Web checkout vs buy anywhere
The simple answer is yes, most brands are still designing for web checkout, and the expectation has dramatically outpaced them. Consumers don't think in channels any more. They think in moments. The expectation is now buy where you discover, when you decide, with zero translation cost between the two.
The data on where commerce is happening is unambiguous. 73% of UK shoppers use multiple channels before purchasing, with around six touchpoints on average. Mobile apps deliver twice the conversion of mobile web, and 85% of UK shoppers prefer apps for speed and personalisation. TikTok Shop took 18.2% of total social commerce in 2025 and is on a trajectory to $30 billion by 2028. WhatsApp Flow now offers guided in-chat journeys with native payments. Connected TV hit 47.5% of total TV usage by December 2025, with 65% of viewers using a second device to research what they have seen. Apple Pay and Google Pay accounted for 28% of in-person transactions in 2023 and are projected to exceed 30% by 2027. The infrastructure for buy anywhere instantly exists. The expectation that you can use it is now table stakes.
But the brand stacks of most retailers haven't caught up. They are still architected around a website as the destination. The PIM, the ecommerce platform, the app, the POS, the social commerce extension, the loyalty programme. All separate stacks loosely stitched together, with the website at the centre. The customer experiences that as friction every time they cross a channel boundary. Cart abandoned on mobile web at 10pm, opened the app the next morning, cart is empty. Started watching a product video on TikTok, can't find the product on the website. Started a return online, can't complete it in store. Every one of those moments costs trust and basket.
What buy anywhere actually requires architecturally is unified identity, unified payment vault, channel-agnostic product feeds and headless commerce. CDP, OMS, identity layer. Stripe, Adyen and Worldpay all support this in their orchestration layers. Shopify's Shop Pay carries credentials across thousands of merchants. Klarna's persistent wallet does the same at the buyer level. The brands that have invested in this plumbing now act like one company everywhere. The brands that haven't act like 10 systems pretending to be one.
The killer point for me is that 88% of users abandon a brand entirely after one bad digital experience. The customer is not patient with the gap between expectation and execution. So the question isn't should we still optimise the website. The website still matters. The question is have we reorganised around the customer's actual journey, or are we still designing as though they only ever come through the front door?
Over-optimising websites
The framing trap in this question is treating it as web versus social. The honest answer is yes, brands are over-optimising websites, but not because websites are the wrong channel. They are optimising at the wrong layer, on the wrong assumption about where the customer is.
Yes, the imbalance is real and the data is stark. 73% of UK shoppers use multiple channels before purchasing, around six touchpoints on average. 40.4% of UK ecommerce sales now sit in omnichannel strategies. Mobile apps deliver twice the conversion of mobile web and 85% of UK shoppers prefer apps for speed and personalisation. TikTok Shop took 18.2% of total social commerce in 2025 and is on trajectory to $30 billion by 2028. WhatsApp Flow is eliminating browser redirects entirely. Connected TV hit 47.5% of total TV usage by December last year with 65% of viewers using a second device to research what they have seen. So if your team is still A/B testing the colour of a checkout button while customers are buying inside TikTok DMs, you have already answered your own question.
But here is the more useful framing. The website is no longer the destination. It is becoming a data source for everything else. Agents are scraping it. AR layers pull from it. Social commerce platforms ingest its feeds. Agentic commerce has five live production deployments today. 75% of NRF 2026 retailers said they are planning or implementing agentic. Amazon Rufus is on a $12 billion annualised run-rate as of Q4 2025. The volume is not theoretical, it is here. So website optimisation has not gone away, but the work has changed. It is structured data, product feed quality, machine-readability and headless architecture. Page-by-page CRO is yesterday's discipline.
The best brands have stopped thinking about channels and started thinking about networks. Single basket across web, app, in-store kiosk and social commerce. Saved payment methods accessible everywhere. Loyalty balances unified. Returns initiated in one channel, completed in another. The customer no longer thinks in channels, so retailers cannot either. That is not a website project, it is an architectural one. CDP, OMS, identity layer, unified payment vault. Stripe, Adyen and Worldpay all support this now. The cost is in the data plumbing, not the front end.
