Should you optimize for
AOV, or for conversion?

Bottom line up front: "should we push AOV or conversion rate?" is the wrong question, because you don't get to set them independently. They're paired variables. Nearly every lever that lifts one leans on the other, so the real decision is which point on the curve earns you the most per visitor. The operators who win treat the pair as one system, price their traffic like inventory, and test their way to the point where the spread between what a visitor costs and what a visitor pays is widest.

This post covers the visitor math almost nobody runs, what actually moves each metric, when each side is the better bet, why they pull against each other, and the testing method that settles the argument with data instead of opinions. For the discounting-specific version of this math, I've already covered the conversion vs margin break-even math separately.

The wrong question
most operators ask.

In almost every DTC audit I run, someone asks which metric deserves the next quarter of attention, AOV or conversion rate. The question assumes they're separate dials. They aren't. It's the ecommerce version of price and volume: raise price and volume slips, chase volume and price erodes. You're picking a point on a curve, whether you admit it or not.

The deeper problem is that most ecommerce advice states a direction without stating the curve. "Raise your AOV" is a claim about where someone else's curve bent, not a strategy. Your product weight, your traffic mix, your margin structure, and your repeat rate all move where that bend sits for you. A tactic that added $40k a month for a bagel brand can quietly drain a furniture brand running the same play, because the two businesses sit on completely different curves.

Run the math on why neither number means anything alone. A store converting 2% of visitors at a $75 AOV and a store converting 3% at $50 both earn $1.50 a visitor. Same revenue, completely different businesses: different shipping costs per revenue dollar, different repeat-purchase profiles, different exposure to a CAC increase. The metric you've been told to maximize is an input, not a goal.

Across the brands I've operated and advised, the costliest testing mistake is picking the metric first and defending it with cherry-picked wins. At WIN Brands Group we tested bundle pressure against buyer autonomy across multiple brands, and the honest scoreboard kept overruling whichever single metric someone in the room was attached to. That scoreboard is where this post ends up. But it starts one level lower, with a number most merchant dashboards never show.

What does a visitor cost,
and what do they pay?

Merchants measure ROAS by campaign and CAC by order. Almost nobody prices the session itself. Yet every store is running the same quiet arbitrage: you buy visitors at one price and monetize them at another. Cost per visitor is your total acquisition spend divided by sessions. Revenue per visitor is conversion rate times AOV. Two numbers, same unit, and they almost never appear on the same dashboard.

Blame the tooling, not laziness. The ad platforms report ROAS and CPM, and they stop at the click. Shopify reports conversion rate and AOV, and it starts at the session. The P&L lives in a spreadsheet that sees neither. Three dashboards, three owners, and the one number that connects them, what a session costs against what a session earns, belongs to nobody. So a merchant can watch ROAS hold steady, watch conversion tick up, and still lose money per visitor for two quarters before the blended math surfaces in the bank account. I've watched exactly that sequence play out in diligence more than once.

Work a realistic example. A store does 50,000 sessions a month and spends $24,000 on paid media. Blended cost per visitor: $0.48. At the 2025 medians of 2.01% conversion and $74.12 AOV, revenue per visitor is $1.49. At a 40% contribution margin, that's $0.60 of contribution per session. Subtract the $0.48 the session cost and the store clears $0.12 per visitor before fixed costs. That's the whole business, compressed into one line: a twelve-cent spread, multiplied by traffic.

Owned and organic traffic complicates the math in a useful way. Email, SMS, organic search, and direct sessions aren't free, you paid for them in content, tools, and time, but their marginal cost per visitor is near zero, which means their spread is structurally wider. That has two implications. First, blended cost per visitor understates how much your paid traffic really costs, so calculate the paid-only version too. Second, the more of your session mix is owned, the more an on-site win compounds: a conversion lift applies to traffic you'll keep getting without paying again. Stores with strong owned-traffic engines get more out of every test they win, which is its own argument for building the list before scaling the spend.

Now connect it to acquisition, because this is where the two metrics stop being interchangeable. CAC is just cost per visitor divided by conversion rate. At $0.48 and 2.01%, this store pays $23.88 for a customer. Lift conversion 20% and CAC falls to about $19.90 without touching a single ad. Lift AOV 20% and CAC doesn't move at all: you paid the same for the customer, they just handed you more revenue once they arrived. Same RPV gain on paper, completely different effect on the P&L, and which one you need depends on which constraint is binding. If CAC payback is your problem, conversion is the lever that reaches it; the bands by category are in CAC payback benchmarks by vertical. If per-order economics are the problem, AOV is the lever that reaches those.

