Most trend lists describe the present tense and call it the future. This one starts from what the data already shows in 2026, states where each line goes next, and rates how confident that call is. Seven charts, 56 unique sources, and a section translating all of it for small businesses — including the three trends you are allowed to ignore.
1. The market is growing steadily. Almost nothing else is stable.
Start with the number everyone quotes. EMARKETER puts worldwide retail ecommerce at roughly $6.88 trillion in 2026 — about 21.1% of all retail spending — rising to approximately $7.38 trillion in 2027 and $7.89 trillion in 2028, when online reaches about 22.5% of total retail.
That is a boring, dependable line: single-digit growth, gently decelerating. It is also the least interesting fact in this article, because it tells you nothing about where the money moves inside the channel. The composition of that $7.38 trillion in 2027 will look meaningfully different from 2026’s — different discovery surfaces, different payment rails, different landed-cost math, different measurement. Every trend below is a claim about composition, not size.
2. Agentic commerce stops being a pilot and becomes plumbing
Call: By end of 2027, protocol support is a procurement checkbox, not a differentiator — while fully autonomous checkout stays a minority of agent-influenced revenue.
Watch: Share of orders where an agent touched discovery, not just checkout.
Confidence: High on infrastructure. Low on autonomous purchase volume.
2026 was the year the rails got laid. Stripe and OpenAI’s Agentic Commerce Protocol shipped, Google launched the Universal Commerce Protocol at NRF in January 2026, Microsoft Copilot Checkout went live in the US, and Anthropic’s Model Context Protocol became the de facto way agents reach live inventory and pricing. Visa published a Trusted Agent Protocol; Stripe introduced shared payment tokens so an agent can pay without holding a card.
What did not happen is equally instructive. OpenAI moved away from its initial Instant Checkout approach inside ChatGPT, shifting toward product discovery and comparison while allowing merchants to use their own checkout experiences. Walmart reported in-chat purchases converting roughly three times worse than redirecting the shopper to its own site. Forrester’s mid-2026 read is blunt: the narrative assumes agent-led buying is already widespread, while actual consumer trust and adoption remain limited and unevenly distributed.
Agents are winning the top of the funnel long before they win the wallet. Planning for 2027 on the assumption that agents will buy is the wrong bet; planning on the assumption that they will shortlist is the right one.
The practical consequence is that the competitive battleground in 2027 is not autonomous checkout. It is whether your catalog is legible to a machine — structured, complete, real-time, and API-exposed. Product data has quietly become a distribution asset.
“Structured, accurate, real-time product data is now a trust signal.”
— commercetools, The agentic commerce radar: Spring 2026 update, 22 April 2026
The counterargument
Agentic commerce is currently a rounding error. In July 2026, EMARKETER lowered its forecast for US retail ecommerce sales originating on AI platforms to approximately $19.74 billion in 2026, about 4% below its previous forecast. The revision reinforces the need to keep the opportunity in perspective. If you are a $3M store, protocol work in 2027 is an option contract on a channel, not a revenue plan. Price it accordingly.
What to do in 2027
- Audit machine readability before you audit design. Every PDP should expose price, availability, variants, shipping terms, and return terms as structured data, not just as rendered text.
- Stop blocking the crawlers you want. Check robots.txt and your WAF for blanket AI-agent blocks put in place in 2024–25.
- Pick two protocols, not five. Support the ones your platform natively brokers rather than hand-rolling integrations.
- Instrument agent traffic separately in analytics now, so you have a full year of baseline before you have to justify budget against it.
3. AI referral becomes the highest-intent channel you own least
Call: Through 2027, AI-referred sessions stay small in volume and disproportionate in value — and the gap between machine-readable sites and everyone else widens into a real revenue difference.
Watch: Revenue per visit from AI referrals vs. organic search, tracked monthly.
Confidence: High.
