701 A/B tests from 52 food & drink brands. What they test, and what they keep.
A benchmark of real A/B tests run by food & drink ecommerce brands over the past 3+ years (Feb 2023 to Oct 2026), collected and anonymized 
Robert V.A/B test developer · Bucharest10 years of building A/B tests, now for ecommerce brands and agencies. This benchmark is how I keep track of what’s worth testing.More about me → for my own benchmark. The 26 most common test ideas, counted down by how often brands try them and how often they keep them. Pick your industry to see what your competitors test.
01 · The map
What food & drink brands test, and what they keep.
Across: how many brands tried the idea. Up: how many of them kept it. Bubble size: number of tests. Tap a bubble for the numbers.
02 · The ranking
26 A/B test ideas for food & drink brands, counted down.
From #26 to #1, ranked by a mix of how many brands test the idea, how often they keep it and how many roll out a winner. Open any idea for a control vs. variation example and how to build it. The numbers for #6–#11 and the top 5 are in the full benchmark.
-
26 Trust & guaranteesSmall sampleGuarantee, free returns and secure checkout, right under the button. 6% tested it 0% kept it
Example only: a generic version of this test.
Doubt peaks at the button. Brands test a guarantee badge, a returns line and payment icons in the buy box, especially for higher prices and new visitors.
- Brands that tested it
- 3 6%
- Kept rate
- 0%
- Tests marked winners
- 0 of 3
- Median days live
- – 0 at 90d+
Small sample: fewer than 8 food & drink brands launched this. Use the all-industries numbers as the guide.
What brands test
- Money-back guarantee badge near Add to Cart
- Free returns / trial line under the button
- Payment and security icons
- Guarantee repeated in the cart
What to measure
Add to cart rate, then conversion rate.
How to build it
Short and real: one line people believe works better than a row of badges. Use your actual policy wording.
Where it runs
-
25 Popups & email / SMS captureSmall sampleWhen and how the signup popup shows: timing, offer and step count. 8% tested it 0% kept it
Example only: a generic version of this test.
Popups trade a little conversion for a lot of email and SMS signups. Brands test the offer, the delay, exit intent vs. timed, and two-step forms.
- Brands that tested it
- 4 8%
- Kept rate
- 0%
- Tests marked winners
- 0 of 9
- Median days live
- 0d 0 at 90d+
Small sample: fewer than 8 food & drink brands launched this. Use the all-industries numbers as the guide.
What brands test
- 10% off vs. free gift vs. contest
- On exit vs. after 10 seconds
- Email first, then SMS (two steps)
- No popup on product pages
What to measure
Signup rate and conversion rate together, plus revenue from the list later.
How to build it
Make sure the popup can’t show on top of the cart or checkout, and that it’s easy to close on mobile.
Where it runs
-
24 Announcement bar & bannersThe top bar: which message, rotating or fixed, and is it worth the space? 19% tested it 0% kept it
Example only: a generic version of this test.
The top bar is seen on every page. Brands test the message (shipping, offer, proof), rotating vs. one fixed message, and whether to show it at all on mobile.
- Brands that tested it
- 10 19%
- Kept rate
- 0%
- Tests marked winners
- 0 of 13
- Median days live
- 4d 0 at 90d+
Small sample: fewer than 8 food & drink brands launched this. Use the all-industries numbers as the guide.
What brands test
- Free shipping vs. offer vs. review count
- One fixed message vs. rotating
- Countdown to offer end
- No bar on mobile
What to measure
Conversion rate across the site.
How to build it
Keep the height the same in both versions so nothing jumps, and make sure it can’t cover the sticky header on mobile.
Where it runs
-
23 Urgency & scarcitySmall sampleReal low-stock and order-by-time messages near the button. 6% tested it 33% kept it
Example only: a generic version of this test.
Brands test true urgency: low stock on a size, a delivery cut-off time, an offer end date. Fake timers are easy to spot and hurt trust, so the tests that last use real data.
- Brands that tested it
- 3 6%
- Kept rate
- 33%
- Tests marked winners
- 0 of 10
- Median days live
- 230d 1 at 90d+
Small sample: fewer than 8 food & drink brands launched this. Use the all-industries numbers as the guide.
