When a site isn't converting, the temptation is to change everything at once: new theme, new photography, new copy. That rarely helps, because you end up with ten changes and no idea which one did anything.
CRO is rarely complicated. It's usually about finding the one or two things that are haemorrhaging money and fixing them in the right order. Something as dull as forced account creation at checkout can sit there for years because someone once decided capturing email addresses mattered more than completing sales, and switching it off is a toggle in settings.
The hard part is knowing where to look.
Start by measuring what's happening
You'd be amazed how many stores don't have basic funnel tracking set up. They know their overall conversion rate but couldn't tell you where people are dropping off.
Before you read a single funnel number, check that the numbers are right. Take last week's orders in your shop admin and compare them with the purchases your analytics recorded for the same week. If the platform shows 100 orders and analytics shows 80, find out why before you trust the funnel. Some gap is expected where visitors decline analytics cookies. A gap far bigger than that, or one that appeared suddenly, points to broken tracking: a checkout change, a pixel that stopped firing. If analytics shows more orders than the platform, look for duplicate purchase events, such as the thank-you page firing again when people refresh it. Reconcile the order counts, check for duplicates and work out which sessions are missing. If the missing ones cluster on one device, browser or traffic source, the rates for that segment are wrong, and you'll fix the wrong thing.
Once the data holds up, you need to know: what percentage of visitors view a product? What percentage add to cart? What percentage start checkout? What percentage finish?
These numbers tell you most of what you need. If people are viewing products but not adding to cart, you've got a product page problem: price, imagery, copy, something. If they're adding to cart but bouncing at checkout, you've got friction there. If they're not even viewing products, your traffic quality might be the issue, or your navigation is failing them.
Then split the funnel three ways before you decide what kind of problem it is.
- By device and browser. If one step collapses on one browser or one phone size and nowhere else, that's a technical fault. Find it, fix it, and stop there.
- By traffic source. If paid social visitors leave at the product page while email and search visitors buy happily, the page is probably fine and the traffic is wrong for it.
- Across everything. If every source and every device stalls at the same step, investigate the offer (price, clarity or trust) and any technical problem they all share, such as a payment or delivery-charge fault that hits everyone.
Those three answers lead to completely different work. A redesign won't fix a broken button on one browser, and a better checkout won't convert people who were never going to buy.
Fix checkout first, not homepages
Designers want to start with homepages. They're wrong.
Someone who's reached your checkout has already decided to buy. They've found a product, liked it enough to add it to cart, and clicked through to give you money. Your only job is to not stop them.
Guest checkout is non-negotiable. Forcing account creation puts a hurdle in front of people at the exact moment they've decided to pay. I don't care about your customer database strategy or your remarketing plans. Let people buy first, offer account creation after. This is a setting change on most platforms and it's probably the highest-return thing in this entire article.
Look at your form fields. Every field is friction. Phone number: do you need it, or does your courier? Company name: unless you're B2B, delete it. Separate first and last name fields: does your carrier even distinguish? The ideal checkout has the bare minimum: email, delivery address, payment details.
Payment options matter more than people realise. If you only take cards, you're missing PayPal users (there are still lots), Apple Pay devotees, and the people who want buy-now-pay-later. Adding Klarna or Clearpay tends to matter more as order values rise, because that's where the full price becomes the sticking point. Check what the provider charges you per order before you assume it pays for itself.
And test your checkout on mobile. Properly test it: go through the whole thing with your thumbs, on a phone with an average screen size, on mobile data. If any part of that is painful, you're bleeding money.
Product pages are where people decide
After checkout is solid, product pages are next. This is where the buying decision happens or doesn't.
Images do more work than copy, and plenty of shops underinvest here. You need multiple angles, context shots showing the product in use, scale references so people can judge size, and zoom that works. For anything complex or mechanical, video is worth considering.
Three photos, all from roughly the same angle against a white background, may do for commodity items. If you're selling anything where quality or design matters, that's not enough.
Pricing clarity is underrated. If you have a discount, show it clearly and strike through the old price. If you offer Klarna, show the instalments: on a £60 item, 'or 3 interest-free payments of £20'. For some buyers that's the difference between 'too expensive' and 'I can do that'. Use the wording your provider gives you, because how you advertise buy-now-pay-later isn't entirely up to you.
Reviews matter, but whether they're your priority depends on what's stopping people. Missing dimensions, unclear delivery information or a buy button that doesn't work can cost more sales than a lack of reviews. Where reviews are the gap, email every customer asking for one, make it a single tap, follow up once. Don't tie any reward to a positive rating. A product with a handful of honest reviews gives a hesitant buyer something to go on; a product with none gives them nothing.
