day window. Week 2 is lower; audience mix and mobile journey remain competing explanations.
  • The first full comparison was possible September 21, 23:59:59 UTC. Before changing budget, compare audience composition and the mobile journey, and confirm which page version visitors saw.