Conversion Dropped After the Redesign. Now Find the Cause.

Step 0: Prove the Number Is Real (Half a Day)

Most redesigns ship a new analytics implementation alongside the new front end. New templates, new dataLayer, new consent banner, sometimes a new GTM container. Which means the first question is not "what did we break in the UX" but "did we break the measurement."

Do this before anything else:

• Reconcile against the source of truth. Pull order count and revenue from the platform back end (Magento, ERP, payment provider) for the 14 days before launch and the 14 days after. Compare against GA4 purchase events for the same windows. If the back end shows a 2% dip and GA4 shows a 22% dip, you have a tracking problem, not a conversion problem. This single check resolves a surprising share of "emergencies."
• Check the denominator, not just the numerator. Conversion rate has two moving parts. A new consent management platform that blocks analytics until interaction, a changed cross-domain linker, or a redirect chain that starts a fresh session can inflate session counts and deflate conversion rate without a single lost order.
• Verify event integrity step by step. view_item, add_to_cart, begin_checkout, add_payment_info, purchase. Not just "does it fire" but does it fire once, with the right parameters, on every template variant, on mobile, and with consent granted and denied.
• Look for duplicate or missing transaction IDs. Deduplication breaking is the classic quiet regression after a checkout rebuild.
• Check the date boundary itself. Deploys rarely happen at midnight. Comparing "before" and "after" using calendar days when the release landed at 3pm Thursday will smear the signal across two periods and hide the size of the step change.

If the drop is a measurement artifact, you are done in a day. If it survives this step, it is real and you continue.

Step 1: Rule Out the Traffic (2 Hours)

The second cheapest explanation is that your visitors changed, not your site.

Hold the mix constant before you interpret anything. The specific trap here is Simpson's paradox: overall conversion rate can fall while every single segment holds steady, purely because the share of low-converting traffic increased. If mobile went from 60% to 72% of sessions during the same fortnight - because a campaign shifted, or because a new PDP started ranking on mobile queries - you will see a blended drop that has nothing to do with the redesign.

So calculate conversion rate within each segment and compare like with like:

• New vs returning
• Device category
• Channel and campaign
• Country and locale
• Logged-in vs guest, and, if you run B2B, dealer vs retail pricing groups

Also check whether the redesign changed traffic quality upstream. New templates mean new title tags, new internal linking, sometimes new URLs. If organic landing pages shifted from commercial pages to blog content, intent dropped and conversion rate followed - correctly.

Add a year-over-year control. Pre versus post is contaminated by seasonality and the promo calendar. Comparing the same fortnight last year gives you a rough sense of how much of the delta was going to happen anyway.

Step 2: Localize the Leak, and Read What the Shape Tells You

Now build the funnel: sessions, product view, add to cart, begin checkout, shipping, payment, order placed. Compute step-to-step conversion for before and after, per device. (For where those step rates should sit in the first place, see our checkout completion rate benchmarks.)

The shape of the drop is diagnostic, and this is the part most checklists skip:

Two places to look that people routinely forget:

On-site search. Redesigns frequently ship a new search implementation, and internal search users convert at multiple times the rate of browsers. If relevance or the autocomplete broke, you lose your highest-intent segment silently, and nothing in the checkout funnel will point at it. Check search exit rate and no-results rate before and after.

Faceted navigation and filters. Same logic. Slower, fewer, or relocated filters on a large catalogue kill product discovery for exactly the people who were ready to buy.

Step 3: The Device and Browser Matrix (Half a Day)

Build a small grid: iOS Safari, Android Chrome, desktop Chrome, desktop Safari, Firefox, plus whatever else shows up meaningfully in your data. Conversion rate before and after in each cell.

A regression isolated to one cell is almost always code, not design. Then go find it:

• JavaScript error rate by browser, before and after. A single uncaught exception in an add-to-cart handler on one Safari version is worth more than a week of UX debate.
• Failed network requests and 4xx/5xx on checkout endpoints.
• Third-party script load order. Redesigns reshuffle it, and a payment SDK or address autocomplete that now loads after the form initialises will fail intermittently.
• Tap target sizes, sticky header height eating the viewport, and whether pinch zoom is still enabled.

Baymard's abandonment survey puts "website had errors / crashed" at 17% of abandoners once you exclude the people who were only browsing. A redesign is, structurally, a mass-production event for exactly that failure mode. We wrote about hunting these silent breakages in detail in our Magento checkout abandonment guide.

