One lever per window, and why it pays: A staging discipline for agent-driven change: one lever per measurement window, a logged boundary, and a read before the next batch is allowed to start.
Read the full article below for detailed insights and actionable strategies.
Channel comparison
Platform-reported vs. causal contribution
Platform-reported numbers double-count assists; causal inference reveals reality
Sequencing costs you a few weeks and buys you the ability to attribute anything at all. Here is the discipline.
The rule
One lever per window. Listings, or ad copy, or creative, or pricing. Not two, and never all four.
A window is long enough for the effect to show and for the noise to average out, which for most DTC catalogues means weeks rather than days.
Why this is not obvious
It feels wasteful. The agent can do all four in an afternoon, and sequencing them looks like deliberately going slower than you have to.
The trade is real, and it is worth making: four changes in one window gives you one number you cannot decompose. Four changes across four windows gives you four numbers you can act on individually. The second is worth more even though it arrives later.
The changelog is half the method
| Log | Why it matters |
|---|---|
| Timestamp of the batch | Defines the before and after boundary |
| What changed, precisely | So the next read knows what it is reading |
| What deliberately did not change | The holdout, recorded before results exist |
| Anything external in the window | Promotions, outages, seasonality, press |
An agent can write all four rows itself, at the moment it acts, which is the only moment the information is complete. Retrofitting a changelog a month later produces something that looks like a record and is not one.
Reading between windows
Before the next batch, read the last one. An estimate with a confidence interval that spans break-even means the window did not resolve, and the correct response is usually a longer window rather than a bigger change.
Stacking a second change onto an unresolved first is how a catalogue ends up in a state nobody can explain.
The read
A causal read on a Google Analytics export, with the changelog to define the boundaries, returns per channel an estimate, an interval, a coverage share, and a label for what could not be resolved. It is 99 euro once, refunded if it does not move a budget decision.
The interactive demo shows the same output on sample data, with no signup.
Related answers
Key Terms in This Article
Analytics
Analytics is the systematic computational analysis of data. It reveals customer behavior and measures campaign performance.
Attribution
Attribution identifies user actions that contribute to a desired outcome and assigns value to each. It reveals which marketing touchpoints drive conversions.
Causality
Causality is the relationship where one event directly causes another, essential for identifying specific actions that drive desired outcomes in marketing.
Confidence Interval
Confidence Interval is a statistical range of values that likely contains the true value of a metric. In marketing analytics, it quantifies uncertainty around estimates, indicating the precision of an outcome or causal effect.
Google Analytics
Google Analytics is a web analytics service that tracks and reports website traffic.
Related Articles
Sixty-second versions of these ideas: Causality Engine on YouTube Shorts.
Ready to see your real numbers?
Own the budget? Upload your GA4 export and see which channels drive incremental sales, with confidence intervals, in minutes. Have to defend it? Start with the live demo and take the read to your CFO.
Full refund if you don't see value.
Stay ahead of the attribution curve
Weekly insights on marketing attribution, incrementality testing, and data-driven growth. Written for the person who owns the budget and the person who has to defend it.
No spam. Unsubscribe anytime. We respect your data.
Frequently Asked Questions
What does one lever per window mean?
Change listings, or ad copy, or creative, or pricing in a given measurement window, but not several at once. Four changes in one window produce one number you cannot decompose; four windows produce four you can act on.
What should an agent log when it makes changes?
The batch timestamp, precisely what changed, what was deliberately left unchanged, and anything external in the window such as promotions or outages. It should write these at the moment it acts, which is the only time the record is complete.
What if a window does not resolve?
Extend the window rather than making a bigger change. Stacking a second change onto an unresolved first is how a catalogue reaches a state nobody can explain.