eCommerce Conversion Rate Optimization Guide
Conversion rate is the only lever that improves every channel at once. This guide covers what a single point of it is worth on your numbers, where stores actually lose people, the four fixes that move it most, and how to run a test programme on traffic you realistically have rather than traffic a case study had.
CHAPTERS
8 chapters
READING TIME
37 min read
LAST UPDATED
19 September 2026
LEVEL
Intermediate
The cheapest revenue in most accounts is already on the site. It is leaving at a step nobody has measured.
EXECUTIVE SUMMARY
What this guide argues, in five points
Conversion optimisation has a reputation problem. It is associated with button colours and pop-ups because that is what is easy to sell, and the easy things do not move a business. The work that does move it is unglamorous: measuring properly, fixing the templates that carry the revenue, and being honest about what a test can and cannot tell you at your traffic.
- Work out what one point of conversion rate is worth on your numbers before doing anything else. It is usually larger than the media budget being argued over.
- Most stores lose people on the category page and in checkout, and optimise the product page anyway because that is where the tooling points.
- Unexpected cost at checkout is a policy decision, not a design problem. No amount of testing fixes a shipping threshold that is wrong.
- At realistic eCommerce traffic, most tests cannot detect the effect sizes people claim. Size the test before running it or the result is noise with a p-value.
- CRO is a programme, not a project. The value comes from a standing cadence of sized tests and a log that survives handover.
The eight chapters below are the detail behind those five points, in the order we work through them when we take conversion on for a store.
- Contents
Eight chapters
Read it in order the first time. After that, the chapter index is the useful part.
01
Before choosing anything to fix, put the prize in pounds. On most stores it is larger than the media budget being debated.
4 min
02
Five steps, measured per device, compared against each other rather than against a benchmark. The biggest relative drop is the work.
5 min
03
The page that decides which product gets seen is usually the least-optimised page on the store.
4 min
04
The remaining wins are rarely in layout. They are in answering the question that is stopping this particular buyer.
5 min
05
The highest concentration of lost revenue on the site, and most of the fixes are decisions rather than designs.
5 min
06
The losses nobody in the office experiences, because they happen on a mid-range phone on a real network.
4 min
07
At realistic eCommerce traffic most tests cannot detect the effects people claim. Size it first, or the result is noise with a p-value.
6 min
08
The compounding comes from cadence and a log that survives handover, not from any single winning test.
4 min
- Before you start
Who this guide is for
Written for the people who own the conversion number, whether or not they build the pages.
Founders and heads of eCommerce
You want to know whether conversion is your constraint at all, and what a realistic improvement looks like on your traffic.
In-house CRO and UX teams
You have a backlog of ideas and want a way to rank them that survives a conversation with finance.
Merchandisers and store managers
You control sort order, filters, imagery and copy — the levers this guide spends most of its length on.
NOT WRITTEN FOR
People looking for a list of psychological tricks. Urgency banners and fake scarcity move a number for a fortnight and cost trust permanently. Nothing in this guide depends on misleading anybody.
CHAPTERS
- 01 What a point of conversion is actually worth
- 02 Finding where the funnel actually leaks
- 03 Category pages and internal search
- 04 The product page, in order of impact
- 05 Cart and checkout: policy before design
- 06 Speed, mobile and the silent failures
- 07 Testing without fooling yourself
- 08 Running CRO as a programme
CHAPTER 01
4 min
What a point of conversion is actually worth
Before choosing anything to fix, put the prize in pounds. On most stores it is larger than the media budget being debated.
Start by making the prize concrete. A relative improvement of ten per cent in conversion rate — from 1.8% to 1.98% — is not a rounding error; it is ten per cent more orders from traffic you have already paid for, across every channel simultaneously. Write that number down in pounds before deciding whether this work deserves a quarter.
Monthly sessions
120,000
Across all channels
Conversion rate
1.80%
2,160 orders
After a 10% relative lift
1.98%
2,376 orders — 216 more
At £26 contribution
£5,616 a month
£67,392 a year, with no extra media spend
Run this on your own numbers. It takes four minutes and it settles most arguments about where the next quarter goes.
KEY TAKEAWAY
Conversion rate is the only lever that improves search, paid, email and marketplaces at the same time. That is why it is usually cheaper per additional order than the next increment of traffic.
Why the average is the wrong number
- Blended conversion rate hides everything. Branded traffic converts several times better than cold paid social. A blended number moves when your channel mix moves, not when your store improves.
- Device changes the answer. Mobile is usually half the desktop rate and most of the sessions. An average across both describes neither.