The killer point: 88% of users abandon a brand entirely after one bad digital experience. Loyalty is now a single-event risk, not slow erosion. So the real question is not are we over-optimising the website. It is are we still applying yesterday's optimisation playbook to fundamentally new commercial geography? Most brands are. The ones that are not, are winning.
Best brands reducing friction
The best brands are not trying to remove friction. They are choosing it more carefully. That is the shift.
Three lenses I use with clients. Stopping friction is the non-negotiable failure: crashes, payment processor errors, postcode lookups breaking, 3DS challenges with no fallback. Baymard's data shows 48% of all abandonment is unexpected costs revealed at checkout. That has been the number one cause for six years running. 26% abandon when forced to create an account. 25% abandon over security concerns. 23% abandon over slow delivery. 18% abandon when checkout is long or complicated. 13% abandon over website errors. Almost every one of those is fixable engineering work, not a brand or marketing problem.
Distracting friction is the UX discipline layer. The average ecommerce checkout has 23.48 form fields. The ideal is 12 to 14. Booking.com runs at around 8. That is the gap. The best brands are ruthless about form economy, wallet-native checkout where the customer never types a card number, transparency-first pricing where total cost is visible before checkout starts, and guest checkout that converts to an account post-purchase. They remove cookie banners, newsletter pop-ups and chat widgets firing simultaneously. They strip out the captchas on every form, the modal dialogs blocking the cart, the upsells that interrupt momentum rather than complement it.
Then there is the underused lever: purposeful friction. The Dunkin' Donuts example is the one I always land on. They removed an auto-order feature because it eliminated the high-margin upsell, the do-you-want-fries-with-that moment. Friction is sometimes how you earn the bigger basket. Reservation models for high-value purchases, like the refundable holds you see in automotive. Optional gift wrap that surfaces add-ons. Order review pages that build confidence rather than slow the journey. Post-purchase loyalty sign-up rather than blocking the purchase with a registration form.
The brands doing it best, in concrete terms. Amazon's one-click is still the gold standard for stored credentials. Apple Pay and Google Pay erase the card form entirely. Shopify's Shop Pay carries credentials across thousands of merchants. IKEA Place AR removes 94% of purchase uncertainty and delivers a 189% conversion lift. Sephora's Virtual Artist does 112%. TikTok Shop's native checkout eliminates browser redirects. Booking.com on transparency, with the total cost in your face at every step. M&S, Next and Boots using BORIS to turn returns into footfall. 85% of click and collect customers buy something extra in store.
The macro number is staggering. Baymard estimates $260 billion is recoverable in US and EU markets alone, just through better checkout design. A 35% conversion lift is available to most retailers. None of it is technology innovation. It is discipline applied at the right layer. And Baymard's benchmark of 60 leading ecommerce sites still finds an average of 39 unresolved improvement areas per site, even among Fortune 500 retailers. The basics are unsolved because brands prioritise new feature work over disciplined friction reduction.
AI agents and commerce
Commerce becomes invisible and the brand becomes a data signal. Those are the two shifts that matter most when consumers start relying on AI agents.
The market data tells you this is here, not theoretical. 75% of NRF 2026 retailer attendees said they were implementing or planning agentic commerce. Amazon Rufus is on a $12 billion incremental annualised run-rate per Q4 2025 earnings. There are five live production deployments as of April 2026: ChatGPT Instant Checkout, Amazon Buy for Me, Mastercard Agent Pay, Visa Intelligent Commerce, and Coinbase Agent.market. Stripe and Tempo launched the Machine Payments Protocol in March 2026, supporting both stablecoin and fiat in one integration. Stripe paid $1.1 billion for Bridge in October 2024, which tells you what the infrastructure layer looks like.
