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 · 5.3

Illustrated torii shrine in violet linked by a bridge to a tiered pagoda, with watermills feeding the river below a golden moon.

The weekly optimization routine

Updated August 2026 6 min read
THIRTY MINUTES, EVERY WEEK, IN THIS ORDER 1 · READ 10 min · four numbers, four weeks 2 · CHECK WHAT MOVED 5 min · updates you did not make 3 · CHOOSE ONE 5 min · from the queue, not the mood 4 · SHIP IT AND LOG IT 10 min · date, expectation, done THE QUEUE ← this week ← still waiting, and that is fine THE ALTERNATIVE IS FIVE CHANGES IN THREE DAYS AND ONE NUMBER NOBODY CAN READ.

The queue is allowed to be as long as it likes. The constraint that makes the routine work is that the week only ever removes one item from it.

The short answer

Thirty minutes, same time every week, four blocks. Ten minutes reading four numbers on a same-weekday comparison. Five minutes checking what changed in your store without being asked: app updates, theme updates, supplier prices, stock. Five minutes taking exactly one item off a written queue. Ten minutes shipping it and logging it with the date and what you expect. The routine's entire value is that it replaces reacting to a bad week with a cadence that keeps producing findings whether the week was good or bad.

What the routine replaces

Most small stores run on a different cycle, and it feels like diligence.

Revenue dips. Panic. Five changes ship in three days. Revenue recovers, or does not. Nothing is learned either way, because the week produced one uninterpretable number, per chapter 5.2. Then a quiet fortnight where nothing is looked at because nothing is wrong.

That cycle has two failures. It only produces work when something breaks, and the work it produces is unattributable. A routine fixes both by decoupling effort from panic.

A routine that only runs when something is wrong is not a routine. It is a smoke alarm. The Dropshipping Playbook

The four blocks

Block one: read, ten minutes

The four numbers from chapter 5.1, on a same-weekday four-week comparison. Contribution per order, blended acquisition cost, repeat rate, and cash against committed spend. Plus one discovery number, usually zero-result search rate.

You are looking for a direction, not a value. A single week is rarely meaningful; a metric that has moved the same way four weeks running always is.

Block two: what changed without asking you, five minutes

Almost no store runs this block, and it comes straight from the silent-changes table in chapter 3.1.

Anything found here goes into the change log, marked as something you did not choose. Unlogged, it becomes your decision by default.

Block three: choose one, five minutes

From a written queue, ordered by expected value, not from whatever is annoying you today. The queue is fed by the audit in chapter 3.7 and by the failures block two surfaces.

One constraint makes the routine work: one item. Not two small ones. Not one plus a quick fix. One.

Block four: ship it and log it, ten minutes

Make the change, or schedule the afternoon it needs. Write the log row first, including what you expect to happen, per chapter 5.2. Then leave it alone for a fortnight.

Building the queue

An ordered queue is what stops the routine degenerating into fixing whatever was mentioned most recently. Ordering by expected value beats ordering by effort, but effort is a legitimate tie-break.

PrioritySourceTypical items
HighestAnything broken, from block twoA flow not sending, a sort reset, a dead variant
HighFree fixes from the auditDelivery window on product pages, guest checkout, policy pages
MediumDiscovery failures with a countTop zero-result queries, empty filter combinations
MediumOrder-value leversThreshold, cart contents, one bundle
LowestAesthetic changesAnything nobody specifically complained about

The bottom row is deliberately blunt. Design work is the most satisfying item on any queue and the hardest to attribute. It is where a routine goes to die.

What sits above the week

Two longer loops keep the weekly one honest.

Monthly, one hour
Run the full audit from chapter 3.7 and refill the queue. Also fill in the "what happened" column for anything logged a fortnight or more ago.
Quarterly, half a day
Re-check the things that only break slowly: margins against supplier prices, break-even ROAS, whether the free-shipping threshold still sits above your current AOV.
After any theme or app change
Run block two immediately rather than waiting for the week. This is when silent changes cluster.
When revenue moves sharply
Stop the routine and run the 30-minute diagnostic in chapter 3.1. Optimising a store that is broken is wasted work.

Two honest limitations

Under 500 sessions a month, block one will not tell you much. Run it anyway, because the history is what makes month four readable, but make changes on judgement and expect the evidence to arrive late. Chapter 2.1 covers why.

Block two is only as good as what is being watched. Several of its items, particularly anything in the discovery layer, are events rather than reports, and chapter 3.4 covers why nothing reconstructs them afterwards. If nothing is recording, block two is a partial check and should be written down as such rather than ticked.

Common questions

Thirty minutes seems too short to optimise anything.
It is not the working time, it is the deciding time. The change itself might take an afternoon; what the half hour buys is that the afternoon goes into the right thing and gets logged. Most stores have the opposite ratio: hours of work chosen in five seconds of frustration.
What if nothing changed this week?
Then the routine took ten minutes and you have a fourth data point on four metrics. That is the normal outcome most weeks, and it is what makes the abnormal weeks legible when they arrive. A routine that only runs when something is wrong is not a routine.
Should I do this daily instead?
No. Under 500 sessions a month, daily numbers are noise, per chapter 2.1, and daily checking produces the reactive multi-change weeks that make attribution impossible. Check daily for breakage, which is a different activity, and read numbers weekly.
How do I stop the queue getting overwhelming?
You do not, and you should stop trying. A long queue is a healthy sign that you can see more opportunities than you have weeks. What matters is that it is written down and ordered by expected value, so the one item you take is the best one available rather than the most recent annoyance.
Illustrated violet dunes with a yellow sun low over the ridge.

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