If you track one new number after reading this, track the spread: contribution per visitor minus cost per visitor, monthly, blended across channels. It reframes every test you'll ever run. A variant doesn't win because conversion went up or because AOV went up. It wins because the spread widened. And it reframes traffic buying too: a channel where cost per visitor exceeds contribution per visitor is renting you revenue, no matter what the platform's ROAS screen says. Your ceiling on the buying side is your max allowable CAC; the spread is the same discipline applied per session instead of per order.

What actually moves
average order value?

AOV is set by offer architecture, not persuasion. It decomposes into two factors, items per order and average item price, and every AOV strategy is a bet on one of them. Attach strategies (bundles, cross-sells, volume breaks) move the first. Price strategies (premiumization, tiers, billing periods) move the second. Customers spend more when the menu makes spending more the natural choice, and they resist when the menu makes it feel like a toll.

That framing matters because it tells you where AOV work actually happens: in the catalog and the offer, not in the checkout. By the time a shopper reaches the cart, most of the AOV decision is already made by what you sell, at what prices, in what quantities. The cart can only nudge the remainder. Brands that try to extract AOV entirely at the finish line, with pop-ups and forced add-ons, are working the smallest part of the problem with the most conversion-hostile tools available.

The strategy menu, roughly in order of how much conversion each lever risks:

• Post-purchase upsells. A one-click offer after payment can't lose you the original order, which makes this the only AOV lever with no conversion tax. It's also the cheapest test on this list to prove, because the downside is bounded at zero. If you're not running one, start here before touching anything upstream.
• Billing structure. Longer billing periods raise order size without changing the product, and they pull cash forward while cutting fulfillment frequency. IM8 pushed new subscribers from monthly to quarterly plans and new-customer AOV jumped 53% in a single quarter, a move I broke down in the IM8 growth playbook. For any subscription or replenishment brand, this is the highest-leverage AOV play available, and most brands never test it.
• Premiumization. Good-better-best tiers, premium SKUs, and anchor pricing raise AOV through mix rather than pressure. The customer picks the bigger number themselves, which is why the conversion risk stays low as long as the entry tier stays visible and viable. Kill the entry tier and you've converted premiumization into a forced minimum, which is a different lever with a different tax.
• Volume discounts. In a published VWO test, office-supply retailer Paperstone added bulk discounts and lifted AOV 18.94% and revenue 16.8% (VWO, ecommerce metrics analysis). The precondition is a product people genuinely consume in multiples. Bulk-discounting a product nobody needs two of just discounts the product.
• Cart and checkout cross-sells. The add-a-battery, add-a-gift-wrap, complete-the-set slot. Done as a single relevant suggestion in the cart, it's a low-tax attach lever; done as a stack of pop-ups between the customer and the pay button, it migrates up this list into the pre-purchase-friction category. The dose makes the poison.
• Free-shipping thresholds. The classic nudge, and the first place AOV strategy starts taxing conversion. Set just above your natural order band, the threshold pads baskets with add-ons. Set too far above it, and shoppers abandon instead of adding. The threshold is also the single most testable number in your store, which makes it strange how many brands set it once at launch and never touch it again.
• Preset bundles and minimums. The bluntest instrument. They raise the average by removing the smaller option, which works only when the customer wanted that much product anyway. When they didn't, the lost orders show up in someone else's metric, which is exactly how the seesaw stays hidden from anyone watching only AOV.

Two cautions that rarely make the strategy deck. First, AOV won through pressure can tax the second order, not just the first. A customer who felt forced into a bigger box than they needed churns quieter and sooner, which is why any AOV push should get checked against the LTV math brands keep getting wrong a quarter later. Second, bigger baskets carry more return risk in try-on categories, and a returned three-item order costs more to unwind than a kept one-item order ever earned; the full cost of a DTC return is the sanity check there.

What actually moves
conversion rate?