The invisible part is the obvious bit. Pre-authorised wallets. Agent-to-agent transactions. Subscription restocking via pre-authorised limits, which Visa's Trusted Agent Protocol now supports. Comparison across thousands of merchants in milliseconds, ranked by criteria the consumer never sees. The pay button as a concept starts dissolving. Walmart's Sparky pivot is the data point worth knowing: their direct chatbot checkout converted three times worse than agent-as-redirect, so today the agent finds and recommends, the merchant still closes. That balance will shift as confidence grows, but right now the merchant is the trust anchor.
The brand-as-data-signal shift is the bigger one. If the agent doesn't know your brand, you don't get bought. AEO, or answer engine optimisation, is now revenue-critical, not a brand exercise. Product feed quality and structured data become core infrastructure. The work is schema.org Product, Offer, Review. Clean enriched product feeds in Google Merchant Center, Amazon, Bing Shopping. Reviews surfaced with provenance. Returns policy machine-readable. The brand premium that historically came from emotion and design now has to be earned in the data, because the consumer isn't seeing the website, the agent is.
The angle I would add is that this is the same shift as headless commerce was five years ago, just at a different layer. Headless made the front end channel-agnostic. Agent-first makes the catalogue and identity agent-agnostic. The brands that built headless early are now a decade ahead. The brands that build agent-first in 2026 will be a decade ahead by 2030.
AI discovery as new SEO
Yes, but it's bigger than SEO. AEO is not the new SEO. It's the new commercial infrastructure. SEO was about being found by humans on a search results page. AEO is about being chosen by an agent inside a conversation.
The mechanics are completely different. SEO is keywords, backlinks, content, page rank, schema, technical site health. AEO is structured product data, feed quality, review provenance, machine-readable spec sheets, returns policy clarity, brand verification, agent-friendly APIs. SEO optimises for ranking on a page where humans then choose. AEO optimises for being the recommendation when an agent answers a question. There's no second page of results when an agent answers a query. You're either the answer or you're not in the conversation.
The stakes are higher because the funnel collapses. In SEO, even if you ranked third or fourth, a human might still click through, compare options, and choose you. With agentic shopping, if the agent picks one or three options and you weren't in them, you don't exist for that purchase. The Walmart and Etsy early integrations through OpenAI captured share before saturation, and the Amazon Rufus $12 billion run-rate tells you the volume is already material. Latecomers will pay an attention tax.
What this means for retailers in practice: audit your structured data implementation. Are you marked up with schema.org Product, Offer, Review and AggregateRating? Are your product feeds clean and enriched across Google Merchant Center, Amazon, Bing Shopping and the emerging agent platforms? Have you tested how your brand surfaces in ChatGPT, Perplexity, Claude and Gemini for shopping queries in your category? Are you tracking agent referral traffic as a separate channel? Have you built feed quality into your commerce roadmap as a KPI, not a one-off project?
The counter-argument worth acknowledging is that SEO didn't die when paid search came along, and it won't die now. AEO is compounding on top of SEO, not replacing it. Brands will need both: traditional SERP discovery for human-led research journeys, AEO for agent-led purchase journeys. Amazon and Walmart, the two brands winning at AEO, are also still winning at SEO. The discipline is being good at multiple discovery layers simultaneously.
The deeper shift, and the one most brands haven't internalised, is that emotionally driven shopping is being replaced by data-driven shopping. If the agent decides on price, spec and reviews, the brand premium has to be earned in the data, not the emotion. That's an uncomfortable thought for brand teams. But it's where we are.
Conversational commerce
A good conversational commerce experience does three things consistently. It removes the search step, it presents only the relevant options, and it closes the purchase without context switching. Get those three right and the experience feels less like shopping and more like having someone who knows what you need and has already paid the bill.
Let me unpack each. Removing the search step means the customer doesn't have to think in keywords. They state intent in natural language. "I need a wedding outfit for September, mid-range budget, somewhere between formal and relaxed." A good conversational interface captures that, interprets the constraints, and gets to recommendations without the customer ever typing a query into a search box. The friction of search becomes the fluency of conversation.