- New and returning are different businesses. Returning customers convert on trust you already earned; new visitors convert on what the page says.
- Category matters more than any benchmark. A considered purchase at £900 will never convert like a £12 consumable, and comparing them is how sensible teams get told they are failing.
Segment by channel, device and visitor type before you form a view. The gap between two segments is usually the finding; the average is usually the thing that hid it.
CHAPTER 02
5 min
Finding where the funnel actually leaks
Five steps, measured per device, compared against each other rather than against a benchmark. The biggest relative drop is the work.
Every store has a funnel, and every funnel leaks at one step more than the others. Finding that step takes an afternoon and saves a quarter, because the alternative is a backlog of improvements distributed evenly across a journey where only one part is broken.
The five-step measurement
Step
What to measure
Where it usually goes wrong
Landing
Sessions reaching a category or product page
Paid traffic landing on the home page rather than the page it was promised
Product view
Sessions that open at least one product
Category page too thin to help anybody choose
Add to cart
Sessions adding an item
Missing size, delivery or specification information at the decision point
Checkout start
Sessions beginning checkout
Cart page that makes the total look worse than expected
Purchase
Completed orders
Unexpected cost, forced account creation, payment method missing
Do this per device. The mobile column is where the answer usually is, and it is almost never the same answer as desktop.
KEY TAKEAWAY
Compare each step against the step above it, not against a benchmark. The biggest relative drop is your constraint, whatever an industry average says the number should be.
What the numbers will not tell you
Analytics tells you where people leave. It does not tell you why, and guessing the why is how teams end up testing button colours. Three cheap sources close that gap and none of them needs a research budget.
- Session recordings on the leaking step. Twenty on mobile is usually enough to see the problem; you are looking for hesitation, repeated scrolling and rage clicks, not for a statistic.
- Support tickets and pre-purchase questions. Every repeated question is an unanswered objection sitting on a template.
- Returns reasons. They tell you which expectation the page is setting wrongly — a conversion problem that shows up as a logistics cost.
- One on-site question, asked on the leaking page: what nearly stopped you buying today. Short, unbranded and answered more honestly than any survey panel.
Our 40-check CRO checklist is the systematic version of this chapter, run against a live store.
CHAPTER 03
4 min
Category pages and internal search
The page that decides which product gets seen is usually the least-optimised page on the store.
The category page is the most under-optimised template on most stores and the one that decides which product a customer ever sees. Its job is not to display inventory. Its job is to get somebody from “I want one of these” to “that one” in as few decisions as possible.
If a customer cannot narrow a hundred products to three, the product page never gets a chance.
What moves conversion on a category page
- Default sort order. The single highest-leverage setting on the page, and on most stores it is whatever the platform shipped. In stock, in margin, and proven to sell beats newest.
- In-stock products in the first row. An out-of-stock hero product at the top of a grid is a silent, daily conversion tax.
- Filters that match how buyers decide, not how the catalogue is structured. Size, fit, compatibility and use case beat a tag taxonomy nobody outside the business understands.
- Enough information on the card to shortlist without clicking: price, key attribute, availability, and a second image on hover or swipe.
- A short orienting paragraph for people who have arrived from search and do not yet know what makes one of these better than another.
KEY TAKEAWAY
Fix the default sort order before testing anything else on a category page. It is a five-minute change that applies to every session and it is almost never deliberate.
Internal search, the forgotten template
Visitors who use internal search convert at several times the site average, because they arrived already knowing what they want. They are also the easiest people to lose, because a no-results page is a dead end with no recovery path.
1
Read the top fifty search terms monthly. They are the most honest statement of demand your store produces.
2
Fix zero-result terms first. Usually synonyms, misspellings or a product you sell under a different name. Each one is a customer who wanted to buy.
3
Never show an empty page. Show the nearest category, the closest matches and a way to ask.
4
Watch for terms you do not stock. A ranked list of products your customers expect you to sell is a merchandising brief, free of charge.
CHAPTER 04
5 min
The product page, in order of impact
The remaining wins are rarely in layout. They are in answering the question that is stopping this particular buyer.
The product page is where the decision is actually made, and it is the template most teams have already been over. The remaining wins are rarely in layout. They are in answering the specific question that is stopping this particular buyer, which is a research problem before it is a design one.
The order in which things matter
Element
What it resolves
Typical failure
Imagery
Is this actually what I think it is?
Too few images, no scale reference, no detail shots
Specification completeness
Will it fit, work, last?
