Ecommerce CRO glossary
29 terms that come up in the actual work, defined in one sentence each, then explained in the way that matters: what it means for your revenue and where people get it wrong. Free, like the rest of it.
Metrics
The numbers a testing programme is actually managed against.
- Conversion rate
-
The share of sessions that end in an order, usually written as a percentage.
Orders divided by sessions. It is the headline number in ecommerce and the easiest one to move dishonestly: a sitewide discount lifts conversion rate while destroying margin. Treat it as one input, never as the goal on its own.
Why conversion rate is the wrong number to worship - Revenue per session
-
Total revenue divided by total sessions: the metric that cannot be gamed with a discount.
Because it multiplies conversion rate by average order value, a change that lifts orders while shrinking baskets shows up as flat or down. This is the number we optimise for, and the one that decides whether a test is called a win.
How revenue per session ties the numbers together - Average order value (AOV)
-
Revenue divided by number of orders.
Moved by bundling, thresholds, upsells and merchandising rather than by persuasion. A free-shipping threshold is the most common lever, and the most commonly set at the wrong number.
Free shipping threshold calculator - Add-to-cart rate
-
The share of product page views that result in an item added to the cart.
A diagnostic, not a goal. A rise in add-to-cart with no rise in orders usually means the cart or the shipping cost is doing the rejecting, and the real problem moved rather than went away.
- Guardrail metric
-
A second metric watched during a test to make sure a win is not being bought with something else.
Order value, refund rate and margin are the usual guardrails. A test that lifts conversion 6% while cutting order value 9% is a loss, and reporting it as a win is how agencies keep contracts they have not earned.
Reading a result honestly - Bounce rate
-
The share of sessions with no meaningful second interaction.
Weak on its own and often misread. A high bounce rate on a page that answers a question completely is fine; on a paid landing page it usually means the ad promised something the page did not deliver.
Testing and statistics
The vocabulary of an experiment that can actually be trusted.
- A/B test
-
A controlled experiment splitting live traffic between an unchanged control and one changed variation.
The split is random and simultaneous, which is what separates a test from a before-and-after comparison. Everything that makes a test trustworthy (sizing, duration, guardrails) exists to protect that comparison.
How A/B testing actually works - Statistical significance
-
A measure of how likely a difference this large would be if the change did nothing at all.
Usually reported as a p-value or a confidence level. Significance says a result is unlikely to be noise. It does not say the result is large, durable, or worth the engineering time. Those are three separate questions people routinely collapse into one.
Check a result - Sample size
-
The number of visitors a test needs before its result means anything.
Calculated before the test is built, from your baseline rate and the smallest lift worth detecting. If the required sample will not arrive within about eight weeks, the honest options are to ship the change without testing or to test something bigger.
Sizing a test before you build it - Peeking
-
Checking a running test repeatedly and stopping when it looks good.
The single most common way stores fabricate wins without meaning to. Every look is another chance for random noise to cross the line, so a test stopped on the first good day is close to worthless.
Why peeking lies to you - Minimum detectable effect (MDE)
-
The smallest improvement a test is capable of detecting at your traffic level.
Low-traffic stores have a high MDE, which is why they cannot test button colours: only a very large change could ever register. Knowing your MDE tells you which ideas are testable and which must simply be decided.
- Inconclusive result
-
A finished test that cannot distinguish the variation from the control.
The most common outcome of a well-run programme, and not a failure. It means the change was too small to matter at this traffic, which is itself a decision: stop spending on that idea.
What to do with an inconclusive result - Hypothesis
-
A testable statement linking an observed problem to a proposed change and an expected effect.
Good ones name the evidence: "because reviews repeatedly ask about sizing, adding a fit guide to the gallery will raise add-to-cart." Bad ones name a preference: "a cleaner hero will convert better."
Hypotheses that deserve traffic - ICE and PIE
-
Prioritisation frameworks that score ideas on impact, confidence and ease (ICE) or potential, importance and ease (PIE).
Useful for forcing a conversation, dangerous when the scores are treated as data. Every input is an estimate, so the output is an opinion wearing a number. Use them to rank, never to justify.