Conversion rate is set by everything that happens between intent and payment: friction, trust, clarity, speed, and how choices are framed. But "conversion rate" is four rates multiplied together, and diagnosing which one is broken matters more than any tactic. A session has to find the product, the product page has to earn an add-to-cart, the cart has to reach checkout, and checkout has to complete. Each stage leaks separately.

The benchmark skeleton, from the same 2025 dataset: the paid-channel conversion median was 2.01% across 33,000+ brands, apparel add-to-cart rates ran 6.33 to 7.12%, and roughly 70% of carts were abandoned (Triple Whale, Ecommerce Benchmarks Report, 2025). The vertical spread is enormous, food and beverage converts around 2.7% while luxury sits near 0.9%, and the device gap is bigger still: desktop converts at 3.9% against mobile's 1.8%, with mobile carrying roughly 73% of traffic. Before testing anything, place yourself against the Shopify conversion benchmarks breakdown and the one-page DTC benchmark card; a "low" conversion rate that's actually at band for your category is not the constraint.

The strategy menu on this side:

• Cut decisions before payment. Every extra choice past purchase intent feeds the abandonment number. The choice research is blunt: adding a second option lifted purchase intent from 9% to 32% in Daniel Mochon's single-option aversion studies, but stacking options past a handful reverses the effect (CXL, choice research roundup). The shape is an inverted U: one option reads as a trap, a few read as a choice, a wall reads as work.
• Give the customer control. Choice architecture beats choice pressure. Build-your-own flows, visible quantity control, and live price updates convert because the shopper is assembling their own order instead of resisting yours. Control is also the rare conversion lever that doesn't require discounting anything.
• Make shipping costs boring. Surprise shipping at checkout is the most cited abandonment trigger in checkout research. In the threshold experiment covered below, sessions that qualified for free shipping completed checkout at 65.06% against 60.62% for sessions that didn't, a 7% completion gap from shipping alone. Transparent thresholds and early cost display defuse it.
• Speed. Slow stores bleed conversions before design ever gets a vote, and the damage concentrates on mobile, exactly where conversion is already weakest. I covered the numbers in store speed and conversion.
• Trust and proof. Reviews near the buy button, clear returns, recognizable payment marks, express-pay buttons that skip the form entirely. Unsexy, compounding.

Match the fix to the stage that's leaking. If sessions die before the product page, the problem is traffic quality or findability, and no checkout tweak will touch it. If add-to-cart trails your category, the PDP isn't earning the click: price framing, imagery, proof. If carts die at the shipping step, it's thresholds and cost surprise. If checkout starts but doesn't finish, it's form length and payment options. Testing a PDP fix against a checkout leak is how brands run six months of A/B tests and conclude, wrongly, that "testing doesn't work for us."

Notice the tax runs the other way here. Several conversion levers quietly bill AOV: discounts convert price-sensitive buyers at lower ticket sizes, lower thresholds shrink baskets, and customer-controlled quantities let people buy less than your bundle wanted them to. Neither side of this menu is free, which is why the next section exists.

Why do AOV and conversion
rate fight each other?

Because nearly every lever that raises order size adds price, friction, or decisions at the exact moment a customer is trying to say yes. And nearly every lever that lifts conversion works by removing one of those three things. Lay the two menus side by side and the seesaw is obvious:

The seesaw isn't symmetric, either, and the asymmetry matters for strategy. Conversion losses compound with paid traffic, because you already paid for every session that abandons. AOV losses compound with fulfillment, because you pay to pick, pack, and ship every smaller box. Which side of that asymmetry your business sits on is most of the answer to "which metric should we optimize," and it's the spine of the two verdict sections coming up.

Zoom out and this is one instance of a pattern that runs through all of commerce: fixed variables that trade against each other. Price trades against volume. Shipping speed trades against shipping cost. Discount depth trades against margin. SKU breadth trades against operational complexity. In every pair, the amateur move is maximizing one side because a benchmark said to, and the operator move is finding the ratio that fits your cost structure. AOV and conversion rate are just the most visible of these pairs, because both numbers sit on the Shopify home screen taunting you daily. Once you see the pattern, "increase your AOV" sounds exactly like "raise your prices" would: not wrong, not right, just incomplete until someone says what it costs on the other side of the trade.

