Start with a number you can explain
Traffic with few purchases is a symptom. Before changing the theme, confirm what your traffic and purchase numbers represent. Compare Shopify orders with the purchase events in your analytics over the same dates, time zone and currency. Exclude test orders and document refunds and cancellations. Check consent behavior and whether a purchase event fires twice. Different systems have different attribution rules, so disagreement does not automatically mean one is broken.
Write down what is reliable, what is missing and what cannot yet be compared. If add-to-cart events are unreliable, repair those events before claiming a template has a low add-to-cart rate. You can still inspect the shopping journey and document reproducible problems while measurement is repaired. Shopify explains its conversion reporting and funnel definitions.
Shopify also updated session measurement in September 2026. A reporting-method change can alter session counts and conversion rates without changing orders. Record that change when comparing periods across the rollout.
Ask whether the visitors can reasonably buy
Segment traffic by source, campaign, landing page, country and device. A broad informational campaign is different from a product-intent search. Check whether ads promise a product, price or delivery option that the landing page actually supports. Look for bot traffic, unavailable shipping regions and out-of-stock variants before interpreting a site-wide average.
For example, a campaign for a particular knife series should land shoppers where they can identify that series and choose an available product. Sending everyone to a general homepage can add a decision the ad already appeared to resolve. This is a diagnostic example, not a measured Configurate result.
Find the first unresolved product question
Walk the main landing and product templates on a phone. Can a new shopper tell who the product is for, what differs between variants, what is included and when it can arrive? Check dimensions, materials, care requirements, compatibility and returns against approved source information. A visible purchase button cannot compensate for a missing answer.
Our Cangshan project included product information, series comparisons and reusable content sections. That delivered work supports a practical approach: structure the information shoppers need to choose and give the merchant a way to keep it current. It does not establish a quantified conversion effect.
Trace add to cart through checkout
Test selecting variants, unavailable combinations, quantity changes, adding from a product page, opening the cart and proceeding to checkout. Try back navigation and a return from checkout. Check whether the selected variant and price remain correct. Watch for discount messages that contradict checkout, unexpected shipping information and third-party widgets that cover controls.
Separate a broken interaction from a persuasion hypothesis. A button that does not add the selected variant is a reproducible defect. A belief that a different button label will improve sales needs testing. Checkout customization depends on the store's plan and platform permissions, so establish those constraints before proposing a change.
Compare devices without blaming mobile
Compare the same traffic source and landing-page intent on mobile and desktop. A lower mobile purchase rate may reflect different visitors or buying stages as well as interface friction. Observe product-option selection, text wrapping, sticky controls, cart drawers and keyboard behavior on actual supported devices. Record viewport, browser and consent state so another person can repeat the problem.
Use performance measurements alongside the journey checks. A slow product image and an unclear variant label are separate findings, even if they affect the same template.
Choose the next change by evidence and consequence
For each finding, record the page, affected journey, evidence, likely consequence, confidence, effort, dependencies and release risk. Fix broken purchasing and missing measurement before running subtle copy experiments. Do not turn a guessed revenue impact into a forecast. With small samples, report uncertainty instead of declaring a winner.
A useful first work item might be: add approved dimensions beside a confusing size selector, verify every variant on mobile and desktop, then compare product-view-to-add-to-cart behavior over a defined period with campaign mix documented. The work item needs an owner and acceptance criteria, not just an idea.
Focused fix, experiment or rebuild?
Use a focused fix when the obstacle can be isolated in the current theme. Use an experiment when two plausible choices exist and traffic, measurement and operating capacity support a fair comparison. Consider a larger rebuild when repeated changes are blocked by the architecture or the required shopping model cannot be supported responsibly.
A rebuild introduces more simultaneous changes and makes attribution harder. Document that tradeoff. Our CRO implementation work connects agreed priorities to preview, QA and measurement. An audit can also be handed to an existing agency or internal developer.