Marketing adjectives where dimensions should be
Delivery and returns on the page
When does it arrive and what if it is wrong?
Buried two clicks away in a policy page
Variant selection
Which one do I need?
Unavailable variants shown as selectable
Social proof
Did people like me regret this?
Five-star wall with no detail, which reads as fake
Questions answered
The thing I am still unsure about
Invented FAQs rather than real support tickets
Roughly in order of impact across the stores we have worked on. Your category may reorder the middle; imagery is first almost everywhere.
KEY TAKEAWAY
Stating delivery date and returns terms on the product page is one of the most reliable conversion improvements available, and it is a content change rather than a test.
Reviews, honestly
Reviews work because they are credible, and they stop working the moment they look managed. A page of identical five-star sentences converts worse than a page with a handful of three-star reviews that name a real trade-off and get a reply. Show the distribution, keep the critical ones, and answer them.
Our short guide on the nine product page elements that move conversion — and the four that rarely do — is the scoring version of this chapter.
CHAPTER 05
5 min
Cart and checkout: policy before design
The highest concentration of lost revenue on the site, and most of the fixes are decisions rather than designs.
Checkout is the step with the highest concentration of lost revenue and the lowest concentration of creative opinion, which makes it the most reliable place to work. It is also where the most important fixes are not design changes at all but decisions somebody in the business has to make.
Most abandonment at this step was decided earlier, by a shipping policy rather than a layout.
The policy problems, first
- Unexpected cost. Consistently the most cited reason for abandoning a cart. Show delivery cost before checkout, or set a threshold and say so on every page.
- Forced account creation. Guest checkout is not a feature request; it is the removal of a barrier you chose to add. Offer the account after the order instead.
- Delivery speed and choice. If the only option is five to seven days in a category where two is standard, that is a fulfilment decision showing up as a conversion number.
- Returns terms. A short, plainly stated policy converts better than a generous one nobody can find.
KEY TAKEAWAY
No amount of testing fixes a shipping threshold that is wrong. Four of the five biggest checkout problems on most stores are policy decisions, and they are free to change.
Then the mechanics
1
Cut fields to what you need to ship and bill. Every optional field is a small tax paid by every customer.
2
Use address lookup and card scanning. Mobile typing is the single biggest friction in checkout and both are cheap to add.
3
Offer the wallets your customers use. Express payment methods skip most of the form and most of the failure modes with it.
4
Keep the order summary visible. Hiding the total does not reduce abandonment; it relocates it to the confirmation step.
5
Show errors inline, in plain words, and never clear the form. A declined payment with a wiped basket is a customer lost permanently.
On a hosted platform, checkout is often the part you control least. Where that is true, the work moves upstream: the cart page, the delivery messaging and the expectations set on the product page, all of which you do control.
CHAPTER 06
4 min
Speed, mobile and the silent failures
The losses nobody in the office experiences, because they happen on a mid-range phone on a real network.
Some conversion losses never appear as a design problem because nobody experiences them in the office. They happen on a mid-range phone on a patchy connection, or on a page where an app failed quietly, and they are invisible until somebody goes looking on the right device.
Measure the templates that carry revenue
- Test category and product, not the home page. The home page is the page your team looks at and the page your customers spend least time on.
- Use a mid-range phone on a real network. A flagship device on office wifi will tell you everything is fine.
- Identify what the largest element actually is. On most product templates it is the hero image, and it is usually larger than it needs to be.
- Audit apps and tags against the pages that use them. A review widget loading on checkout is pure cost. Unowned tags accumulate in every account we inherit.
KEY TAKEAWAY
Speed is not a ranking checkbox. It is a conversion input, and the relationship is steepest exactly where most of your sessions are: mobile, over cellular, on a mid-range device.
The quiet failures worth checking monthly
- Payment or address validation failing for one country or card type. Invisible in aggregate, total for the customers affected.
- A discount code field that rejects valid codes — or advertises that codes exist to people who do not have one.
- Out-of-stock variants still selectable, producing an error only after the customer has committed.
- Cookie and consent banners covering the add-to-cart button on small screens. More common than it sounds, and it costs every mobile session.
- A broken step in one browser. Check the two or three that actually appear in your analytics, including in-app browsers.
CHAPTER 07
6 min
Testing without fooling yourself
At realistic eCommerce traffic most tests cannot detect the effects people claim. Size it first, or the result is noise with a p-value.
Most eCommerce stores do not have the traffic to run a clean A/B test on anything except their highest-volume templates. That is not an argument against testing. It is an argument for testing fewer, larger things, and for being honest about what a result means when the sample is small.