The single most striking data point of 2026 came from Adobe Analytics, drawn from more than a trillion visits to US retail sites. In March 2025, visitors arriving from AI assistants converted roughly 38% worse than everyone else. Twelve months later, in March 2026, the same channel converted 42% better. By May 2026 Adobe measured the premium at 54%. Same channel, same stores, opposite sign.
An ~80-point swing in twelve months. Adobe attributes it to pre-qualification: the comparison, filtering, and objection-handling now happen inside the assistant, so the click that reaches you is late-stage. AI-referred visitors also spent 48% longer on site, viewed 13% more pages and had a 12% higher engagement rate.
“AI is quickly becoming the primary interface between consumers and their favorite brands.”
— Vivek Pandya, Director, Adobe Digital Insights
Adobe also published something less flattering: an AI visibility scorecard for retail pages. Returns, contact and FAQ pages scored at or above 80% (82%, 81% and 80% respectively) — they are text-heavy and easy to parse. Homepages averaged 75%, category pages 74%. Product detail pages, the most numerous and most commercially important pages on any store, averaged 66%. The pages you most need an assistant to read correctly are the ones it reads worst.
Average AI visibility score for retail product detail pages, March 2026 — the worst-performing page type Adobe measured. Homepages: 75%. FAQ pages: 80%. Returns pages: 82%.
Keep it in proportion
Contentsquare, measuring 99 billion sessions, put AI-referred traffic at roughly 0.2% of total traffic in Q4 2026 — growing more than 600% year over year, but from almost nothing, and converting at 1.3% absolute versus email’s 1.9%. Adobe reports a relative premium; Contentsquare reports an absolute rate. Both are true. Neither justifies moving your budget wholesale in 2027.
What to do in 2027
- Treat the PDP as an answer, not a listing. Sizing, materials, compatibility, warranty and return terms in plain text, not in a tab loaded by JavaScript.
- Write the comparison content the assistant is looking for — “X vs Y”, “best X for [constraint]”, “does X work with Y” — with substantive, self-contained answers.
- Check server-side rendering. Content that only appears after hydration is content an assistant may never see.
- Segment AI referrers in GA4 as a named channel group and report on revenue per visit, not sessions.
Is your catalog readable by the systems that now recommend it?
4. Retail media stops being free money
Call: 2027 is the year retail media growth ex-Amazon slows to single digits and mid-tier networks are forced to differentiate or consolidate.
Watch: Your incremental ROAS on retail media, measured against a holdout — not platform-reported ROAS.
Confidence: High.
WARC Media forecasts global retail media investment passing $200.4 billion in 2026 and reaching $223.4 billion in 2027 — 15.2% of all worldwide advertising investment. Impressive, until you read the second line: growth of 11.5% in 2027 falls to 9.8% excluding Amazon, the lowest annual rate since WARC began tracking the sector. In 2025, Amazon alone accounted for roughly three-quarters of all US retail media spend, per EMARKETER.
The category is maturing and concentrating at the same time. For everyone outside the top two networks, 2027 is about vertical specialization and honest attribution rather than inventory scale.
“While retail media excels at converting existing demand, it underperforms on long-term brand building.”
— Alex Brownsell, Head of Content, WARC Media
There is a second, sharper risk that most 2027 planning ignores. Retail media monetises human eyeballs scrolling human-readable pages. WARC’s James McDonald, Director of Data, Intelligence & Forecasting, flagged in the firm’s November 2025 forecast that agentic AI “threatens the high human traffic volumes” that have funded these networks in the first place. If an assistant shortlists three products and the shopper never browses a category page, the sponsored placement on that page was never seen. Retail media and agentic commerce are, structurally, in tension.
What to do in 2027
- Run GEO or audience holdouts at least twice a year. Closed-loop attribution inside a network is not incrementality.
- Shift a defined share off-site. Off-site and CTV formats reach the shopper before an agent narrows the field.
- Renegotiate on ad load. As networks push more units per page, conversion quality degrades. Make placement quality contractual.