Prioritising without the theatre
Research
How you find out what is broken before you have an opinion about it.
- Conversion research
-
The structured process of finding where and why a store loses buyers, before proposing any fix.
Combines quantitative sources (funnels, device splits) with qualitative ones (reviews, support tickets, recordings, surveys). The output is a ranked list of problems, not a list of ideas.
The research chapter - Review mining
-
Reading your own and competitors’ product reviews to find the objections that cost orders.
The cheapest research available and the most consistently skipped. Buyers explain in public exactly which question the product page failed to answer before they bought anyway, or did not.
- Session recording
-
A replay of one visitor’s interactions with your store.
Best used to explain a number you already found, never as a browsing activity. Watching fifty recordings without a question in mind produces anecdotes and a strong, unearned sense of certainty.
- Heatmap
-
An aggregate visualisation of clicks, movement or scroll depth on a page.
Good for finding what people never reach and what they click that is not a link. Weak as evidence of intent, and easily misread when the sample skews to one device.
- Funnel analysis
-
Measuring the drop-off between each step from landing to purchase.
Tells you where the leak is, not why. Its main value is deciding where to point the qualitative research, and it should always be split by device before anyone draws a conclusion.
Shopify and page types
The parts of a store, and what each one is actually for.
- Product detail page (PDP)
-
The page for a single product, where most ecommerce buying decisions are made or lost.
Its one job is to answer the buying questions in the order a buyer asks them: is this right for me, what does it cost delivered, what happens if I am wrong.
Product page optimisation - Collection page
-
A Shopify category page listing products that share an attribute.
Usually treated as navigation and used as a landing page. Because paid and organic traffic lands on them directly, they carry more conversion weight than most stores give them.
Navigation and collections - Cart drawer
-
The slide-out cart that appears after an add-to-cart without a full page load.
Keeps momentum but hides the cart page, which is where shipping expectations are often set. Whether a drawer beats a cart page is genuinely store-specific and worth testing.
Cart and checkout - Shopify Plus
-
Shopify’s enterprise tier.
Relevant to CRO mainly because it opens up checkout customisation and higher API limits. It does not change where most losses happen, which is still before the checkout.
- App bloat
-
Accumulated third-party Shopify apps whose scripts slow the storefront.
Each app adds requests and often render-blocking JavaScript. The conversion cost of a slow page is real and compounding, and reviewing installed apps is usually the cheapest speed work available.
Theme performance and app bloat - Core Web Vitals
-
Google’s published set of page-experience measurements: loading, interactivity and visual stability.
Largest Contentful Paint, Interaction to Next Paint and Cumulative Layout Shift. They matter to conversion for the obvious reason (people leave slow pages) more than for any ranking effect.
Paid traffic and landing pages
Where money is spent, and where it leaks fastest.
- Message match
-
The degree to which a landing page repeats the promise, wording and imagery of the ad that sent the visitor.
The cheapest fix in paid ecommerce. A visitor who clicked a specific claim and lands on a generic homepage has to re-orient, and a meaningful share simply leave.
Landing pages for paid traffic - Awareness level
-
How much a visitor already knows about their problem, your category and your product.
Decides the page structure. Someone who does not know the product category needs an argument; someone who searched your brand name needs a buy button and shipping information.
- Advertorial
-
A long-form editorial-styled landing page that sells by explaining rather than by displaying.
Effective for unaware and problem-aware traffic where a product page has nothing to work with. Requires disclosure and genuine substance; the version that pretends to be journalism is both a legal and a brand risk.
- Customer acquisition cost (CAC)
-
What it costs, all in, to acquire one new customer.
CRO is the lever that lowers CAC without touching the ad account: the same spend, more of the clicks converting. This is the whole commercial argument for the work.
Know the words. Want the work done?
Twenty minutes on your numbers. We tell you where the store is leaking, what we would fix first, and whether it is worth hiring anyone at your traffic.
Got it. Now pick a time.
Open the calendar →Twenty minutes on your store and your numbers. If we cannot help we will say so.
- 437+ stores built
- 2,000+ tests shipped
- No call needed to start