The buy button should be obvious, and sticky on mobile. If someone has to scroll past three paragraphs of copy to find add-to-cart, move it.
Speed matters, perfection doesn't
Slow sites lose sales, and they lose them hardest where you can least afford it: on phones, on patchy connections, and among people arriving from ads who have no particular loyalty to you yet.
Run PageSpeed Insights on your site. The score isn't gospel, but it gives you a sense. The tool labels scores of 0 to 49 as poor, 50 to 89 as needs improvement and 90 or above as good. Those bands describe the lab test, not what a fix is worth: whether to pay for one depends on its cost and what your own data says slow pages are costing you. If your site has enough traffic, look at the field data from real visitors as well as the lab score, because that's what your customers experience.
Speed issues often start with images. People upload huge files straight from a phone or camera. Compress everything, use modern formats like WebP, and enable lazy loading so images below the fold don't block the first render.
Apps are the next suspect, especially on Shopify. Each app can load JavaScript on every page, whether or not that page uses it, and uninstalling an app doesn't always remove the code it added to your theme. Audit your apps. Delete anything you're not using, then check the theme for leftovers.
Themes matter too. Shopify's own free themes are built to be lean, which makes them a useful benchmark: if a stripped-back free theme loads far faster than yours with the same products, your theme is part of the problem. Older themes with lots of animations and sliders are often much slower. Sometimes a theme switch is the right call.
Don't obsess over perfect scores. Getting a slow site to reasonable is where the gains are. Before chasing the last points from 70 to 95, check the field data: a better score on its own doesn't show that visitors notice the change or that it repays the engineering.
Prioritisation: what to fix first
You can't fix everything at once, and not all changes are equal.
Use a simple framework: multiply potential impact by your confidence it'll work, then divide by effort. High impact, high confidence, low effort goes to the top.
Adding guest checkout: big impact, near-certainty it helps, five minutes to enable. Do it today.
Complete site redesign: big potential impact, but uncertain whether it'll improve conversion (sometimes redesigns make things worse), and months of effort. Rarely worth it compared with targeted fixes.
Adding Klarna or Clearpay: moderate impact, reasonable confidence if your order values are high enough for instalments to matter, maybe a day of setup. A solid quick win if the fees stack up.
Run through your list of potential changes this way. The answers usually become obvious.
The trap is wanting to do the big impressive stuff when the boring foundations haven't been addressed. Nobody wants to brag about enabling guest checkout or compressing their images. But that's usually where the money is.
Testing: overrated for small stores
Everyone talks about A/B testing. But you need a lot of traffic for A/B testing to tell you anything.
How much depends on four things: your current conversion rate, the smallest improvement you'd want to detect, how sure you want to be, and how likely you want to be to spot a real effect. A common rule of thumb, at the usual settings of 95% confidence and 80% power, is that each variation needs roughly 16 × p × (1 − p) ÷ d² visitors, where p is your conversion rate and d is the difference you want to detect.
Work it through. Say your conversion rate is 2% and you want to detect a lift to 2.4%, a 20% improvement. That's 16 × 0.02 × 0.98 ÷ 0.004², which comes to 19,600 visitors per variation, or 39,200 in total. At 1,000 visitors a day, that's about 39 days. Want to detect a 10% lift instead? Halving the difference you're looking for roughly quadruples the visitors you need.
So whether to A/B test comes down to traffic, not revenue: can you detect a change of the size you expect in a sensible time? Put your own numbers into a sample-size calculator before you start. If it says four months, don't run the test.
Below that point, make one change at a time and keep a dated log of what changed. Then compare before and after carefully, knowing what that comparison can't do: it can't separate your change from a promotion, a shift in traffic mix, a bank holiday or the weather. Compare like with like (same traffic sources, same devices, a few weeks either side), avoid changing things during a sale, and treat the result as a direction rather than proof. You'll be wrong sometimes. You'll also learn faster than you would agonising over a test your traffic can't support.
When CRO won't save you
CRO can't fix fundamental problems.
If nobody wants your product, optimising your product page won't help. If your prices are 40% higher than competitors without a compelling reason, no amount of urgency copy will overcome that. If you're buying rubbish traffic from irrelevant audiences, the best checkout in the world won't convert them.
Before starting CRO work, ask honestly: is there genuine demand for this product at this price? Is the traffic I'm paying for relevant? If the answer is no, CRO is a distraction.
If the fundamentals are sound (real demand, reasonable pricing, relevant traffic), CRO is about capturing more of what's already there. If the data shows a fault, fix it. If it doesn't, investigate demand before you change the site.
How we help: Conversion Rate Optimisation · Shopify Consulting
Want to talk this through for your own business? Dax reads every message personally and usually replies within a day.
Tell us what you needOr get the next article by email, every couple of weeks.