- For SMB retailers: the ROI case for launching your own retail media network in 2027 is thinner than the vendor decks suggest. Check whether you have the traffic to sell before you buy the tech to sell it.
5. Checkout goes wallet-first and agent-readable
Call: Digital wallets pass 60% of global ecommerce transaction value in 2027, while payment-method coverage remains a measurable source of avoidable cart abandonment for SMB merchants.
Watch: Checkout completion rate segmented by payment method offered.
Confidence: High on wallet dominance, Medium on clearing 60% specifically by 2027 — the supporting projection predates the 2025 actuals. Medium on agent-native payment volume.
Worldpay’s 2026 Global Payments Report, built on a survey of more than 63,000 consumers across 42 markets, found digital wallets already carrying 56% of global ecommerce transaction value and 33% of in-store spending. Worldpay’s earlier modeling projected wallets reaching 61% of global ecommerce value by 2027 and more than $25 trillion in combined online and point-of-sale transaction value by 2027. Treat that 61% with care: it was published before the 2025 actuals landed, and the current GPR series forecasts to 2030 rather than 2027. Getting from 56% to 61% in two years is a steeper climb than the latest data implies — which is why the call below is rated Medium on timing even though the direction is not in doubt.
The US is a laggard, not a leader, in wallet adoption. If you sell cross-border, the payment mix that works in Chicago will quietly cost you conversions in São Paulo, Amsterdam or Jakarta.
Two structural shifts sit underneath this. First, account-to-account rails — Pix in Brazil, UPI in India — have made instant bank transfer a first-class ecommerce method in entire markets. Second, agents need a way to pay that is neither a stored card nor a human tapping Face ID. Stripe’s shared payment tokens and Visa’s Trusted Agent Protocol are early answers. Stablecoins remain, per Worldpay’s own framing, a developing story whose clearest near-term use case is cross-border settlement rather than consumer checkout.
What to do in 2027
- localize the wallet stack per market, not per platform default. Offering Apple Pay in a Pix market is a conversion tax.
- Measure approval rates, not just fees. A processor that is 40 basis points cheaper and declines 2% more transactions is more expensive.
- Treat BNPL as a category decision. It earns its keep on higher-AOV considered purchases and costs you margin on impulse items.
- Ask your payment provider directly what its roadmap is for delegated or agent-initiated payments. If they cannot answer in 2027, that is data.
6. Landed cost becomes a conversion lever, not a back-office line
Call: Not a forecast — a statutory date. The US de minimis exemption is terminated by law on 1 July 2027. Duty-inclusive pricing at checkout becomes standard practice, and the low-value direct-ship model does not come back.
Watch: International cart abandonment rate at the shipping step.
Confidence: High. This one is written into legislation.
Plain English — what is de minimis?
What it was: For most of the last century, customs authorities waived duty on parcels too cheap to be worth taxing. The US drew that line at $800 per person per day. A shopper in Illinois ordering a $40 phone case direct from a supplier in Shenzhen paid $40 — no duty, almost no paperwork, and the parcel cleared customs on the shipping manifest alone. De minimis is legal shorthand for “too small to bother with”, and the US rule dates to 1938, when chasing a few cents of duty genuinely cost the government more than the few cents were worth.
Who it affects: Anyone whose goods cross a border. That means brands shipping direct from overseas manufacturers, dropshippers, marketplace sellers, and any business importing inventory in lots of small consignments — plus the shoppers buying from them. If you stock domestically and sell domestically you are not directly affected, though your suppliers almost certainly are and their prices will reflect it. Genuine gifts under $100 and goods travellers carry in personally are untouched.
Why it matters: Two things changed, not one. Duty is now owed on that $40 phone case — but the bigger operational hit is that every parcel needs a real customs entry, filed by a qualified party, with real data behind it. Shipping tens of thousands of cheap parcels a month straight from an overseas supplier was a viable business model largely because that paperwork did not exist. It does now. And in the EU and UK, the party legally on the hook for getting it right has shifted from the carrier to the merchant.
