Which Website Metrics Actually Help You Make Better Decisions?
Understand conversion rates, traffic mix, technical performance and measurement errors so that your team can tell a useful improvement from a misleading dashboard.
The Short Answer
Start with the decision you need to make and use a small set of clearly defined measures. Compare similar visitors and conditions, check that tracking is working, and look at the commercial and customer consequences alongside purchase completion.
An overall conversion rate can fall simply because the store attracts more new visitors. A speed score can improve while product information remains confusing. Before-and-after results are useful but do not isolate a design change when promotions or advertising changed too. My preference is a short review that states what happened, how confident we are and what action follows.
Key takeaways
- Define the numerator and denominator before comparing conversion rates.
- Separate changes in traffic mix from changes in website performance.
- Use technical performance as a foundation, not as proof of an effective offer.
- Check tracking and business side effects before declaring a winning change.
A store owner asks, “Our conversion rate is 2%. Is that good?” My first question would be, “Two percent of which visitors, buying which products, under what conditions?” Without that context, a benchmark can send the team toward the wrong work.
Website metrics are useful when they help choose an action. They are less useful when a team collects dozens of numbers but cannot explain what it would change if one of them moved. Start with the business question, keep the definitions consistent and compare similar situations.
This guide explains how I would evaluate website experience for a commerce business. The principles apply across platforms; where Shopify has a specific reporting definition, that distinction is called out.
Decide what success means beyond more orders
The website needs to help the right customer make a good purchase, and the business needs enough money left to serve that customer. A higher purchase rate is not automatically a better result if it came from a discount that erased the margin or from misleading information that increased returns.
Choose one main question for the change you are evaluating. For a product-information improvement, it might be whether comparable visitors are buying the intended product more often without a rise in avoidable returns. For a registration repair, successful account creation is the direct outcome. For a delivery-information change, repeated support questions may be especially useful.
Then choose a few checks that would reveal a harmful side effect. A bundle test might increase order value but make fulfillment more expensive. A sticky purchase button might receive more taps while hiding necessary information. The extra metric should protect a real business or customer interest, not simply make the dashboard look thorough.
Use a small set of numbers with clear meanings
| Measure | What it helps you understand | What it does not establish by itself |
|---|---|---|
| Purchase completion for comparable visits | Whether a similar audience is completing the buying journey | Why a visitor left, or whether the order was profitable |
| Cart and checkout progression | Where to investigate an interruption | That the visible interface caused the interruption |
| Net revenue and order-related costs | Whether the resulting orders make commercial sense | The isolated effect of one design change |
| Support questions and return reasons | Whether customers understood what they bought | The experience of customers who never contact you |
| Loading, responsiveness and layout stability | Whether technical behavior makes the website harder to use | Whether the offer or product explanation is convincing |
Write the definition beside each metric. “Checkout conversion” can mean completed orders divided by checkout starts, while “store conversion” can use all sessions as the denominator. Those are different questions. A session is a visit as defined by the reporting system; it is not the same as a person.
Shopify’s report definitions also distinguish purchasing sessions from order counts. One session can contain more than one order. Use the definition shown in the report instead of mixing its rate with a different calculation in a spreadsheet. See Shopify’s behavior-report definitions.
Compare the same kinds of traffic before judging the design
Imagine a store with two groups of visits. This is an illustrative calculation, not a client result or industry benchmark. Returning visitors generate 800 visits and 80 purchasing visits. New visitors generate 200 visits and 4 purchasing visits. The combined rate is 84 divided by 1,000, or 8.4%.
Next month, both groups keep exactly the same individual rates, but the mix changes: returning visitors generate 200 visits and 20 purchasing visits; new visitors generate 800 visits and 16 purchasing visits. The overall rate is now 3.6%. The website did not become worse in this example. The business reached more people who were less ready to buy.
This is why I would examine at least device, traffic source, market and new versus returning customers where the reporting supports those distinctions. Do not split the data into so many tiny groups that a single order dominates every result. Begin with the few differences most likely to affect the decision.
Keep a note of stock availability, pricing, promotion dates and advertising changes. If the best-selling size sold out, the product page’s performance may decline even though its design did not change.
Give technical benchmarks their proper role
Core Web Vitals describe three parts of website experience: how quickly the main content appears, how promptly the page responds to an interaction and how much the layout unexpectedly shifts. Google’s current good-experience thresholds are LCP at or below 2.5 seconds, INP at or below 200 milliseconds and CLS at or below 0.1, assessed at the 75th percentile of visits. The Web Vitals documentation explains the measures and assessment.
A synthetic speed test is useful for finding technical problems under its test conditions. It is not the same as data from actual customer visits. Keep those two views separate when a developer explains progress.
I would treat a passing technical score as a foundation. The store still needs accurate product information, useful navigation, working controls and a credible offer. Improving a speed score while making the product harder to understand is not a complete improvement.
Check that your measurement still describes reality
Before interpreting a sudden change, ask whether the tracking changed. A purchase event firing twice can look like a successful redesign. A broken cart event can look like a customer-experience problem. Consent choices, blockers and different platform definitions can also make reporting systems disagree.
Repeat a known customer action and inspect whether the corresponding event appears as expected. Compare order records with reported purchases while allowing for timing and the definitions used. They need not match perfectly, but a large unexplained difference should be investigated before making decisions from the rate.
Google Analytics treats events as records of interactions or occurrences on a site or app. Naming a click “purchase” does not turn it into a real transaction. The implementation must represent the action you intend to measure. Google’s event overview provides the underlying distinction.
Keep one written definition for the main outcome so that the owner, marketer and developer are comparing the same thing. If it changes, mark the date instead of silently joining the two series.
Judge a change with the strength of evidence you actually have
A before-and-after comparison can be useful, but it does not automatically prove cause. If the team changed the page, introduced a discount and increased branded advertising in the same week, the result belongs to that combined situation. It should not be presented as the isolated gain from the page design.
A controlled test can provide stronger evidence when the traffic, implementation and test design support it. It is still possible to draw the wrong conclusion by stopping as soon as the result looks positive or changing the main outcome halfway through. Agree on the decision rules before viewing the result.
For a low-traffic store, concentrate first on things you can verify directly: broken controls, missing product facts and repeated misunderstandings. Use customer observations and a longer period of comparable business data. Avoid announcing a universal winner from a few purchases.
Turn the review into a decision
I would end the weekly review with a short statement: what changed, what we think it means, how confident we are and what we will do next. For example: “More shoppers opened the size guide, but purchases are too few to judge. The reported confusion is lower, so we will keep the clearer explanation and continue observing returns.”
That is more useful than declaring the website healthy because a single number is green. If the team has plenty of metrics but no clear problem to investigate, return to the customer-journey audit. The point of measurement is to improve the next decision, not to replace judgment with a dashboard.
Sources
Still have a question?Ask the DTCUP Community ↗

