How to Run a Conversion Optimization Audit for a Shopify Store

A conversion optimization audit ends in a scored finding log

A conversion optimization audit is a time-boxed review of the path from session to order. It ends when you have a scored finding log: which products, landing pages, or funnel stages leaked orders, what evidence supports that, and which three items have an owner. It does not end in a redesign, a theme brief, or a list of page edits.

Keep this separate from the Shopify conversion rate optimization program. That guide is where accepted findings become fixes: copy, proof, collection order, checkout friction, and a 30-day implementation sequence. This audit only decides what is allowed onto that list. If a sentence in the log tells someone how to rewrite a page, it belongs in the other document.

Leave these jobs out of the audit file:

• Product-page rewrites, review placement, size charts, and mobile button fixes.
• Collection reorders, filter builds, and homepage merchandising.
• Checkout field cuts, guest checkout, and payment-method changes.
• A week-by-week optimization plan. That sequence already lives in the optimization guide.

Freeze the window and the evidence packet

Pick a window that still matches the current catalog. Thirty days is a practical default. Use 60 or 90 days when order volume is thin. Write down every sale, stockout, launch, and ad-budget change inside the window before you interpret a rate. A one-week spike is an exception, not a permanent leak.

Export or screenshot this packet before anyone opens a product page:

• Store sessions, orders, and conversion rate for the frozen window.
• The same product-level figures, plus revenue, for products with meaningful traffic, from whatever product report you already trust.
• Sessions by landing page, so you know which collections, campaigns, or other URLs sessions actually start on.
• The four funnel counts below, with a note of whether you used the open or closed view.
• An exception list: products that were unavailable, deeply discounted, or newly launched during the window.

Shopify's Conversion rate breakdown report is the funnel to cite. In the admin, go to Analytics, then Reports, filter Category to Behavior, and open Conversion rate breakdown. The funnel chart supports four metrics, in order: all sessions, sessions with cart additions, sessions that reached checkout, and sessions that completed checkout. Conversion rate over time uses the same stage names and defines conversion rate as the percentage of sessions that resulted in a purchase.

Read the stage names literally. Shopify defines sessions that reached checkout as sessions with user input during checkout, such as a key press or a click. Sessions that completed checkout are sessions where a customer purchased a product. That report does not show abandonment by contact, shipping, and payment steps. Do not write a finding that names one of those steps unless a separate observation, outside this report, supports it.

Record whether the funnel was open or closed

The same report can show an open or a closed funnel. If the log does not say which view you used, the next reader will treat a skipped step as a drop-off.

Sessions by landing page, also in Behavior reports, tells you where sessions start. A steep gap between all sessions and sessions with cart additions can mean people never reached a product worth buying. Log that as a discovery-stage finding. Do not jump to a checkout edit because the store conversion rate looks weak.

Gate every row before it becomes a finding

A row in the export is not a finding. It becomes a finding only after three gates.

1. Sample size. A page with a few dozen sessions and zero orders is a thin sample. A page with a large session count and a conversion rate far below similar in-stock products is an audit target. Set the cutoff before you sort, and write it on the log.
2. Inventory. If the product was unavailable for part of the window, split the dates or drop it from the comparison. A stockout is an exception, not a page finding.
3. Traffic mix. When you can separate paid and unpaid sessions, do it. A product that converts from collection browsing and fails from a cold ad is a traffic-quality note, not automatic proof the page is broken.

After the gates, give every accepted row one classification:

• Wasted demand: meaningful traffic, weak conversion versus peers, confounds cleared.
• Underexposure: strong conversion, thin traffic, product in stock.
• Discovery drop: sessions are high, cart additions are not, and landing-page data points at a collection, search, or home entry.
• Post-cart drop: cart additions are healthy, and reached checkout or completed checkout is not. Leave the cause unspecified unless you observed it.
• Inconclusive: thin traffic, a stockout, a sale, or mixed sources. Park it. Do not assign an owner.

The product conversion rate guide explains how to read that metric without treating every low rate as a page defect. Use it when a row is borderline. Do not paste its advice into the finding.

Write findings that can be rejected

Each accepted finding needs the same fields, in the same order. If a field is blank, the row goes back to inconclusive.

• ID, date, and frozen window.
• Object: product, landing page, or funnel stage.
• Evidence: the counts, the peer comparison, and open or closed funnel.
• Classification from the list above.
• Confound check: stock, discount, launch, and traffic mix, each marked clear or not.
• Rank: estimated revenue at stake, effort, and confidence. Label the revenue figure as an estimate, not a forecast. A workable estimate is sessions on that object, times the gap versus a peer rate you can defend, times price or average order value.
• Owner and handoff date. One primary observation only.

An acceptable line looks like this: Product A, last 30 days, large session count, conversion well below same-price peers, in stock the whole window, paid and unpaid both weak, classification wasted demand, confidence high, owner named. An unacceptable line looks like this: make Product A more premium.

Rank the leak, not the idea. A finding without evidence is a preference, and preferences do not get an owner.

Ship only high-estimate, low-effort, high-confidence findings to the optimization program. Hold theme rebuilds until the log shows the largest leak is actually on a page people reach. If the largest leak is discovery, a new homepage hero is not the audit outcome.

Close the audit and hand off the fixes

When the top three findings have owners, the audit is done. Page, collection, and checkout changes happen after that, in the optimization guide, and only for accepted rows. A discovery or underexposure finding that names a collection should go to collection optimization as a ranking job, not as a template redesign started inside the audit.

Map the stage name to the ecommerce funnel so the team shares one label for discovery versus checkout. Mixing those labels is how a post-cart leak turns into a week of product-page edits.

Repeat the same packet every month, or after a large traffic or catalog change. Keep a one-page close log:

1. Window, exceptions, and the funnel view used (open or closed).
2. Rows that failed a gate, with the gate named.
3. Accepted findings, one classification each, ranked by the estimate.
4. Top three owners. Everything else is deferred, not half-started.
5. Retest date. If the rate did not move and traffic was stable, replace the finding. Do not add a second fix on top of a wrong diagnosis.

If you want the product and collection cuts refreshed between monthly exports, Skymetrics product analytics can keep traffic-without-orders and underexposed winners on a ranked list, so the next audit starts from last month's findings. The evidence rules above still apply. A ranked list is not a finished finding until the gates are checked.