The one-line version
The cheapest route from an overseas factory to a customer’s doorstep has been closed — in the US, permanently, from 1 July 2027.
This is the only item on the list with a date already fixed in statute. Section 70531(b) of the One Big Beautiful Bill Act, enacted 4 July 2025, terminates the US de minimis exemption effective 1 July 2027. Everything before that date has been executive action; from that date it is law, and it does not lapse when an administration changes.
The sequence is worth having straight, because it has been widely misreported. Executive Order 14256 suspended duty-free de minimis for imports from China from 2 May 2025. Executive Order 14324 extended the suspension to all countries from 29 August 2025. On 20 February 2026 the Supreme Court decided Learning Resources, Inc. v. Trump, holding that IEEPA does not authorise the President to impose additional tariffs — but the ruling did not address the de minimis suspension. Executive Order 14389 duly ended the IEEPA duties while Executive Order 14388, signed the same day, continued the de minimis suspension. CBP then codified an indefinite suspension in its own regulations by two companion interim final rules on 24 June 2026 — one covering all modes other than the international postal network (91 FR 37789), one covering postal — under its independent statutory authority, explicitly so that the suspension survives regardless of what happens to the executive orders.
De minimis shipments processed by US Customs and Border Protection in FY2024 — nearly ten times the 139 million processed in 2015. CBP’s stated reason for ending the exemption is that the volume made the original cost-benefit logic obsolete. CBP, 91 FR 37789.
Europe is moving on the same trajectory. The EU abolished its €150 exemption effective 1 July 2026, replacing it with a flat €3 customs duty per item category, with standard tariffs expected once the Customs Data Hub goes live around 2028. The change comes as low-value ecommerce imports continue to surge. According to the European Commission, nearly 5.9 billion low-value items entered the EU in 2025, up 26% from 2024. The UK plans to remove customs-duty relief for low-value imports of £135 or less by October 2028 at the latest.
What to do in 2027
- Show duties and taxes in the cart, before the shipping step. Landed-cost transparency is now a conversion mechanic, not a compliance nicety.
- Clean your HS classifications. Wrong codes are now expensive on every order, not just on bulk imports.
- Model forward stocking. For steady-volume SKUs, in-country inventory often beats per-parcel duty once you price the new fees honestly.
- Re-underwrite your product mix. Sub-$25 imported items may simply no longer work as standalone cross-border SKUs. Bundle or drop them.
7. Returns get engineered instead of policed
Call: Return rates bifurcate through 2027: merchants investing in fit technology, product data and returnless thresholds push the low end down, while policy-only merchants stay flat and lose margin to rising reverse-logistics costs.
Watch: Cost per return and recovery rate — not return rate.
Confidence: Medium. Direction is clear; the size of the gap is not.
NRF and Happy Returns projected US retail returns at roughly $849.9 billion in 2025 on a 15.8% return rate, with ecommerce projected higher at about 19.3%. In 2024, returns totaled approximately $890 billion on a 16.9% return rate. Those are retailer estimates for the year rather than audited actuals, which matters when you are benchmarking against them. For context, NRF’s projected online return rate of 19.3% provides a more defensible blended ecommerce benchmark than an unsupported DTC estimate.
The blended average is not useful in isolation. Coresight Research found a 23.4% average return rate for US online apparel, while ICSC research found a 5% return rate for in-store purchases. Electronics and beauty benchmarks vary by source. Benchmark against your own category and channel mix, not against a single universal number.
Cost is moving the wrong way, though this is the part of the returns picture with the least published data behind it. There is no authoritative public benchmark for a flat dollar cost per parcel return. Deloitte reports that returns can cost merchants approximately 15–30% of the returned item’s value, depending on the product and process. We are flagging that rather than dressing it up. The practical implication is unchanged: nobody can tell you your cost per return, so you have to calculate it yourself.