Presenting only the relevant options means three to five curated picks, not 200 grid results. The cognitive load shifts from filtering to choosing. And critically, the agent should explain why it picked them. Source transparency is the trust mechanism. Walmart's Sparky surfaces options inside ChatGPT with rationale, not just a list. That's the model. The agent shows working, the customer trusts the recommendation, and the brand's role becomes earning the inclusion rather than fighting for a click.
Closing the purchase without context switching is the hardest part. The customer should not be redirected to a website to complete the transaction. That's where the Walmart Sparky pivot becomes instructive. Their direct chatbot checkout converted three times worse than agent-as-redirect, which means today the agent finds and recommends, but the merchant still closes the transaction. The reason is trust: customers need a brand checkpoint at the moment of payment. Over time as trust in agents grows, this will invert and checkout will become genuinely invisible. But for now, the design challenge is to make the handoff feel seamless rather than disruptive.
The components of good conversational commerce in 2026: inline rich media (image, AR preview, short video) so the customer can verify what they're buying without leaving the conversation. Trust signals that are machine-presentable (returns policy, delivery time, brand verification, review provenance). One-tap or zero-tap payment confirmed by biometric, with the agent holding the pre-authorisation. Post-purchase context (order status, tracking, support) in the same surface. The conversation continues after the purchase rather than handing off to email.
The brands building this well right now: Amazon with Rufus, integrating discovery and purchase into one experience. ChatGPT Instant Checkout for partnered merchants. TikTok Shop's chat-to-cart with creator amplification. The pattern they share is that they treat the conversation as the entire commerce surface, not a discovery layer that hands off.
What bad conversational commerce looks like: the chatbot that asks you 12 questions before showing you a product. The recommendation that doesn't explain itself. The checkout that boots you out to a separate webpage. The support flow that loses context the moment you reach the order page. Each of those is a redirect cost, and every redirect cost compounds against conversion.
5-year delegation
Yes consumers will browse less and delegate more, but delegation is a spectrum, not a binary.
The high-value, low-frequency purchases will still be browsed. Cars, holidays, houses, jewellery, gifts. Trust requires human judgement at that stake. The 48-hour reservation model used in automotive is a sign of how those journeys evolve, not a sign they go away. The browsing becomes more structured, the consideration becomes more guided, but the human stays in the decision.
The low-value, high-frequency purchases will be almost fully delegated. Groceries, replenishment, consumables, subscriptions. The customer sets a policy: budget, preferences, ethics, brand affinity, dietary requirements. The agent executes inside that policy. Visa's Trusted Agent Protocol now supports the pre-authorised limits for exactly this pattern. Five years from now, a typical UK household will probably have agents managing four or five recurring purchase categories silently in the background.
The middle band is where it gets interesting. Considered purchases like clothing, electronics, gifts. The agent surfaces options, makes recommendations, sometimes pre-authorises. The human verifies and confirms. The Walmart Sparky pivot tells us where the friction sits today: customers want a brand checkpoint at the moment of payment, even when they trust the agent for discovery and selection. That balance shifts over time but not uniformly across categories.
My prediction: in five years, most UK consumers will delegate roughly 60 to 70% of their purchase decisions by volume, but only 20 to 30% by value. The bulk of routine commerce becomes invisible. The high-stakes commerce becomes more guided but still human-led.
The strategic shift for brands is from competing for eyeballs to competing for agent affinity. Less store loyalty, more agent loyalty. Less brand-customer relationship, more brand-platform relationship. The retailers who set up well in 2026 will be the ones still selling in 2030.
Security vs conversion
The answer is to invert the problem. Stop treating security and conversion as opposing forces. The best security UX is invisible security, where the friction sits with the system, not the customer.
