The trade literature on e-commerce UX is, by 2026, sufficiently large and sufficiently repetitive that the dominant patterns it recommends have acquired the status of canonical practice. The canonical practice is, on the data I have available from approximately one hundred retail sites at Praxis Digital 3D, partly correct and partly incorrect, in proportions that vary by category and by audience composition. The patterns that consistently produce conversion uplift are, in my observation, a smaller subset of the canonical recommendations than the literature implies; the patterns that produce no measurable uplift, despite the literature’s confident endorsement, are correspondingly a larger fraction.
Easy E-com UX is the checklist I have assembled, over the course of approximately four years of consulting on retail UX engagements, of the patterns that the data supports and the patterns that the data does not. This post is an account of the checklist in some detail, with the conversion-data evidence for each pattern’s inclusion or exclusion.
The patterns that consistently work
Five patterns produce measurable conversion uplift on essentially every retail site I have measured. The patterns are not novel; the consistency with which they work is the property worth noting.
The first is sticky add-to-basket on the product page. A button that remains visible in the viewport regardless of the visitor’s scroll position consistently outperforms a button that scrolls out of view, with conversion-rate improvements typically in the range of three to seven per cent. The mechanism is straightforward; the button is, by hypothesis, the page’s principal commercial action, and the visitor’s ability to perform the action without re-orienting the page is consequential for whether the action is performed. The pattern is canonical and the data supports it.
The second is the visible price comparison on category pages. The display of a strikethrough original price alongside a discounted current price produces measurable uplift on essentially every retailer I have worked with, with the uplift concentrated on the items that bear the strikethrough rather than on the category as a whole. The pattern is, again, canonical; the data supports it; the only caveat worth mentioning is that the strikethrough produces no uplift unless the discount is genuine, and the search engine’s increasingly aggressive efforts to identify and penalise fictitious discount displays have, in the previous twelve months, made the pattern operationally riskier than it previously was.
The third is the trust-signal cluster near the basket button. A small visual cluster including the retailer’s return policy, delivery options, payment-security badges and customer-service contact information produces measurable uplift on the conversion of the visitors who reach the product page. The cluster’s individual elements are not, on the data, decisive; the aggregate visual signal of the cluster is what produces the uplift, with the implication that the cluster should be visually unified rather than presented as a collection of separate widgets.
The fourth is the post-add-to-basket continuation prompt. When the visitor adds an item to the basket, the page should present the visitor with two distinct options — to proceed to the basket and complete the purchase, or to continue browsing — rather than committing the visitor to one or the other implicitly. The presentation of the options consistently produces a higher proceed-to-basket rate than either implicit pattern, with the explicit-continuation pattern producing approximately six to nine per cent uplift on the basket-progression rate.
The fifth is the explicit display of in-stock quantities, where the quantity is sufficiently low to credibly motivate urgency. The pattern is canonical and the canonical caveat applies: the display produces uplift only when the quantity is genuinely low; the display of a fictitiously low quantity produces, in the data I have collected, an active regression as visitors recognise the manipulation and lose confidence in the broader site.
The patterns the literature recommends but the data does not
Five patterns are, on the data, less productive than the trade literature claims them to be.
The first is the live-chat widget. The widget is sold as a high-value conversion intervention; the data, on the deployments I have measured, shows it producing essentially no measurable uplift on conversion rate when the widget is anchored to the bottom-right of the viewport in its conventional position. There exists a class of better-targeted live-chat implementations — typically those that surface only on the cart and checkout pages, and only after a defined dwell time — which do produce uplift; the conventional pattern of the widget being available on every page produces, on the data, no measurable improvement and occasionally produces an active regression as the widget’s visual presence is treated by visitors as a signal that the site requires assistance to use.
The second is the persistent promotional banner across the top of every page. The banner is sold as an opportunity to communicate the current promotion at every page-view; the data shows that visitors habituate to the banner within approximately two pageviews and that the banner’s contribution to conversion, after the second pageview, is essentially nil. The retailers who have removed the banner have, on the data, observed no regression in the promotion’s uptake and have observed a small improvement in the visitor’s perception of the broader page hierarchy.
The third is the “you might also like” widget appended to the product description. The widget is canonical; the conversion data on its specific position — at the bottom of the product description, below the fold on most devices — shows essentially no contribution. Recommendation widgets do produce uplift, as Content Recommendation Widget argues, but the position of the widget is consequential and the bottom-of-description position is not, on the data, a productive one.
The fourth is the size-guide modal that is presented as a separate document. The modal is canonical for fashion retail and is sold as a way of reducing return-rate by improving fit accuracy at the point of purchase; the data shows that the great majority of visitors do not open the modal, and that the visitors who do open it convert at approximately the same rate as those who do not. The pattern that does produce measurable improvement is the inline size guide presented within the product page itself, which visitors engage with at considerably higher rates and whose engagement correlates with measurable reduction in subsequent return rates.
The fifth is the multi-step checkout. The trade literature has, for several years, presented the multi-step checkout as a deliberate design choice with various justifications relating to mobile-form ergonomics and progressive cognitive load; the data shows, on essentially every site where I have compared the two, that the single-page checkout produces higher conversion rates against the same visitor population. The multi-step pattern is preserved on a great many sites for reasons of organisational inertia rather than of evidence, and the inertia is, in resource terms, expensive.
The patterns the literature does not discuss
Two patterns produce measurable uplift on the data and are, in my observation, rarely discussed in the canonical literature.
The first is the explicit display of estimated delivery date, computed from the visitor’s location and the current stock and dispatch state of the product. The display produces measurable uplift on conversion rate, with the magnitude depending on the specificity of the estimate and the credibility of the source. The pattern is straightforward to implement and is, on the data, more productive than the canonical “free delivery” badge that the literature treats as the headline trust signal.
The second is the visible recent-customer-activity display, where the display shows the number of customers who have recently viewed or purchased the product, on a continuously updating count. The pattern is occasionally treated, in the literature, as a deceptive practice; the data shows that, when the count is genuine — that is, when it reflects actual recent activity rather than fictitious figures — the display produces measurable uplift without any of the trust-degradation effects that the deceptive variant would produce. The retailer’s first task, in implementing the pattern, is to ensure that the count is honest; given the honesty, the pattern is one of the more productive interventions available on a typical product page.
An advisory close
The e-commerce UX checklist as it appears in the trade literature is, in 2026, a mixed inheritance: a substantial portion of the checklist consists of patterns that the data supports, and a substantial portion consists of patterns that the data does not. The retailer’s first task is to distinguish between the two; the distinction is, in the canonical literature, frequently obscured by the equal-weight presentation of patterns whose evidence is decisively unequal.
It is recommended that any retailer auditing its UX against the canonical checklist do so with the data from its own site rather than against the literature’s confident endorsements; the patterns that the data supports for the retailer’s specific audience composition are the patterns worth investing in, while the patterns the data does not support are, on the same audience composition, the patterns worth removing despite the literature’s recommendations to the contrary.