A verdict should be arithmetic. If the test was underpowered, that is the finding, whatever the p-value says.
Size the test before you run it
The sample needed to detect an effect scales with the inverse square of the effect size. Detecting a 10% relative lift takes roughly four times as many sessions as detecting a 20% one. A rough approximation is enough to decide whether a test is worth starting: at a 2% baseline, detecting a 10% relative change needs somewhere near thirty thousand sessions per variant.
What you are testing
Realistic effect
What to do at low traffic
Checkout step removal
Large
Testable on most stores
Delivery messaging on product pages
Medium to large
Testable; run it site-wide if not
Category default sort order
Medium to large
Testable, and often the biggest single win
Button colour or copy tweak
Very small
Not testable. Do not spend a month finding out.
Full template redesign
Large but confounded
Ship it and measure before and after, honestly labelled
Where a test is not viable, a before-and-after with a stated caveat is more honest than an underpowered test presented as evidence.
KEY TAKEAWAY
Peeking at a running test and stopping when it looks good is the most common way teams convince themselves of a lift that is not there. Set the duration before you start and read it once.
Rules that keep a programme honest
- Run whole business cycles. At least two full weeks, so weekday and weekend behaviour are both represented.
- One change per test unless you are deliberately testing a whole concept and will accept not knowing which part worked.
- Log everything, including losers. Losing tests are the cheapest information you own and the first thing lost at handover.
- Re-check winners after a quarter. Novelty effects fade, and some winners quietly stop winning.
- Report in contribution, not conversion rate. A lift that shifts people to a lower-margin product is not a win.
Our CRO testing template carries the sizing approximation and the significance calculation, so a verdict comes out of the spreadsheet rather than a meeting.
CHAPTER 08
4 min
Running CRO as a programme
The compounding comes from cadence and a log that survives handover, not from any single winning test.
The stores that compound conversion improvements are not the ones with the best ideas. They are the ones with a standing cadence: a prioritised backlog, a fixed number of tests in flight, a readout everybody attends, and a log that survives the person who started it.
A cadence that works at most sizes
1
Monthly: read the funnel. Five steps, by device. Note what moved and what did not.
2
Monthly: groom the backlog. Each item carries an expected effect, the traffic needed, and the evidence behind it.
3
Continuously: one or two tests in flight. More than that on eCommerce traffic means interference and slower reads.
4
Quarterly: ship the unfixable-by-test items. Policy changes, speed work, silent-failure fixes — measured before and after.
5
Quarterly: re-check last quarter’s winners. Cheap, and it catches the ones that faded.
KEY TAKEAWAY
Prioritise by expected contribution divided by effort, not by how interesting the idea is. The dull items win this calculation almost every time, which is why they are still undone.
What the log should hold
- The hypothesis in one sentence, naming the evidence that prompted it.
- The sample you planned for and the sample you actually got.
- The result, including the losers and the inconclusive ones. An organisation that only records winners learns nothing and repeats itself.
- What you shipped as a result, which is not always what you tested.
EXPERT VIEW
“The most valuable artefact a conversion programme produces is not a lift. It is a written record of what did not work on this store, so the next person does not spend a quarter rediscovering it.”
Dazzle Commerce · Conversion team
- Worked examples
Three decisions, before and after
The three arguments that come up on almost every store we take conversion on.
MERCHANDISING
The default sort nobody chose
Before
Category pages sorted by newest, first row frequently out of stock
After
Sorted by a blend of availability, margin and proven sell-through
The setting applied to every session on every category page and had never been a decision. It was a five-minute change that beat the tests queued behind it.
CHECKOUT
Shipping cost revealed at the last step
Before
Delivery charge first shown on the payment step
After
Threshold stated in the header, cost shown on the product page and in the cart
The abandonment was not caused by the charge but by the surprise. Nothing about the policy changed; only when the customer learned about it did.
TESTING
A winner that was not one
Before
Test stopped on day four at 96% confidence, rolled out site-wide
After
Test resized, run for two full weeks, and recorded as inconclusive
Stopping a test when it looks good is not measurement. The honest verdict freed a month of roadmap that would have been spent defending a lift that was not there.
Anonymised from stores we manage. The shapes recur; the numbers will not match yours exactly.
- Take these with you
Three checklists
Written to be worked through rather than read. Nothing here needs a tool you do not already have.
Diagnose, once a quarter
An afternoon with analytics and a phone.