Three responses are consolidating into standard practice. Returnless refund thresholds are rising and are increasingly set per SKU tier rather than as one flat rule. AI fit and size prediction is moving from pilot to default in apparel, though we have not found a credible published adoption figure and would treat any specific percentage you are quoted with suspicion. And return fraud — 9% of returns, per NRF, which is the one hard number in this paragraph — is being met with machine-learning policy engines rather than blanket tightening, because blanket tightening costs you good customers along with the bad ones.
What to do in 2027
- Calculate your true cost per return including support labour and unsellable write-offs, then set a returnless threshold per SKU tier.
- Fix the top three return reasons at the PDP before touching policy. Fit, expectation mismatch and damage are product-content problems.
- Move returns to self-service. Every return that starts as an email costs an agent’s time on work a portal should do.
- Report recovery rate — resale value recovered per dollar returned — alongside return rate in your monthly numbers.
8. B2B quietly becomes the bigger ecommerce prize
Call: US B2B ecommerce reaches roughly $3 trillion in 2027 — about a quarter of all US B2B sales — and self-service portals move from competitive advantage to table stakes for distributors and manufacturers.
Watch: Share of reorders placed without a sales rep touching them.
Confidence: High on the trajectory. Medium on the exact figure.
Forrester’s forecast has US B2B ecommerce growing from $1.7 trillion in 2021 to approximately $3 trillion by 2027 at a 10.7% compound rate — reaching about 24% of total US B2B sales, up from 16% in 2021. EMARKETER measured B2B ecommerce site sales at $2.297 trillion in 2024, up 10.5% year over year.
B2B is not becoming B2C, but it is adopting B2C’s mechanics. McKinsey’s B2B Pulse research finds buyers now move through roughly ten interaction channels in a single journey, up from about five a decade ago.
The behavioral shift is the real story, though it is usually overstated. Gartner’s finding is that 67% of B2B buyers prefer a rep-free experience for at least part of their purchase — not that two-thirds want sellers out of the process altogether. Read correctly it is a claim about sequencing: most buyers want to complete some stages without a rep, so a mandatory sales touch placed at the wrong stage is friction rather than reassurance. The same research house projects that by 2030, 75% of B2B buyers will prefer human interaction for complex or high-stakes transactions. The 2027 sales motion is bimodal: frictionless and digital for research and reorders, expert and human for the decisions that swing a deal.
Paul Okhrem, Co-Founder and CEO of Elogic Commerce, makes the point that for B2B in 2026–27, operational readiness now matters more than front-end features — clean, API-exposed pricing and catalog data is the prerequisite for everything from self-service quoting to agentic reordering inside existing contracts.
What to do in 2027
- Ship the reorder flow first. Repeat purchasing is the highest-volume, lowest-risk portal use case and it frees rep time immediately.
- Expose contract pricing to the portal. If a logged-in buyer cannot see their price, the portal is a catalog, not a channel.
- Publish the specs. B2B buyers research with generative AI now; Content locked behind forms or authentication is largely inaccessible to the crawlers that power AI search and discovery.
- Keep the human path obvious for high-value and configured orders. Rep-free is a preference, not a rule.
9. Social commerce graduates from brand spend to a revenue line
Call: US social commerce continues expanding in 2027, driven more by spend depth per buyer than buyer growth, with video remaining one of the strongest conversion formats.
Watch: Annual spend per social buyer in your own cohort data.
Confidence: Medium. Forecasts in this category vary by an order of magnitude between vendors.
A note on the numbers in this section
This is the thinnest-sourced trend on the list, and we would rather say so than pretend otherwise. The figures below circulate widely and trace back to EMARKETER social commerce forecasts, but we have only been able to confirm them through secondary compilations rather than the primary releases. Treat them as directional. Every other trend in this article is sourced to a document we have read.