Step 4: Diff the Design Against What Was Earning Money

This is where a checkout specialist earns their fee, and it is not "the CTA is less visible." It is an inventory exercise.

Take the old page and the new page side by side, and list every element that has disappeared, moved below the fold, or become one click further away. Not the elements you were thinking about - all of them. Redesigns lose things by accident: a component gets dropped in the rebuild, and nobody notices because nobody had written down that it was load-bearing.

The elements most worth auditing, in rough order of how often they cause damage:

Cost transparency. Extra costs being too high is the single largest recoverable abandonment reason - 40% of abandoners, excluding the just-browsing segment (Baymard). But the underlying failure is usually when the cost appears, not what it is. If shipping cost, tax, or fees now surface one step later than before, or the free-shipping threshold message no longer appears in the cart, you have added a surprise. A further 12% of abandoners cite not being able to see or calculate the total up front.

Guest checkout. 18% of abandoners cite being asked to create an account. Redesigns love a clean "Sign in or create an account" screen. If guest checkout is now a secondary link, or below the fold on mobile, or worded as "continue without an account" next to a big primary "Create account" button, you have introduced a wall.

Form field count. Baymard's testing puts an ideal checkout at 12-14 form elements, roughly 7-8 actual fields, against a benchmarked US average of 23.48 elements. New design systems add fields quietly: a separate "confirm email," a split first/middle/last name, an always-visible "Address line 2," a mandatory phone number, a company field on a B2C flow. Count them. Before and after.

Address handling. Did address autocomplete survive the rebuild? Did postcode auto-detection of city and region survive? Is the new validation rejecting formats it should accept - Polish and UK postcodes, Irish Eircodes, addresses without a house number? This is the failure mode that produces zero complaints and a large hole, because affected users cannot complete and cannot explain why.

Payment method visibility. 9% of abandoners cite insufficient payment methods, and express wallets are frequently casualties of a redesign. Is the Apple Pay or PayPal express button still on the product page and in the cart, not just at the payment step? Are card icons still displayed?

Trust and reassurance. 19% of abandoners say they did not trust the site with their card details. Trust is carried by unglamorous components - security copy near the card field, a visible returns window, delivery estimates, review counts, contact information, order cut-off times. These are exactly the components a designer removes to reduce clutter.

Error and validation behaviour. If the new build replaced inline validation with on-submit validation, or clears entered data on error, or shows a generic error at the top of a long form, you have converted recoverable mistakes into abandonments. This is worth testing by hand: deliberately submit a bad card number, a mismatched billing address, an out-of-stock item, and an expired coupon, and see what the user is left with.

Coupon field prominence. A newly prominent promo code field on the cart page sends people off-site to hunt for a code. Some of them do not come back.

Step 5: Watch Twenty Sessions (Half a Day)

Only now do you open session replay - because now you know which step to watch, which means twenty sessions is enough instead of two hundred.

Filter to: the affected step, the affected device, sessions that entered the step and did not complete it. Watch twenty. Then watch five that did complete, as a control.

What you are looking for is not "friction" in the abstract. It is specific: rage clicks on a non-interactive element, repeated attempts on the same field, scroll-hunting for something that used to be there, a keyboard covering the field being typed into, a form that resets. Pair it with form field analytics if you have it - time per field, error rate per field, abandonment by field - which turns "checkout feels worse" into "63% of mobile users who touch the phone field never submit."

Twenty replays chosen well beat a week of unfiltered browsing. This is why replay comes fifth and not first.

Step 6: Performance (2 Hours)

Redesigns ship heavier pages. Check Core Web Vitals on real user data, not lab data, split by device, before and after: LCP, INP, CLS on homepage, category, product, cart, checkout.

The things that break: a hero image no longer preloaded, a web font blocking render, a client-side rendered price or availability block, lazy loading applied above the fold, a third-party tag added during the launch. Mobile INP regressions in particular read to users as "the button doesn't work," and they respond by tapping again or leaving.

Now You Have Hypotheses. Testing Them Is the Hard Part.

Every generic diagnostic list ends with "run an A/B test." That advice does not survive contact with a redesign, because you already rolled out to 100% of traffic. There is no control group left. You cannot A/B test your way backwards out of a big-bang release.

