Case 02 · Kesha Select · Conversion · Jun to Aug 2026
Three initiatives that lifted cart-to-checkout 29%
Add-to-cart was healthy and checkout completion was normal, yet four out of five carts never reached checkout. The leak sat in the cart itself. Two months, three initiatives, no developer sprint.
Artifact and decisions
one issue · 3 linked fixes · numbers match the callouts
The issue
Four in five carts never start checkout
Add-to-cart and checkout completion are both normal for the category. The loss sits in the cart drawer itself, so every fix in this group lives there.
Simplified recreation in English. Item names and prices are illustrative; the mechanics are the real ones.
One sitewide threshold with a progress bar that names the gap, replacing three per-campaign minimums nobody could see.
Upsell moved to the product page and the post-purchase page, where it adds instead of interrupts.
Orders from one campaign merged after close; excess shipping refunded as a gift card, which also brings the customer back.
Locate the leak before theorizing about it. The step rates isolated it: add-to-cart and checkout completion were both normal for the category, so the loss sat between them.
step conversion, Apr to Aug 2026 · overall about 2%
Set up Microsoft Clarity with cart-page segments and rage-click filters, pulled the Shopify funnel by device, and added the cart-to-checkout step to the daily dashboard fed by the order sync. Then three sources that fail differently: 57K session recordings, 15 VIP interviews, and a screenshot one customer sent of her cart.
Five hypotheses written before opening a recording, so the evidence could rule things out. Three survived, and together they described the cart a customer actually saw: a per-campaign minimum she could not see, a grey untranslated button when she missed it, recommendation cards pulling her away, and a second order that charged shipping twice.
Ranked by cost and confidence into three tiers. Tier A: a copy change or a setting, this week. Tier B: layout changes with the designer. Tier C: anything needing an app or a developer. Every Tier A item shipped before a single Tier C item was discussed. The owner approved the list on one page.
Traced which threshold logic lived in the theme, which in a discount app and which in Shopify settings, so the ranking reflected where each change actually had to be made. Two overlapping apps were the source of the threshold states and, later, a double-discount incident.
One threshold, explained
Tier A
Recommendations out of the cart
Tier B
Merge split orders
Tier A · ops
metric: sessions reaching checkout ÷ sessions with an add-to-cart
The jump in multi-order customers is the signature of the threshold change: people stopped abandoning and started splitting on purpose, which is why Initiative 3 exists. Shop Pay users also showed up as the strongest repeat signal in the data (80.7% checkout completion, four times the orders); that went on the backlog as a test on new customers, not into this group.
One sitewide threshold, not one per campaign. Cheapest-first held: nothing in the recovery needed a developer sprint. And the next constraint is already visible in the data: navigation. Food SKUs grew from a few dozen to several hundred during the period, search became the second most visited page, and VIPs who did not buy said "can't find it", never "too expensive".
Overlapping apps are a product risk, not just a cost. The pair that produced the threshold states later produced a double-discount incident, which led to a coupon naming convention and a QA checklist for every promotion launch.
What I would not claim
The recovery happened over two months in which campaigns, assortment and season also changed, and group-buy traffic is too bursty for a clean user-level A/B test. I found the leak, ranked the fixes and shipped them. I do not claim sole cause for the number.
Rule this produced
One threshold, not many. Per-campaign minimums create split orders, double shipping and abandoned checkouts.