the calendar takes care of volume. margin is whatever survives the discounting, and that comes down to how big the basket was when the traffic hit. a point off the price is margin gone. twelve dollars more in the cart isn’t. bundles, thresholds and gift ladders need a decision rather than an experiment — so they still ship before the freeze, and they keep paying right through december.
conversion rate is the number the category organised itself around, so it’s the one with a dashboard, a weekly meeting and someone’s name against it. average order value moves the same arithmetic and usually sits between merchandising, email and whoever last touched the cart. that isn’t neglect — it’s just how the category grew up. it does mean there’s room in it.
— patterns, not one store. every one of these has shown up more than once.
same traffic, same 2.5%. one sells a $40 single, the other a $95 bundle. conversion rate alone can’t tell them apart, so it points both at the same work.
conversion up, order value down, revenue flat. it happens when a change nudges people toward the cheapest thing on the page. the test was read correctly — it just wasn’t measuring the whole outcome.
it sits between merchandising, email and whoever set up the cart. the half with the least competition for attention is also the half with no name against it — which is usually why there’s slack in it.
a bundle launches, basket size jumps, and a slower, more considered purchase takes its place. judged on rate alone, the best month of the year looks like a loss.
this is not a forecast. it’s the arithmetic of your own store: revenue per visitor is conversion rate times average order value, so a gain on each compounds rather than adds. set the two levers and the panel shows what each one is worth on its own, and what they’re worth together.
checkout, product page, trust, speed. measured on sessions — the bigger, noisier denominator.
bundles, cart thresholds, gift ladders, merchandising order. measured on orders, and it lands in the p&l the month it ships.
what one visit is worth once both halves have moved.
on the traffic you already pay to acquire, with no increase in spend.
— the first two slices are what each lever is worth alone. the third is the part that exists only because both moved — real, and smaller than people expect. the point isn’t that slice — it’s the gap between working one lever and working two.
the second lever isn’t harder to move. it just never had anyone assigned to it.
— why patch works both
most of the order-value surface needs no experiment to authorise — it needs someone to decide, build it properly and declare what it will be judged against. the conversion surface is where testing earns its fee, when your traffic can carry one.
built, priced, photographed and still sitting in draft is the most common finding we have.
when the bar and the shipping rates disagree, every cart pays for it.
set at the basket sizes your orders actually cluster around, not at a round number someone liked.
routing, not redesign. the budget going to the weaker of two product pages is a free fix.
the one surface with a denominator big enough to test properly on low traffic.
moves what people find and what they put in the basket. on a supplement store this one moved revenue per visitor 48% and order value 32% in a ten-day test.
— the 48% figure is graded directional: click-through was the only one of its figures to reach significance, and the menu shipped alongside other changes in the same wave. the case page says so at length.
no term and no notice period. the first wins ship in week one — not after a month of analysis — because most of what we would otherwise test has already been tested and logged. ninety days is what we recommend, because that is how long the order-value half takes to read cleanly against a baseline, and you can stop at the end of any month of it.
revenue per visitor, conversion, order value and margin per order for the ninety days prior, agreed before a single change goes live. this is the number everything is judged against, and it is signed off before we can be tempted by a better one. it takes days, not weeks.
the changes that need neither a decision meeting nor an experiment — a bundle that is built but unpublished, a free-shipping bar that disagrees with the shipping rates, an ad pointed at the weaker of two product pages. these come straight off the quick wins library, so they go live the same week the baseline is signed.
both halves, with what each finding is worth a month and an evidence grade against it. it arrives after the first wins are already live, not instead of them. yours whether or not you continue.
bundles, thresholds, ladders, merchandising order. the work that needs a decision rather than an experiment. guardrails hold throughout: conversion can’t fall, refunds can’t rise, margin per order can’t drop.
shipped where it’s plainly broken, tested where your traffic can carry a test — cart, checkout, email. we’ll tell you which of the two a given question is, and why.
revenue per visitor against the declared baseline, split by half, with a grade on every figure. if it didn’t move, the readout says that, and you don’t renew.
the same five cases, read the way this programme reports: which half of the equation did the work, and how much weight each figure carries.
revenue per visitor, with order value up 32% and click-through up 13%. click-through is the only one that reached significance, and the menu shipped alongside other changes — the case page holds both caveats.
read the case → verified · before/afterconversion more than doubled, cart-to-checkout up 208%, monthly revenue €20k to €50k. one of the four changes was a price cut; the case page says so up front.
read the case → verifiedvisitors per variant, and a full test programme ran on it anyway. four tests, six weeks, three shipped. the case that proves the method rather than a number.
read the case →published, because you should be able to work out whether this is worth your time without booking anything. the teardown is free, the sprint has a fixed end, and the founder rate exists for brands where the money is tighter than the opportunity — it doesn’t cost less, it costs later.
— there is no term on any line and no notice to serve. we recommend giving it about ninety days, because that’s how long the order-value half takes to show up cleanly against a baseline — at the standard rate that’s the sprint plus three months, $15,000 — but it’s a recommendation, not a commitment, and you can stop at the end of any month.
the teardown reads both halves and prices every finding. it takes five working days and read-only access to your analytics, and you keep it either way.
it can be, and that version stops working by month three. what we do is make the basket people already wanted easier to assemble — publishing the bundle that exists, putting the threshold where orders actually cluster, ordering the range by the problem rather than the supplier. margin per order is a guardrail throughout, so a gain that comes from discounting shows up as a failure rather than a win.
then the teardown says so and the ninety days goes to the conversion half instead. single-sku stores and strictly one-per-customer categories are real. we would rather tell you that in week one than charge you for a lever that isn’t there.
because at most dtc traffic levels a site-wide test can’t close, and because half the surface we work doesn’t need one. where your traffic can carry a test we run one and say so. where it can’t, the change ships against the declared baseline and is measured at the business level.
no. no minimum, no notice period, nothing to serve out. we recommend about ninety days because the order-value half needs a few cycles to read cleanly against a baseline, and a month is not long enough to tell a real gain from a good week — but that is a recommendation and you can stop at the end of any month. if the day-90 readout is good you carry on; if it isn’t, you have a document saying why. we would rather lose a renewal than argue about one.