- Value of a 10% relative lift calculated in contribution
- Five funnel steps measured separately on mobile and desktop
- Conversion split by channel, device and new versus returning
- Twenty session recordings watched on the leaking step
- Top fifty internal search terms read, zero-result terms listed
- Repeated pre-purchase questions and returns reasons collected
Fix before testing
Changes that need a decision, not an experiment.
- Category default sort order chosen deliberately
- First row of every top category in stock
- Delivery cost and date stated on the product page
- Returns terms stated in plain words on the page
- Guest checkout available
- Unavailable variants not selectable
Monthly on a real phone
Twenty minutes, mid-range device, cellular data.
- Category and product templates measured, not the home page
- Largest element on the product template identified
- Consent banner not covering add to cart on small screens
- Checkout completed end to end with a real card
- Apps and tags audited against the pages that use them
- One full purchase attempted in the second most-used browser
Copy these into your own tracker. There is no download form and no email wall.
- Frameworks
Two frameworks worth keeping
The two decisions that get reargued every quarter, written down so they stop being reargued.
FRAMEWORK ONE
The testability triage
Before an idea joins the backlog, decide which of three routes it takes. Most disagreements about CRO are really disagreements about which row an item belongs in, and naming the row ends the argument.
Route
When it applies
How it is judged
Just fix it
A defect, a policy problem, or an obviously correct change
Ship it. Measure before and after, labelled as such.
Test it
Plausible either way, and your traffic can detect the effect
Sized in advance, read once, logged whatever it says
Park it
Effect too small for your traffic to ever resolve
Not tested. Revisit if traffic grows materially.
The third row is the one teams skip, and it is where most testing budget quietly goes.
A change that is obviously right does not need a test to justify it. A test is for when reasonable people disagree.
FRAMEWORK TWO
The conversion priority order
When everything needs doing, the order matters more than the list. Each step assumes the one above it has cleared, because a test at step six is unreadable while step one is wrong.
1
Measurement integrity. Funnel steps tracked per device, revenue reconciled to the platform. Nothing below can be judged without it.
2
Silent failures. Broken payment paths, unselectable variants, banners over buttons. Pure loss, free to fix.
3
Policy. Delivery cost, thresholds, guest checkout, returns. Decisions rather than designs, and usually the largest single lever.
4
Merchandising defaults. Sort order, stock in the first row, filters that match how buyers decide.
5
Content at the decision point. Imagery, specification, delivery date, real answers to real questions.
6
Speed on revenue templates. Measured on a mid-range phone, on the pages that carry the money.
7
Testing. Last, not first — on a store where the six above are already true.
Teams that start at step seven produce a year of inconclusive tests on a store with an unmeasured funnel.
- In summary
Key takeaways
Eight sentences you could hand to somebody who is not going to read the guide.
1
Calculate what a 10% relative lift is worth in contribution before deciding whether this work deserves a quarter.
2
Never judge a blended conversion rate. Split by channel, device and visitor type, because the gap between segments is the finding.
3
Measure five funnel steps against each other rather than against a benchmark. The biggest relative drop is your constraint.
4
The category page decides which product is ever seen, and its default sort order is the highest-leverage setting most stores have never chosen.
5
State delivery cost, delivery date and returns terms on the product page. It is a content change and one of the most reliable improvements available.
6
Four of the five biggest checkout problems are policy decisions. No test fixes a shipping threshold that is wrong.
7
Size every test before running it, read it once, and record the losers. An underpowered test presented as evidence is worse than no test.
8
CRO compounds through cadence and a written log, not through any single winning experiment.
If you want a single first action: open your top category page on a mid-range phone and try to buy something.
- Go deeper
Where this work sits in the practice
Three services that pick up the chapters above.
Funnel measurement, sized testing and the template work that moves conversion on the pages carrying your revenue.
Checkout, theme and performance work — the fixes from chapters five and six that need engineering rather than an experiment.
A review of your store, ad accounts and funnel with findings ranked by revenue impact rather than severity.
- Keep reading
Related guides
The rest of the guide library, all published in full — and the conversion rate calculator if you want the arithmetic in this guide done for you.
Complete guide · 37 min
Where conversion sits in the wider mix, and how to tell whether it is your binding constraint.
Complete guide · 38 min
The other multiplier: what happens after the order, and why it changes what you can pay for traffic.
Complete guide · 38 min
The category pages from chapter three, treated as an organic search problem as well.
Want this run on your store rather than read about it?
Our conversion engagements open by measuring the funnel step by step, then sizing the tests that your traffic can actually settle.