What brands test
- “Only 4 left in size 10” from real stock
- “Order in 2h 14m for delivery Thursday”
- Offer end date in the hero and cart
- Back-in-stock and “selling fast” badges
What to measure
Add to cart rate and conversion rate.
How to build it
Read stock and cut-off times from real data and hide the message when it isn’t true.
Where it runs
-
22 Benefits & USPsThree short benefits with icons, high on the page, instead of a paragraph. 17% tested it 17% kept it
Example only: a generic version of this test.
Shoppers scan. A paragraph of description loses to three short lines a shopper can read in two seconds. Brands test benefit bullets, icon rows and “why us” blocks near the buy box.
- Brands that tested it
- 9 17%
- Kept rate
- 17%
- Tests marked winners
- 0 of 13
- Median days live
- 40d 0 at 90d+
Small sample: fewer than 8 food & drink brands launched this. Use the all-industries numbers as the guide.
What brands test
- 3 benefit bullets with icons under the price
- “Why us” comparison row
- Benefits as image overlays in the gallery
- Ingredient / material callouts
What to measure
Add to cart rate.
How to build it
Keep benefits in product data (metafields), so each product shows its own and the block hides when there’s nothing to show.
Where it runs
-
21 Reviews & social proofStar rating under the title, review snippets near the button, UGC photos. 21% tested it 14% kept it
Example only: a generic version of this test.
Most product pages have reviews, but at the bottom. Brands test moving proof up: the rating next to the title, one strong review next to the button, customer photos in the gallery.
- Brands that tested it
- 11 21%
- Kept rate
- 14%
- Tests marked winners
- 0 of 17
- Median days live
- 28d 1 at 90d+
Small sample: fewer than 8 food & drink brands launched this. Use the all-industries numbers as the guide.
What brands test
- Stars + review count under the product title
- One review quote near Add to Cart
- Customer photos in the gallery
- Review count on collection cards
What to measure
Add to cart rate on the product page.
How to build it
Read ratings from the review app’s data already on the page, and hide the element when a product has no reviews.
Where it runs
-
20 Mobile-specific UXSmall sampleMobile-only changes: shorter product page, sticky Add to Cart, thumb-friendly controls. 15% tested it 50% kept it
Example only: a generic version of this test.
Most traffic is mobile, but many sites are designed on desktop. Brands run mobile-only tests: collapse long content, keep the button in reach, make variant buttons big enough for a thumb.
- Brands that tested it
- 8 15%
- Kept rate
- 50%
- Tests marked winners
- 0 of 10
- Median days live
- 55d 0 at 90d+
Small sample: fewer than 8 food & drink brands launched this. Use the all-industries numbers as the guide.
What brands test
- Long content in accordions on mobile
- Sticky Add to Cart bar
- Bigger tap targets for size and colour
- Mobile-only menu and filters
What to measure
Mobile conversion rate. Segment the results by device.
How to build it
Target by device in the test settings, not only with CSS, so desktop visitors don’t count in a mobile test.
Where it runs
-
19 Quiz & product finderSmall sampleA short quiz that recommends the right product, for catalogues where people need help choosing. 10% tested it 33% kept it
Example only: a generic version of this test.
Rarely tested, but kept often when it is. For skincare, supplements, mattresses or gifts, a 3–5 question finder can beat a collection page. A good example of an idea that is underrated in the data.
- Brands that tested it
- 5 10%
- Kept rate
- 33%
- Tests marked winners
- 2 of 11
- Median days live
- – 0 at 90d+
Small sample: fewer than 8 food & drink brands launched this. Use the all-industries numbers as the guide.
What brands test
- Quiz entry in the hero vs. collection page
- Quiz result with a bundle offer
- 3 questions vs. 7 questions
- Finder on the collection page
What to measure
Conversion rate of quiz takers vs. the same traffic on the collection page.
How to build it
Keep it short and link results to real products and bundles, so the result page can add to cart in one tap.
Where it runs
-
18 Variant & option selectorsSize and colour as buttons and swatches instead of dropdowns, sold-out sizes shown clearly. 23% tested it 13% kept it
Example only: a generic version of this test.
Dropdowns hide the options and the stock. Brands test buttons and swatches, showing unavailable sizes, size guides next to the selector, and the default variant.