Baymard's data tells you 25% of cart abandonment globally is due to security concerns, which is a tied second behind hidden costs at 48%. That's a quarter of revenue walking out because the security signal felt wrong, or the security challenge felt suspicious. Meanwhile false declines are a major hidden cost. Industry estimates put the false decline rate at 5% to 15% of all transactions, which is multiples larger than actual fraud loss. So the security versus conversion trade-off is usually framed wrong. The biggest threat to conversion is often badly-designed security, not insufficient security.
The principle that works is layered risk-based authentication. The more signals you have, the less friction you need to apply at any single layer. Device recognition, network tokens, behavioural biometrics, transaction history, payment method tokenisation. All of these run in milliseconds in the background and the consumer never sees them. The challenge only fires when the system genuinely doesn't have enough signal to authorise without it. 3DS 2.0 with intelligent SCA exemptions is the EU and UK manifestation of this. Network tokenisation through Visa Token Service and Mastercard MDES measurably lifts authorisation rates and reduces false declines simultaneously.
The brands doing this best treat security as a UX problem, not a compliance problem. Apple Pay is the gold standard. Biometric at the device level, network tokens at the payment level, no card number ever entered. Faster than typing a card and more secure than entering one. Open Banking in the UK is doing the same at a different layer: bank-level security with checkout-level UX, lower MDR, instant settlement. Stripe Radar runs ML-driven fraud detection at the network level so individual merchants don't have to.
When friction does have to be applied, framing matters. "We're protecting your card" lands very differently to "verify your identity". The customer's tolerance for security friction is proportional to whether they feel the friction is for them or against them. Optional 3DS confirmation as a trust gesture, framed positively, increases conversion. Mandatory 3DS challenge with a default-fail UI destroys it.
The deeper architectural shift is moving fraud detection from blocking to scoring. Old models were binary: pass or fail at the point of payment. Modern fraud systems score every transaction continuously, with thousands of signals, and the score determines the path. Genuine high-confidence transactions go through invisibly. Medium-confidence get light challenge (biometric, OTP). Only low-confidence get hard friction. Most brands are still running on yesterday's binary architecture and paying the conversion cost.
UX mistakes pre-payment
Most abandonment isn't at payment, it's well before. By the time the customer reaches the pay button, they've already decided. The real damage happens in the steps leading up to it, and that's where most retailers under-invest.
The Baymard data is the place to start. 48% of all abandonment is unexpected costs revealed at checkout, the single biggest cause for six consecutive years. 26% is forced account creation. 25% is security concerns. 23% is slow delivery. 22% is long or complicated checkout. 18% is website errors. 17% is returns policy concerns. 13% is no preferred payment method available. Almost every one of those failure modes happens before the customer reaches the payment screen.
The biggest pre-payment mistakes I see across clients fall into a few categories. The first is hidden total cost. Shipping, tax, fees surfaced at the last step. The fix is straightforward: show the total cost in the cart, not at the payment step. Booking.com and the airlines do this well now. Most retailers still don't.
The second is forced account creation. Blocking guest checkout is catastrophic for first-time buyers. The conversion uplift from offering guest checkout is typically 35%. The fix is offer guest checkout, then convert to an account post-purchase with a single-click sign-up using the email and address they've already given you.
The third is excessive form fields. The average ecommerce checkout has 23.48 form elements. The ideal is 12 to 14. Booking.com runs at around 8. Every unnecessary field is a percentage point of drop-off. The fix is form-field discipline. Every field has to justify its existence: address autocomplete, country-detection, smart defaults.
The fourth is distrust signals. No security badges, ugly design, hidden returns policy, no clear contact information. 25% of abandonment is security-related, and a lot of that is perception not reality. The fix is trust signals layered through the experience, not buried in the footer.
The fifth is page speed and mobile UX. Mobile cart abandonment runs at over 80% versus 66% on desktop, and the gap is widening, not closing. Apps deliver 2x the conversion of mobile web partly because of speed. Slow page loads, layout shifts, broken touch targets, all of these compound.
The sixth is the late-revealed out-of-stock. Letting a customer fill a cart and reach checkout before telling them the item is unavailable. Worst single experience in ecommerce. The fix is real-time inventory at the product detail page, not at the checkout step.