Two public tests show the seesaw paying out in opposite directions, which is exactly the point. In a published free-shipping threshold experiment, lowering the threshold lifted conversion 6% and revenue per visitor 2.03%, worth roughly $92,915 in added monthly revenue against about $28,720 in new shipping costs (Intelligems, threshold testing guide, 2024). The threshold had been set to protect AOV and was quietly taxing conversion harder than the AOV was worth.

The second is the cleanest public demonstration I've seen of deliberately trading AOV away. Growth marketer Harry Molyneux published results from a DTC bagel brand that swapped forced bundles for a build-your-own flow with quantity control. AOV fell 3.91%, conversion jumped 20.14%, revenue per visitor rose 11.44%, and profit per visitor rose 14.03%, worth an estimated $42,405 in added monthly revenue (Harry Molyneux, LinkedIn, 2025). His one-line summary of the lesson: "You don't always need bigger baskets. Sometimes you need more baskets." Credit where due: most operators never publish numbers this clean. Note that neither test "won" a metric. Both found a better point on the curve.

When AOV is
the better bet.

AOV is the better side of the trade when the economics of an order dominate the economics of a session. Five situations where that's true:

When fulfillment costs are heavy. If your product is bulky, heavy, cold-chain, or expensive to pick and pack, every order carries a fixed cost that has nothing to do with its size. A conversion win that adds more, smaller orders multiplies that fixed cost; an AOV win amortizes it. A furniture brand and a sticker shop can run the identical test and get opposite profit outcomes at the same RPV. If contribution margin per order is thin, order count is not your friend.

Put numbers on that. Two brands each earn $1.50 per visitor. Brand A ships $15 snack boxes that cost $6 to fulfill; brand B ships $300 side tables that cost $70 in freight and handling. A test that trades 4% of AOV for 12% more orders is a clear win for the snack brand: fulfillment barely notices, and the new customers feed a replenishment cycle. The identical result costs the furniture brand real money, because twelve percent more freight bills arrive attached to smaller invoices. Same test, same RPV movement, opposite verdicts. The RPV line can't see the loading dock.

When traffic is capped or expensive. Niche categories, small TAMs, and brands already saturating their best audiences can't just buy more sessions at yesterday's price. When the next thousand visitors cost meaningfully more than the last thousand, monetizing the buyers you already win becomes the growth lever. This is also the late-stage condition: mature brands with rising CPCs tend to migrate from conversion plays toward order-economics plays as their acquisition curve steepens.

When you bill on a cycle. Subscription and replenishment brands hold the single most powerful AOV lever in commerce, billing period, and it barely touches conversion because the decision happens at the plan-selection step, not the checkout button. Quarterly and annual plans raise effective order value, pull cash forward, cut shipping frequency, and reduce cancellation decisions per year. If you bill monthly and have never tested a quarterly default, that's the first test.

When conversion already sits at the band ceiling. Conversion optimization has diminishing returns like everything else. If you're already converting at or above your vertical's band, the next conversion point costs more to find than the last one did, and the honest read is that your curve has more room on the order-value axis. The reverse read applies too, which is the next section.

When the category monetizes through order value. Look at how differently the verticals solve the same equation. Luxury converts under 1% and still out-earns nearly everyone per visitor, because a $350+ AOV does the work. Food converts near the top of the table on small baskets. Neither category is wrong; they're different points on the same curve, and each one's levers follow from its position:

The chart carries the strategic warning too: a luxury brand that chases conversion with discounts is attacking the one variable its category monetizes through. Price integrity is an asset with compounding value, and a conversion lift bought by eroding it is the most expensive conversion lift there is.

When is conversion
the better bet?

Conversion is the better side when the economics of a session dominate the economics of an order, and when a new customer is worth more than their first receipt. Five situations where that's true:

When your traffic is mostly paid. Every abandoned session is a session you paid for, and the CAC identity from Plate 02 makes the logic mechanical: CAC equals cost per visitor divided by conversion rate, so conversion is the only on-site lever that cuts acquisition cost. A brand spending heavily on Meta and Google with a below-band conversion rate has an arbitrage sitting on its own website: the cheapest "media buy" available is converting more of the traffic already bought.