- Brands that tested it
- 12 23%
- Kept rate
- 13%
- Tests marked winners
- 2 of 22
- Median days live
- 11d 1 at 90d+
What brands test
- Size buttons instead of a dropdown
- Colour swatches with the name on select
- Sold-out sizes crossed out, not hidden
- Size guide / fit finder next to the sizes
What to measure
Add to cart rate and fewer “no size selected” errors.
How to build it
Keep the theme’s variant logic underneath, so price, image and stock update the same way in both versions.
Where it runs
-
17 Cart & cart drawerOpen a cart drawer after Add to Cart, with reassurance and a clear path to checkout. 33% tested it 8% kept it
Example only: a generic version of this test.
What happens after Add to Cart: stay on the page, open a drawer, or go to the cart page? Brands test the drawer, what it shows (savings, delivery date, reassurance) and how fast it gets people to checkout.
- Brands that tested it
- 17 33%
- Kept rate
- 8%
- Tests marked winners
- 1 of 36
- Median days live
- 5d 0 at 90d+
What brands test
- Cart drawer vs. cart page
- Delivery estimate and returns line in the drawer
- Savings summary (“You save $24”)
- Checkout button always visible in the drawer
What to measure
Checkout starts per cart, and conversion rate.
How to build it
Use the theme’s cart API so totals, discounts and line items stay correct. Test with discounts, bundles and subscriptions in the cart.
Where it runs
-
16 Upsells & cross-sellsUnderratedOne relevant add-on in the cart or near the button: “Complete the look”, “Add socks for $14”. 27% tested it 33% kept it
Example only: a generic version of this test.
Brands test one well-chosen add-on against a row of recommendations, and where to show it: product page, cart drawer or after the purchase. A test of order value.
- Brands that tested it
- 14 27%
- Kept rate
- 33%
- Tests marked winners
- 1 of 28
- Median days live
- 138d 1 at 90d+
Small sample: fewer than 8 food & drink brands launched this. Use the all-industries numbers as the guide.
What brands test
- One add-on in the cart drawer
- “Frequently bought together” on the product page
- Add-on with a small discount
- Upgrade offer (bigger size / pack)
What to measure
Average order value and revenue per visitor. Check that conversion rate doesn’t drop.
How to build it
Pick add-ons by rule (category, price) so the block is never empty or irrelevant, and add to cart without reloading the page.
Where it runs
-
15 Subscription offersSubscribe & save as the default, with savings, perks and “cancel anytime” spelled out. 21% tested it 25% kept it
Example only: a generic version of this test.
For products people reorder, brands test how the subscription is offered: selected by default or not, the savings in dollars, the perks and how clearly cancelling is explained.
- Brands that tested it
- 11 21%
- Kept rate
- 25%
- Tests marked winners
- 1 of 20
- Median days live
- 29d 1 at 90d+
What brands test
- Subscribe & save selected by default
- Savings in $ per delivery instead of %
- Perks list + “cancel anytime”
- Delivery frequency picker vs. one fixed option
What to measure
Subscription rate and revenue per visitor. Watch first-order conversion so the default doesn’t scare people off.
How to build it
Work with the subscription app’s own widget and data, not a copy, so the right selling plan is always sent to the cart.
Where it runs
-
14 Personalized by audienceUnderratedDifferent hero or offer for returning customers, ad traffic or a country. 19% tested it 60% kept it
Example only: a generic version of this test.
Not one page for everyone. Brands test showing returning customers a different hero, matching the page to the ad source, or adapting offers and copy by country.
- Brands that tested it
- 10 19%
- Kept rate
- 60%
- Tests marked winners
- 1 of 21
- Median days live
- 46d 3 at 90d+
Small sample: fewer than 8 food & drink brands launched this. Use the all-industries numbers as the guide.
What brands test
- Returning visitors: “Welcome back” + reorder
- Ad traffic: hero matched to the ad angle
- Country-specific shipping and currency messages
- New visitors: first-order offer
What to measure
Conversion rate inside the targeted audience, compared with the same audience in control.
How to build it
Define the audience in the test targeting (UTM, country, new vs. returning) so control and variation see the same people.
Where it runs
-
13 Add to Cart & CTA buttonsUnderratedButton copy, size, colour and what sits right next to it. 23% tested it 43% kept it
Example only: a generic version of this test.