US social commerce continues to mature into a meaningful ecommerce channel. EMARKETER estimates US social commerce sales reached $87.02 billion in 2025, up 21.5% year over year, and forecasts another 18% increase in 2026, pushing the market above $100 billion for the first time. TikTok Shop is becoming an increasingly important part of that market, generating an estimated $15.82 billion in US sales in 2025 and projected to surpass $20 billion in 2026.
But the number that matters more than headline market size may be depth. Spend per social buyer is growing faster than the number of buyers, which creates a different operating challenge. Growth in basket depth and repeat purchase rewards retention work; growth in buyer count rewards acquisition work. For ecommerce teams planning for 2027, social commerce increasingly needs to be treated as a measurable sales channel rather than simply an extension of brand or influencer marketing.
Treat the market-size numbers with care
Estimates of the global social-commerce market vary dramatically, from hundreds of billions to several trillion dollars, largely because researchers define social commerce differently. When a vendor cites a headline number at you, ask whether it counts total marketplace GMV, in-app checkout only, or social-influenced purchases. Those are three different businesses.
What to do in 2027
- Measure the second purchase. Social acquisition is only economic if the cohort repeats; track LTV by acquisition surface, not blended.
- Build for vertical video first in product photography and creative briefs, then adapt to other placements.
- Sync catalog and inventory properly. Out-of-stock items in a social storefront burn trust faster than they do on your site.
- Test creator partnerships as a channel, not a campaign — with negotiated whitelisting rights so you can put paid behind what works.
10. Composable by surface replaces the big-bang replatform
Call:2027 is a correction year. Incremental, surface-by-surface decoupling becomes the default recommendation, and full-stack composable is reserved for genuinely complex operations with the engineering teams to match.
Watch: Time from decision to live for a new customer-facing surface.
Confidence: Medium. This is a judgement call about industry practice, not a measured trend.
Plain English — Composable, headless, MACH?
What the words mean: Most ecommerce platforms are monolithic: one system handles the catalog, search, cart, checkout, content and admin together. Think of an all-in-one stereo — it works out of the box and you cannot swap the speakers.
Headless splits that in two. The shopfront customers see is detached from the engine that processes orders, and the two talk over APIs. You can rebuild the shopfront without touching the engine.
Composable goes further and breaks the whole stack into separate specialist services — search from one vendor, checkout from another, product information from a third — assembled like hi-fi separates. MACH is just the industry’s acronym for the principles behind it: Microservices, API-first, Cloud-native, Headless.
Who it affects: Anyone choosing, renewing or outgrowing an ecommerce platform. In practice it becomes a live question when your platform stops being able to do something you need, or cannot reach a place you need to sell. If your store works, your team is small, and you have no developers on staff or on retainer, this is a decision you can reasonably defer — despite what the vendor decks imply.
Why it matters: Composable buys flexibility and bills you in engineering time. Every piece you unbundle is a component you now own the wiring for, plus a vendor relationship, a contract and an upgrade cycle. That trade is worth making when your requirements genuinely exceed what one platform can deliver, and expensive when they do not. The reason it is back on the agenda for 2027 is not fashion — it is that you increasingly need to publish the same catalog to a website, an app, social storefronts and AI agent endpoints at once.
The one-line version
Composable means assembling your store from separate specialist parts instead of buying one system that does everything. Ask what it lets you do that you cannot do today, and price the wiring honestly.
Newer data paints a more useful picture of how composable commerce is actually being adopted. In a Shopify and IDC survey of more than 1,000 enterprise organizations, 45% reported using a composable front end with a full-stack back end, while 27% reported using a fully headless or modular architecture. Rather than showing a uniform march toward fully composable stacks, the data points to a more mixed architecture landscape where businesses are selectively decoupling the parts of the experience that benefit most from flexibility.
