Search ecommerce conversion rate optimization and you get research institutes, platform vendors and app companies. The research is genuinely good. What none of it can tell you is what happens when you try to act on it inside a real store with imperfect tracking, a product in decline, and a client who disagrees with you.
That gap is what this covers. Not the benchmark, the middle part.
Start from the number that is actually documented
The most reliable figure in this field is cart abandonment, and it is 70.22%, calculated across 50 separate studies (Baymard). It is worth trusting because it is a meta-average rather than one company's sample.
There is no equally reliable "good conversion rate". Category, price point and traffic source move that number more than site quality does, and any article giving you a single benchmark is comparing your store to an average it has not defined. Your own trend against your own baseline is the only comparison that describes your traffic.
So the useful question is not what should my rate be. It is which change is most likely to move it.
The one result that repeats
Across multiple stores, A/B tested repeatedly, one change has produced a 10 to 20% conversion lift more consistently than anything else I have run: three benefit bullets on each product card.
Not the product page. The card, in the grid, before the click.
It works because a product card is a decision surface that most stores treat as a thumbnail. A shopper scanning a collection is choosing what to open, and a name plus a price gives them almost nothing to choose on. Three short benefit lines turn a browsing decision into an informed one, and they do it before the visitor has spent a click.
It is also the least glamorous change on any list, which is roughly why it stays available.
The results that were real and could not be proven
Here is the part the tidy research cannot include, and the part I would want to read.
On a sea-moss and shilajit brand that had gone viral on TikTok, a set of changes lifted add-to-cart clicks by around 50%. I cannot tell you what it did to revenue. The store's tracking was poor and the whole business was in a sales downtrend at the time, so the lift was real and simultaneously invisible in the only number anyone cared about.
On another store, a redesign produced roughly a 50% lift in call-to-action click rate, on a dying viral product. Same shape: the change worked, the trend swallowed it.
Neither of those is a case study. Both are honest, and both are more useful than a clean win, because this is what optimisation frequently looks like: a real improvement inside a business moving in the other direction, measured through instrumentation nobody set up properly.
Two practical consequences worth taking seriously:
Fix tracking before you optimise anything. A lift you cannot measure is a lift you cannot defend, repeat or get paid for. This is dull work and it comes first.
Separate the trend from the change. If the store is declining, a flat month after your work may be a win and a good month may be seasonality. Decide before shipping what window you will read and against what control, or you will argue about attribution afterwards with nothing to point at.
The surface worth testing fifteen times
On a supplements brand I ran roughly fifteen A/B tests on a single header bar.
That sounds obsessive until you notice the arithmetic: 100% of visitors see the header, on every page. It is the only surface on a store with complete reach. A 2% improvement there touches every session, while a 20% improvement on a page 6% of visitors reach does not.
Most testing programmes work the opposite way, spending scarce traffic on pages far down the funnel where sample sizes are smallest and the reach is lowest. If your traffic is limited, and at $50k to $1m a month it is, test the surfaces everybody sees first.
Three positions I will defend, each with its counter-argument
Contrarian is cheap unless the opposing view is stated, so here it is stated.
Quick-add is a metric trap. The counter-argument, held by competent people: fewer clicks means more conversions. In my own A/B tests, sending shoppers to the product page consistently beat quick-add tiles on both conversion and average order value. Quick-add optimises for speed on a catalogue where the buyer has not yet decided, and it removes the surface that does the persuading. It genuinely does win on high-frequency, low-consideration repeat purchases. It loses almost everywhere else.
Colour does not convert. Anchoring order does. The counter-argument: button and price colour testing is a standard, cheap test. It is cheap and it is close to the lowest-value test available. What moves a pricing decision is the order options are presented in and what the first number teaches the buyer to expect. Spend the same traffic on the anchoring order and you are testing something that can move.
Manufactured urgency is now a negative trust signal. Not neutral, negative. Buyers have learned what a countdown that resets on refresh means, and the inference generalises to everything else on the page. Real scarcity still works. If the deadline is not real, the cost is not zero.
When the data says one thing and the decision goes the other way
On a subscription brand, heatmaps and research showed roughly 9% of shoppers clicked "buy once".
Nine percent is small. The conclusion was still to keep buy-once as the default and visible.
The reason is that the 9% is not the whole effect. Removing the one-time option does not convert those shoppers to subscribers, it converts a share of them to nobody, and it changes how the offer reads for people who never click it: a store that hides the one-time purchase looks like a store that is trying something on. The visible alternative is doing work for the people who do not choose it.
I have also lost this argument. A client stayed 100% subscription-first against my recommendation, on the strength of retention data I had not seen, and their call stood. Which is the honest note to end the data section on: a number tells you what happened, not always what to do.
How to sequence the work
- Fix tracking. Everything below is unmeasurable without it.
- Establish the baseline and the window you will read results over, in advance.
- Start with total-reach surfaces (header, navigation, collection cards) because they use your scarce traffic most efficiently.
- Then the highest-value single page, usually the product page.
- Change one thing, against a control, measured on the metric that matches the change.
- Pick winners on profit, revenue per visitor, or margin-adjusted order value. Never raw sales. A variant that sells more units at a worse margin is a loss with a green arrow next to it.
On step five, the constraint most articles skip: a meaningful A/B test needs roughly 200 to 250 conversions and about 5,000 sessions per variation, and more for split tests. Most stores at this size sit below that floor. Below it you cannot test your way to an answer, and pretending otherwise produces confident decisions built on noise. Fix the documented friction, ship it cleanly, and measure a clean before-and-after over a window you set in advance, and say out loud that is what you are doing.
What platform-specific advice adds
Everything above is platform-agnostic, because the buying psychology is. Where platform matters is implementation: what your theme lets you change, whether your cart is a drawer or a page, how your checkout constrains you.
If you are on Shopify, the mechanics are covered in the complete Shopify conversion rate optimization guide, and the checkout-specific friction in Shopify checkout optimization.
Related reading: The Complete Shopify CRO Guide · Shopify Checkout Optimization · Conversion rate optimization audit
If you want to know which of these applies to your store before spending traffic finding out, book a free 30-minute call. I will tell you which surface is worth testing first, including when the honest answer is that you do not have the traffic to test at all and should rebuild instead.