A classic, and still common. The button itself rarely makes a big difference. What helps is what surrounds it: price on the button, reassurance under it, a clear state once it’s been tapped.
- Brands that tested it
- 12 23%
- Kept rate
- 43%
- Tests marked winners
- 1 of 18
- Median days live
- 111d 2 at 90d+
Small sample: fewer than 8 food & drink brands launched this. Use the all-industries numbers as the guide.
What brands test
- “Add to cart · $128” with the price on the button
- Full-width, high-contrast button on mobile
- Reassurance line under the button (free returns, ships tomorrow)
- Buy now vs. Add to cart
What to measure
Add to cart rate.
How to build it
Change the theme’s button, not a copy of it, so variant changes and apps that listen to it keep working.
Where it runs
-
12 Bundles & quantity breaksBuy more, save more: 1 / 2 / 3 packs with the middle one marked most popular. 31% tested it 17% kept it
Example only: a generic version of this test.
A test about order value more than conversion. Brands test quantity tiers on the product page, pre-built bundles and “complete the set” offers. Common in consumables, supplements and gifting.
- Brands that tested it
- 16 31%
- Kept rate
- 17%
- Tests marked winners
- 4 of 48
- Median days live
- 1d 2 at 90d+
What brands test
- 1 / 2 / 3 pack tiers with savings shown
- Middle tier pre-selected as “most popular”
- Mix & match bundle builder
- Bundle as the default on landing pages
What to measure
Revenue per visitor and average order value, not conversion rate alone.
How to build it
Tie the tiers to real discounts or bundle products, so the cart shows the same price as the page. Test the cart and checkout totals every time.
Where it runs
-
11 Headlines & copyBenefit-led headlines and product copy instead of names and specs. tested it kept it
Example only: a generic version of this test.
The cheapest test to build, and it runs everywhere: hero headlines, product titles, button labels, descriptions. Brands test what the product does for the customer against what it is called.
What brands test
- Benefit headline vs. product name
- Short description first, specs in a tab
- Problem / solution framing
- Copy matched to the ad angle
What to measure
Depends on the place: clicks for a hero, add to cart for a product page.
How to build it
Text swaps are quick, but test them one at a time where the copy is visible on the first screen. Copy far down the page rarely moves anything.
Where it runs
-
10 Discounts & offersOffer framing: % off vs. $ off, first-order offers, auto-applied codes, tiered offers. tested it kept it
Example only: a generic version of this test.
Brands test the offer and how it is shown: the same discount framed as a percentage or a dollar amount, applied automatically or with a code, shown in the hero or at the cart.
What brands test
- 15% off vs. $20 off
- Auto-applied discount vs. a code
- Spend $X, get $Y tiers
- Offer shown on the product page vs. only at the cart
What to measure
Profit per visitor, not conversion rate: a bigger discount almost always “wins” on conversions.
How to build it
Show the offer from the real discount logic, so the price in the cart always matches what the page promised.
Where it runs
-
09 Checkout & payment optionsExpress checkout buttons and pay-in-installments messaging near the price and in the cart. tested it kept it
Example only: a generic version of this test.
The last steps are where intent is highest and any friction costs the most. Brands test which payment options to show and where: express buttons in the cart, installment messaging under the price, fewer steps to checkout.
What brands test
- Express checkout buttons in the cart drawer
- “Or 4 payments of $32” under the price
- Skip the cart page: straight to checkout
- Cash on delivery or local methods in some markets
What to measure
Checkout starts and conversion rate. Check average order value too: installments often move it.
How to build it
Checkout itself is locked on most plans, so these tests live in the cart and product page. Test on real devices: express buttons render differently per browser and wallet.
Where it runs
-
08 Product images & videoLifestyle or UGC images first, a short video, swipeable gallery with visible thumbnails. tested it kept it
Example only: a generic version of this test.
Images do most of the selling online, so the order and type of images is a constant test. A pack shot first or a person using it? A video in position two? Thumbnails or dots? These tests stay live longer than most, which usually means the winner is not obvious.
What brands test
- Lifestyle / in-use image first instead of a pack shot
- Short video or GIF in the gallery
- Thumbnails visible on mobile instead of dots
- Before / after or comparison image
What to measure
Add to cart rate. Watch gallery interaction as a secondary event.