When repeat economics are strong. An AOV lift pays you once, on this order. A conversion lift adds a customer, and a customer with a healthy repeat rate keeps paying long after the test ends. If your 90-day repeat rate is strong for your category, check yours against the DTC retention benchmarks, then a converted first order is a discounted ticket into a revenue stream, and conversion carries LTV that never shows up in the test readout. This is the single most under-weighted argument in the AOV vs conversion debate, because the test window is too short to see it.

When AOV sits above band while conversion lags. This is the pressure diagnosis, and it's the bagel case exactly: forced bundles held AOV artificially high while taxing one in five potential buyers out of the funnel. If your AOV beats your vertical's band and your conversion trails it, the most likely explanation is that your order-value machinery is squeezing too hard at the wrong moment.

When your traffic skews mobile and mobile underperforms. With desktop converting at 3.9% against mobile's 1.8% and mobile carrying roughly 73% of traffic, most stores' blended conversion rate is mostly a mobile number. If your mobile gap is wider than two-to-one, conversion work aimed specifically at the mobile checkout path is usually the highest-expected-value test on the board, and it's invisible in blended reporting.

When orders are cheap to serve and the buyer base compounds. Light products, high margins, digital-adjacent goods: when fulfillment barely notices an extra order, more orders is nearly pure upside. Early-stage brands get a second compounding effect, because every new customer is also a review, an email subscriber, a lookalike-seed data point, and a cohort for the retention curve. In year one, the customer count is worth more than the basket size, and the testing agenda should reflect that.

And if you read both verdict lists and found yourself in neither camp, that's the honest position for most brands between $2M and $20M: moderate fulfillment costs, mixed traffic, decent but not exceptional repeat rates. The answer there is to alternate instead of picking a religion. Run a conversion-side test this cycle, an AOV-side test next cycle, keep the same referee for both, and let two or three quarters of results tell you which side of the curve keeps paying you. The diagnosis doesn't have to precede the testing; done honestly, the testing is the diagnosis.

Revenue per visitor
is the referee.

Revenue per visitor is conversion rate multiplied by AOV, which makes it the one number that can't be gamed by moving value from one side of the seesaw to the other. Multiply the 2025 medians, 2.01% and $74.12, and the median store earns roughly $1.49 per visitor. Every test you run either moves your version of that number or it doesn't, and a test that moves a single metric without moving RPV has just relocated revenue, not created it.

But the full referee stack has three layers, and each one catches a failure the layer above it misses. RPV catches metric-shuffling. Profit per visitor, RPV adjusted for margin and per-order costs, catches wins bought with discounts or shipping subsidies. And the spread from Plate 02, contribution per visitor minus cost per visitor, catches the quietest failure of all: a store whose on-site numbers improve while its traffic economics rot. The immediate pushback in any operator conversation about the bagel numbers is the right one: if the brand made 10% profit before, does the new mix still make it? That test survived the question, profit per visitor rose 14.03%, but plenty of RPV wins don't. Run the same check with contribution margin per order before celebrating anything.

This layered scoreboard is also what "ideal outcome" actually means for a testing program. The goal is not a higher conversion rate or a higher AOV. The goal is a wider spread at stable or better margin, quarter over quarter. Write that sentence at the top of your testing doc and half the arguments about which metric matters resolve themselves.

Practically, this is a five-line dashboard, and it fits on one screen:

Five formulas, all computable from numbers you already have. Most brands I audit can produce every input in twenty minutes and have never put them in one place. The brands that do stop having the AOV vs conversion argument entirely, because the scoreboard answers it monthly.

Which side should
you test first?

Diagnose before you test. Your position against your vertical's benchmarks tells you which side of the seesaw is constrained. The verdicts from Plates 06 and 07 add which side your cost structure favors, and the lever's funnel position sets what the test will cost to run:

Whatever the diagnosis says, sequence by conversion tax. Post-purchase levers charge none, so they're free lunches: prove them first with minimal traffic. Pre-purchase AOV levers all charge checkout something, and friction-removal tests give up basket value on purpose. Those true tradeoffs deserve the full referee treatment, and a product page audit usually shows you where the pressure sits before you spend a test on it.

Traffic decides how ambitious the test can be. Detecting a small AOV shift needs thousands of orders per variant, which is why the threshold experiment above waited for 5,000 orders before calling a sub-10% lift. At 300 orders a month, don't test button colors and $5 threshold moves; test big structural swings where the answer shows up at your volume. Quantify what a conversion point is actually worth to you first with the conversion revenue leak calculator.

