Can't tell why your sales dropped? Sledge reads your store data and ranks the leaks by what fixing them is worth. Free for 14 days →

// Level 05 Making It Last · 5.2

Illustrated treehouse village in violet linked by a bridge across island rocks, with waterfalls below a golden moon.

Attributing revenue to changes

Updated August 2026 7 min read
THE SAME FIVE CHANGES, TWO CALENDARS ALL AT ONCE 5 changes revenue moved. by which one? ONE RESULT. NOT REPEATABLE. NOT REVERSIBLE. ONE AT A TIME 1 2 3 4 5 change 4 was negative. revert it. FOUR RESULTS. EACH REPEATABLE. EACH REVERSIBLE. THE SLOWER CALENDAR IS THE FASTER ONE, BECAUSE IT KEEPS WHAT IT LEARNS.

Illustrative. The asymmetry is the argument: one calendar produces a story, the other produces four facts and one revert.

The short answer

This is the cheapest habit in the entire playbook and the one most stores never adopt: change one thing, write down the date, wait, and compare like with like. Ship five changes in a week and revenue will move, but the movement belongs to all five, which means you cannot repeat it, cannot revert the harmful one, and have bought a result instead of a finding. A dated change log costs ten seconds per entry and is the difference between a store that compounds and one that thrashes.

The habit three chapters have already promised

Chapter 3.1 said a written change log turns step six of the 30-minute diagnostic from an hour of remembering into thirty seconds. Chapter 4.11 said sort order is unusually easy to change and unusually hard to attribute. Chapter 3.7 ended by insisting on one fix at a time.

All three were pointing here, and the underlying idea is one sentence: a result you cannot attribute is a result you cannot repeat.

Five changes in one week is not five experiments. It is one anecdote with five possible explanations. The Dropshipping Playbook

Why simultaneity is so expensive

"You cannot tell which one worked" undersells what is actually lost.

What you loseConsequence
The ability to repeat itNext month you cannot do it again on another product
The ability to revertIf one change was harmful, its damage is hidden inside a net positive
The negative findingsA change that lost money looks like a smaller win, so you keep it
The transferable lessonNothing you learned applies to the next store or the next season
The ability to diagnose laterWhen revenue moves in three months, week 12 is a blur

The third row is the quietly worst one. A batch of five changes where four help and one hurts nets out positive, so the batch is kept, and the harmful change stays in your store permanently, doing damage nobody will ever look for because the week it shipped was a good week.

The log itself

It does not need a tool. A spreadsheet, a text file, or a note in whatever you already open every day. Five columns:

ColumnExampleWhy it earns its place
Date2026-08-26The whole point. Everything else is comparison against it
What changedFree-shipping threshold $50 to $65Specific enough to revert exactly
WhereShipping settings, all regionsSo a later reset is detectable
Expected effectAOV up, cart abandonment flatWritten before the result. This is the important one
What happenedFilled in two weeks laterTurns a log into a record of judgement

Column four is the one people skip and the one that does the work. Writing what you expect before you see the outcome is what prevents the universal habit of deciding, afterwards, that whatever happened is what you were going for. It also makes a surprise legible: an unexpected result is only visible as a surprise if the expectation was written down.

Log what you did not choose, too

App auto-updates, theme updates and supplier price changes all move revenue and none of them are your decision. Chapter 3.1 lists them among the changes that never notify you. Add them to the log when you notice them, because in three months they will be indistinguishable from things you did on purpose.

Comparing honestly

The log tells you when. Reading the result correctly is a separate discipline, and there are three ways it usually goes wrong.

Why not just A/B test

Because most dropshipping stores do not have the traffic, and a badly powered test is worse than no test: it produces a confident answer from a sample that could not support one.

A split test needs enough conversions in each arm to distinguish a real difference from chance. At a 1.4% conversion rate and a few hundred sessions a week, reaching that on a modest effect takes months, during which the store cannot change anything else. That is rarely the right trade at this scale.

Sequential comparison, honestly logged, with like-for-like weekday matching, is the realistic method. It is weaker evidence than a properly powered split test and it is much stronger evidence than what most stores currently have, which is a memory.

What this looks like in practice

  1. One change per week. If that feels slow, note that the alternative is a month with no findings in it.
  2. Write the row before you make the change, including what you expect.
  3. Do not touch it for two weeks, unless something is obviously broken.
  4. Compare same-weekday, four weeks, on the metric you nominated.
  5. Fill in what happened, including when it was nothing, which is a finding and the most commonly discarded one.
  6. Revert anything negative immediately, which is only possible because you changed one thing.

That cadence is the spine of chapter 5.3, which turns it into a repeatable half-hour rather than a resolution.

Common questions

How long should I wait between changes?
Long enough to accumulate a readable sample, which for most small stores means two weeks rather than two days. Under 500 sessions a month a single change may never become readable at all, in which case make the change on judgement, log it, and accept that the evidence will come later or not at all.
Isn't this too slow when I need results now?
It is faster in the only sense that matters. Five simultaneous changes buy you one uninterpretable month. Five sequential changes buy you four facts you keep forever and can apply to the next store, the next product and the next season. The slow calendar overtakes the fast one somewhere in month three.
What if two changes have to ship together?
Then log them as one change and accept that the result is joint. That is a legitimate compromise, and it is completely different from shipping five things and telling yourself you will remember which was which.
Do I need an A/B testing tool?
Almost certainly not at dropshipping volumes. A proper split test needs far more traffic than a store doing a few hundred sessions a week can supply, and a badly powered test is worse than none because it produces confident nonsense. Sequential comparison against the same weekday, honestly logged, is the realistic method at this scale.
Illustrated violet forest at dusk, with a yellow sun between the trunks.

One number pays for the rest

What a single order is worth decides what every other fix can afford. Chapter 4.1 is where that number moves.

Go to chapter 4.1