How to build it
Swap image order through the product data or the theme’s gallery, not by hiding images with CSS. Lazy-load the video so it can’t slow the first image down.
Where it runs
-
06 Collection pagesRating and a key benefit on product cards, quick add, better filters, 2 columns on mobile. tested it kept it
Example only: a generic version of this test.
Collection pages are where shoppers compare. The card decides which product gets the click: brands test what the card shows (rating, benefit, badge, second image), how many cards fit on a mobile screen and how filtering works.
What brands test
- Star rating + one benefit line on each card
- Quick add button on the card
- 2 columns on mobile instead of 1
- Sticky filter / sort bar, filters as chips
What to measure
Click-through to product pages, then add to cart rate from the collection.
How to build it
Pull ratings and badges from data the theme already has (product metafields or the review app), so new products are covered without manual work.
Where it runs
- 05 tested it kept it
- 04 tested it kept it
- 03 tested it kept it
- 02 tested it kept it
- 01 tested it kept it
The idea pack · full benchmark + store audit
The top 5, applied to your store.
A list of ideas is the easy part. What’s worth testing on your store depends on what your pages do today. So I audit your store against the full benchmark for food & drink, and turn the ideas that fit into 5 tests built around your pages, your products and your traffic.
- Store audit: your homepage, collection, product page, cart and checkout, on mobile and desktop, checked against what brands in your industry test and keep
- The full benchmark for your industry: all 26 ideas with every number, the top 5 included
- Your top 5 tests: written for your store, each with the hypothesis, a control vs. variation mockup on your own pages and the metric to watch
- A build order: ranked by effort and expected impact, so you know what to launch first
03 · Run time
How long do A/B tests stay live?
How long the live tests at food & drink brands had been running when the data was collected. Half had been running for 20 days or less.
23% had been live for 90 days or more. Those are rarely still “tests”: most are winners running at 100% that were never built into the theme, or tests nobody switched off.
The rule of thumb still holds: run at least two full weeks, so every weekday is counted twice, and stop at the sample size you planned, not at the first green number.
04 · The data
How this benchmark works.
Where it comes from
I built this benchmark for my own work: I collected and anonymized the publicly visible A/B test data of 679 ecommerce stores across different industries, the test setup each store’s own website loads in every visitor’s browser. It covers 9,075 tests from the past 3+ years (Feb 2023 to Oct 2026). Only totals are shown: no brand, store or test is named.
What “kept” means
Of the brands that launched a type of test, the share that kept a version live for 30+ days or marked it as a winner / rolled it out to 100%. It’s a signal, not a lift: long-running tests can be winners or forgotten tests, so read it with the winner count.
How ideas are grouped
Each test is sorted into one or more of 26 types by its name and settings (a bundle test on a landing page counts for both). Industries come from the store’s catalogue. Industries with fewer than 25 brands are only in “All industries”.
What are the best A/B test ideas for food & drink stores?
The 26 ideas on this page are counted down by how often brands test them and keep them. The top 5 are in the full benchmark; just below them in food & drink are collection pages, navigation, menu & search, product images & video.
How often do A/B tests win?
Across 9,075 tests, 749 (8%) were marked as winners or rolled out to 100%. On average, 32% of brands that launched a type of test kept one live for 30+ days or rolled it out.
How long should an A/B test run?
When the data was collected, half of the live tests had been running for 18 days or less. A good rule: at least two full weeks, so every weekday is covered twice, and until you reach the sample size you planned. 19% of live tests have run 90+ days: usually winners waiting to be made permanent.
Where does this data come from?
I collected and anonymized it myself, for my own benchmark: the publicly visible A/B test data of 679 ecommerce stores across different industries, covering 9,075 tests from the past 3+ years (Feb 2023 to Oct 2026). Tests are grouped into 26 types by their name and settings. Only totals are shown: no brand, store or test is named.
Built by Robert V. · A/B test developer
Pick an idea. I’ll build the test.
I build A/B tests for ecommerce brands and agencies, and nothing else: clean, flicker-free variants, QA’d on desktop and mobile, with the tracking set up. Send me a brief and you get the build, usually by the next morning in the US.