The seventh is the cognitive overload moment. Cookie banner, newsletter pop-up, exit-intent modal, chat widget invitation, all firing simultaneously at the cart page. The customer's attention fractures. The fix is sequencing: one prompt at a time, none of them at the moment of decision.
The structural mistake under all of these is treating checkout as the destination. It's not. It's the end of a journey that started on the product page, on the category page, on the homepage, or off-site entirely. The conversion problem is upstream of the pay button, and that's where the investment needs to sit.
3-year underestimated expectation
The expectation that will catch most businesses out over the next three years isn't AI, isn't personalisation, isn't even instant support. It's cross-channel continuity. The simple expectation that what I started in one channel I can finish in another with no translation cost.
The customer already thinks across channels. 73% of UK shoppers use multiple channels before purchasing, with six touchpoints on average. But most brands still treat each channel as a separate stack. The customer abandons a cart on mobile web at 10pm. They open the app in the morning. The cart is empty. They've watched a TikTok video about a product, can't find it on the website. They started a return online, can't complete it in store. Every one of those moments is silent abandonment, and the brand never sees it in their analytics because nothing technically failed. The customer just gave up.
The expectation hardens faster than businesses realise. Five years ago, this was a nice-to-have. Today it's table stakes. In three years it'll be a brand-damaging absence if it isn't there. 88% of users abandon a brand entirely after a single bad digital experience, and a broken cross-channel handoff is one of the most common single-event experiences that triggers that abandonment.
What it requires is mostly architectural, not UX. CDP for unified identity. OMS for orchestrating fulfilment across channels. Single payment vault across web, app, social and in-store. Stripe, Adyen and Worldpay all support this in their orchestration layers. The investment isn't in the front end, it's in the data plumbing. And that's why most brands underestimate it: it's expensive, it's invisible, it's not glamorous, and it doesn't have a single visible KPI tied to it.
The brands getting this right today: John Lewis with cross-channel inventory and identity. Argos through Sainsbury's integration. Sephora carrying loyalty and saved items across web, app and store. These aren't the most innovative retailers in their categories. They're the ones who quietly invested in the architecture five years ago and are now invisibly winning.
Will checkout disappear?
No, but it'll become invisible. There's an important difference.
Checkout as a separate UI moment will disappear for most transactions. The mechanic underneath, a price, an authorisation, a settlement, won't go away. It'll just be embedded in everything: ambient, agent-driven, pre-authorised, biometric. We're already 70% of the way there. Apple Pay, Amazon 1-click, TikTok Shop chat-to-cart, Spotify auto-renewal, Uber pulling payment without ever showing you a screen. Each of these is checkout that has become invisible without disappearing.
What will remain visible for the foreseeable future is the high-trust moment. Cars, mortgages, holidays, expensive gifts, things where the customer needs to feel the weight of the decision. The 48-hour reservation in automotive is a sign of this evolution: the moment of commitment becomes more structured, more guided, but still a moment.
What will also remain visible is the dispute and refund path. When something goes wrong, customers need a transparent process. Invisible commerce works only as long as the trust holds. Break that trust and you need a visible repair mechanism. That's the unsolved part of agentic commerce right now: when the agent buys the wrong thing, who handles the return, on whose authority?
The architectural shift behind invisible checkout is significant. Pre-authorisation of agents. Biometric anchoring at the device. Tokenised payments at the network level. Programmable money via stablecoin and eventually the UK Digital Pound. Each of these removes a step from what the customer experiences as commerce. The cumulative effect over the next five years is that for the bulk of routine purchases, customers will stop noticing they're transacting at all.
For brands, this is both an opportunity and a threat. The opportunity is that customer friction collapses, repeat purchase rises, lifetime value climbs. The threat is that brand loyalty doesn't survive invisible commerce easily. If the agent makes the choice and the payment happens silently, what's left of the brand-customer relationship? That's the question retailers will spend the next five years answering.