Every plan we write opens one lever first. Not all four, and not the one the store asked about on the enquiry form. Choosing which one is the most consequential judgement in the document, because it sets what the first month is spent on and what the second month can build on.
The naive rule is to pick the weakest number. That is a start and it is not enough, because weakest is not the same as most fixable. A store can have a poor repeat rate because it sells something people genuinely buy once, in which case the number is a fact rather than a leak. It can have low traffic and a store that would waste more of it.
So we ask two questions of every lever, not one. Is this number holding the business back, and is there a fix here that we can build, measure and read before the next report. A lever only goes first when both answers are yes.
Each lever has an entry condition
Before ranking anything, we check whether each lever is even available to work on, because some are not, and starting on one that is not is how a month gets wasted.
Visitors needs a store that can hold the people you send it. If a known share of visitors is lost at a specific step, more visitors means paying twice for the same loss. The entry condition is that the path from landing to payment is not obviously broken.
Conversion needs enough traffic for a change to produce a readable signal. A store with very few sessions a week can still improve its pages, and it should, but it will not be able to prove the improvement quickly, which changes how the work is sequenced.
Order value needs a catalogue with adjacency. Things that go together, sizes that scale, consumables that come in multiples. A single-product store with no accessories has less to work with, though quantity and subscription usually still apply.
Repeat purchases needs a product with a reason to come back and enough customers to see a pattern. If your customers reorder yearly, the flow still gets built early, but it will not report early.
Weakest is not the same as most fixable, and a plan has to answer both.
The question we ask of each number
Once we know which levers are available, we interrogate each with a specific question rather than a general sense that it looks low.
For visitors: where do people come from, what does each source cost, and what happens to this business if the biggest source gets more expensive or disappears.For conversion: at which step do people leave, and is the drop concentrated on mobile. A store-wide conversion rate is an average across very different journeys, and the useful version is the funnel split by device and by source.
For order value: what does the distribution of order sizes look like, how many orders contain one item, and where does margin sit across the catalogue. Average order value with no distribution behind it hides everything interesting.
For repeat purchases: what share of customers ever order twice, and how long is the gap. Then the harder one: is there anything in the product that gives them a reason.
Those four questions produce evidence. The lever with the clearest evidence usually goes first, because a fix built on evidence can be judged.
Dependency usually beats size
The levers are not independent, and that changes the order more than the size of any single gap does.
Conversion sits underneath everything. Improving it raises the value of every visitor from every source, including the ones you have already paid for and the ones you have not bought yet. Fix it first and every later traffic pound buys more.
Order value sits underneath acquisition economics. If each order is worth more, the amount you can afford to pay for a customer goes up, which can turn an unaffordable channel into an affordable one without touching the campaigns.
Repeat purchases sit underneath everything on a longer timescale. A store where customers come back can pay more to acquire them than a store where they do not, and it can weather a rise in advertising costs that would end a business dependent on first orders.
Visitors is the lever that multiplies whatever the others are worth. That is exactly why it usually goes later, and why sending more traffic into an unfixed store is the most expensive mistake in this list.
When we start with visitors anyway
There are real cases where traffic goes first, and pretending otherwise would be a rule applied for its own sake.
The first is a store that genuinely converts well and simply has too few people. If the pages answer the questions, the checkout is short, the flows exist and the numbers say people buy when they arrive, then the constraint is arrivals.
The second is an imminent loss. A channel that is about to become unavailable, a feed that is disapproved and taking the whole shopping presence down with it, a policy change in a restricted category. Those are fires, and fires get put out first regardless of the ranking.
The third is a window. A seasonal peak, a launch, a retail moment that only comes once a year. Work that has to be in place before a date moves up the order, and the plan says which date is driving it.
The fourth is when the traffic problem is a measurement problem. If tracking is wrong, every other judgement is built on bad numbers, so fixing what you can see comes before deciding what to change.
When the data is too thin to choose
Small stores and new stores present a real problem: not enough orders for any comparison to mean much. The temptation is to choose confidently anyway, since confidence sells.
What we do instead is choose the lever with the cheapest evidence. If we cannot tell from the numbers where the loss is, we look for it qualitatively: walk the store on a phone, read the support inbox for pre-purchase questions, check what happens after an order, look at the feed and what searches it is showing for.
Those sources do not need volume. A product page that fails to state delivery cost is a problem whether you get forty visitors a day or four thousand.
Then we pick fixes that are cheap and reversible, and we accept a longer measurement window rather than a shorter, noisier one. If a change needs a quarter to prove itself at your volume, the plan says a quarter.
One lever at a time, for a boring reason
We open one lever first because two levers opened together cannot be told apart afterwards.
If conversion work and a campaign restructure land in the same fortnight and revenue rises, nobody can say which one did it, and the next decision is a guess. Sequencing costs a little speed and buys the ability to learn, which is what the whole method is built on.
There is an exception for work with a long lead time. Search work and repeat purchase flows both mature slowly, so they start early and run in the background while faster work happens in front of them. That is not really two levers at once, since they report on completely different timescales and different numbers.
So the plan opens one, names the second, and says what would have to be true for the second to move ahead of the first.
The choice is written down so it can be wrong out loud
The reason all of this appears on page one and two of the plan, rather than in a conversation, is that a written choice can be checked later.
The plan states which lever is weakest, what evidence points there, what we will change, and what would count as it having worked. A month later the report either shows that or it does not. If it does not, we have a specific thing to revisit rather than a general sense that things are slower than hoped.
Being wrong in public is the cost of working this way, and it is worth paying. A diagnosis that was never written down cannot be tested, and an agency whose reasoning is never recorded can always claim to have meant something else.
It also lets you disagree at the start. You know things about your store that no outside reading can find: a supplier problem, a seasonal quirk, a wholesale order that distorted a quarter. If our choice of lever ignores something like that, page one is where you will see it, before any money has changed hands.