What you actually have:

1. Fix and measure forward. For a clear technical defect - a JS error, a broken validation rule, a missing express payment button - do not test it. Fix it, and watch the specific step-conversion metric for that specific segment. The pre-fix period is your baseline.
2. Re-introduce, don't revert. Instead of rolling back the whole redesign, restore individual removed components onto the new design one at a time, with a few days between each. Slower, but it produces attributable evidence and it does not undo work that was fine.
3. Partial rollback as a diagnostic tool. If one template is the prime suspect, serve the old template to a slice of traffic. On Magento this is straightforward per store view or per customer group. Ugly, and worth it if the daily revenue loss is large.
4. Use a structural split. If you run multiple store views, locales, or brands and only some got the redesign, you already have a natural experiment. Use it.
5. Five moderated usability sessions. For anything where the mechanism is unclear - "why do people stall on the shipping step" - five sessions with real users on real tasks will tell you more than another week in the analytics. Cheap, fast, and it produces the recording that ends the internal argument.

Prioritise by Expected Money, Not by How Annoying It Is

For each candidate issue, estimate:

Affected sessions per month × step-drop delta × close rate to order × AOV = monthly revenue at risk

Then rank by that figure divided by engineering effort, and discount by your confidence in the evidence. Confidence matters: a documented JS error on 8% of sessions is not the same class of finding as a designer's opinion that the CTA has lower contrast.

Two rules that keep this honest:

• Fix the specific cause, not the whole redesign. "Revert everything" is a decision made to end discomfort, not to recover revenue.
• Write down what you expect each fix to do before you ship it, and check afterwards. A fix list where nobody recorded predictions becomes folklore within a quarter.

Where the Standard Checklist Gets It Wrong

If you have already asked an AI or a generalist agency for a diagnostic plan, you have a list of thirty questions. The list is not wrong, it is unordered - and three specific gaps cost real money:

1. It treats all questions as equally cheap. Reconciling GA4 against the order table takes two hours and resolves a meaningful share of cases. Usability testing takes two weeks. Sequencing by cost is the entire value of the process.
2. It recommends A/B testing something already at 100% rollout. See above.
3. It never mentions the traffic mix. Segment-level analysis is listed as "who is affected," but the real point is that blended conversion rate is not a reliable signal when the mix moved. Teams have spent quarters redesigning their way out of a problem that was a shift in mobile share.

The Actual Point: This Should Not Have Been a Mystery

The reason a post-redesign drop is hard to diagnose is that it was shipped as a reveal instead of as a deploy. Fix that for next time:

• Instrumentation parity as a release gate. Every conversion event verified on the new templates in staging, plus a reconciliation check in the first 24 hours after launch. Not a task on someone's list - a gate.
• Staged rollout. 5%, 25%, 50%, 100%, with guardrail metrics at each step. Every one of the diagnostics above is trivial when you have a concurrent control group.
• A written inventory of load-bearing elements before the rebuild starts: trust signals, shipping messaging, express payment placements, guest checkout entry point, address autocomplete, field count per step. This is what stops components from disappearing by accident.
• Guardrails with a named owner and a rollback trigger defined in advance, in numbers. "Checkout entry rate drops more than 5% on mobile for 48 hours" is a trigger. "It feels worse" is a meeting.

Baymard's benchmarking of 344 leading sites finds 65% of checkout flows performing at "mediocre" or worse, with an average of 32 documented improvements available per site, and a potential conversion gain of 35.26% from checkout design alone. Redesigns are not the problem. Redesigns shipped without a control group, without instrumentation parity, and without an inventory of what was already working - that is the problem.

The 48-Hour Triage Checklist

Print this. Work top to bottom. Stop when you find it.

Hours 0-4

• Back-end orders and revenue vs analytics, 14 days pre and post
• Session count integrity: consent banner, cross-domain, redirects
• Purchase event dedupe and transaction IDs
• Exact deploy timestamp, re-cut the periods around it

Hours 4-8

• Conversion rate by device, channel, new/returning, locale - not blended
• Traffic mix shift check
• Year-over-year control
• Organic landing page mix before and after

Hours 8-16

• Full funnel, step by step, per device, pre vs post
• Read the shape (see table above)
• On-site search: usage, no-results rate, exit rate
• Category and filter usage

Hours 16-24

• Browser × device conversion matrix
• JS error rates before and after, by browser
• Failed requests on checkout endpoints
• Core Web Vitals, field data, by template and device

Hours 24-36

• Element-by-element diff of what disappeared or moved
• Form element count, before and after, per step
• Guest checkout entry point, on mobile, above the fold
• Express payment placements on PDP and cart
• Manual error-path testing: bad card, bad address, out of stock, dead coupon

Hours 36-48

• Twenty session replays on the identified step and device
• Form field analytics on the identified step
• Ranked fix list with revenue estimates, effort, and confidence
• Written predictions for the top three fixes