How do you run the test
without fooling yourself?

Tooling first. For Shopify brands I default to Shoplift, the first A/B testing app to earn Shopify Plus certification. The reason is structural: Shoplift duplicates your theme templates and splits traffic natively instead of repainting the page with a JavaScript overlay, so a bundle-builder variant renders clean with no flicker and no speed penalty. AOV vs conversion experiments are template experiments, and template experiments are exactly what native testing is built for. It tracks revenue per visitor, AOV, conversion, and add-to-cart out of the box, matches results 1:1 to Shopify orders, and predicts significance with machine learning as the test runs. The vendor reports a +21.7% median conversion lift across 2.1 billion analyzed sessions, and plans start at $99 a month. Weigh vendor numbers accordingly, but the platform mechanics are the point. If the lever you're testing is price itself or the shipping threshold, that's available on Shoplift's Advanced tier, where price changes sync across product pages, collections, cart, and checkout.

Structure the variant as one coherent hypothesis, not a grab bag of tweaks. "Giving customers quantity control lifts revenue per visitor" is testable. "New PDP design" isn't a hypothesis, it's a redesign wearing a lab coat. Then the discipline, which matters more than the tool:

• One lever per test. If you change the bundle flow and the threshold together, the result tells you nothing about either.
• Run two full business cycles. Most valid tests need 2 to 4 weeks; a Tuesday-only read is a coin flip, whatever the dashboard says (Convert, test duration guide).
• Wait for 100+ conversions per variation and 95% significance. Peeking early inflates false positives, and a false winner compounds: it feeds a wrong lesson into every test after it.
• Segment mobile and desktop. With a 3.9% vs 1.8% conversion gap between devices, a variant that wins on desktop can lose on the device carrying most of your traffic.
• Set guardrail metrics before launch. Contribution margin per order, shipping cost per order, and refund rate. A test that wins RPV and breaches a guardrail is a loss that photographs like a win.

Then read the combinations, not the single metric that moved your way:

The discount-driven conversion win is the most common trap in DTC testing, and the break-even math from the margin post linked earlier is blunt about why: at a 40% margin, a 10% price cut needs a 33% conversion lift just to break even. The AOV-up-conversion-down win deserves its own follow-up too, because a customer pressured into a big first order doesn't always come back for a second one.

Last, set expectations for the program itself, because this is where testing cultures die. Most tests won't produce a shippable winner, and that's cartography, not failure. A flat result tells you a lever your competitors are busy "optimizing" doesn't matter to your customers, which is worth knowing at $99 a month. A good testing quarter looks like this: three or four clean single-lever tests, each run to significance, each logged against the referee stack, one or two shipped changes, and a written picture of where your curve bends that didn't exist in January. The ideal outcome is knowing, with receipts, which side of the seesaw your next dollar should land on, not a lucky 20% lift.

One more discipline: the curve moves. Your Q4 shopper tolerates a higher threshold than your February shopper; a product that's gift-heavy in December is replenishment-driven in March. A threshold or bundle configuration that won a test in one season is a reading from that season's curve, not settled law. Retest your highest-leverage levers when the calendar or the traffic mix changes materially, and log every result, winners and losers, in one running doc. Twelve months of logged tests is a proprietary map of your demand curve that no agency, competitor, or benchmark report can give you.

The bigger mindset shift is accepting that AOV and conversion rate were never numbers you could set independently. Any brand can spike either one on demand. The skill is finding the point on the curve where the spread between what a visitor costs and what a visitor pays is widest, and no benchmark, guru, or LinkedIn case study can tell you where that point sits for your brand. Your customers can. That's the whole reason to test.

Questions worth
answering straight.

Should I optimize AOV or conversion rate first?

What is a good revenue per visitor?

What is cost per visitor and why should I track it?

Can decreasing AOV actually increase total revenue?

How long should an AOV vs conversion rate test run?

What's the best tool for testing AOV vs conversion rate on Shopify?

Next time someone tells you to raise your AOV, ask them one question: what does that do to revenue per visitor? If they can't answer, they're optimizing a variable, not a business. Price your sessions, diagnose which side is constrained, pick one lever, run it clean for two cycles, and let the referee decide.