That hybrid pattern also reflects a broader correction in the composable conversation. Practitioners who delivered the 2022–24 composable wave now report that organizations going fully composable without the right team structure found integration complexity multiplying engineering backlogs rather than shrinking them. The broader evidence supports a more cautious conclusion: fully composable architectures can increase integration complexity and may not be appropriate for organizations that lack sufficient business complexity or technical maturity.
What is genuinely new is the reason to decouple. Through 2027 you will need to publish the same catalog to a website, a mobile app, one or more social storefronts, and at least one agent-facing endpoint. That is an argument for an API-addressable commerce core and swappable presentation layers. It is not an argument for replacing your checkout, search, PIM and CMS simultaneously.
What to do in 2027
- Refuse the big-bang replatform. Migrate surfaces, not the engine. Keep every phase independently reversible.
- Baseline before you build. Capture field Core Web Vitals and conversion per template, so post-launch claims are provable.
- Audit your apps and extensions first. Classify each as backend-only, frontend, or dead weight. Uninstall the third category regardless of what you decide next.
- Treat vendor conversion-lift figures as marketing. Headless does not raise conversion; the performance and merchandising work you do during a headless build does.
11. Measurement rebuilds around incrementality, not clicks
Call: By late 2027, last-click is a diagnostic rather than a decision tool for any merchant above roughly $5M, replaced by a stack of server-side conversion data, media mix modeling, and periodic holdout tests.
Watch: Gap between platform-reported ROAS and modelled contribution.
Confidence: Contested. The direction is agreed; the destination is not.
Three pressures are converging. Platform signal keeps narrowing — Meta deprecated its 7-day-view and 28-day-view attribution windows in January 2026, with Meta’s standard attribution defaulting to 7-day click-through, 1-day engage-through and 1-day view-through for applicable conversion campaigns, and GA4 has defaulted to data-driven attribution since retiring first-click, linear, time-decay and position-based as primary models in November 2023. Regulation keeps expanding — by 2026, twenty US states had comprehensive consumer privacy laws in effect, with Indiana, Kentucky and Rhode Island joining on 1 January.
The third pressure is the one nobody has solved. In agent-mediated commerce, the entire discovery and consideration phase happens inside the assistant. The behavioral data stream you receive starts at the add-to-cart moment. You cannot see the comparison, the refinement, or the objection that nearly lost you the sale — and neither can your personalization engine or your retail media targeting.
Blind above the cart
The measurement problem of 2027 is not attribution modeling. It is that a growing share of the decision now happens on infrastructure you do not own, do not instrument, and cannot query.
The workable 2027 stack is unglamorous: Meta Conversions API and Google Enhanced Conversions to improve conversion measurement by supplementing browser-based tracking with server-side or first-party data signals; zero-party data collected through quizzes, preference centres and post-purchase surveys, which also captures word-of-mouth and offline touchpoints digital attribution misses entirely; media mix modeling on aggregated data for budget allocation; and incrementality tests to settle arguments that models cannot.
What to do in 2027
- Implement server-side conversion tracking first. It is the highest-leverage single fix — it recovers conversions you are already generating but cannot currently see.
- Add a post-purchase “how did you hear about us” survey. Low effort, and it can help uncover AI-assisted discovery that traditional referral and click-based attribution misses.
- Run one holdout test per quarter on your largest channel. Model disagreement is normal; unfalsifiable ROAS is not.
- Report directional confidence, not false precision. Perfect attribution is not available in 2027 and pretending otherwise makes worse decisions than admitting it.
12. What this actually means if you run a small business
Every number in this article comes from enterprise-scale data. Adobe measured a trillion visits. WARC tracks global ad investment. Forrester models a $3 trillion market. Those figures describe the market you sell into — they are not a budget, a headcount, or a roadmap you can lift.
So here is the translation. If you do somewhere between $250,000 and $5 million online, with no dedicated ecommerce team and a developer who is either part-time or on retainer, the question is not which of these trends are real. They all are. The question is which ones will change anything you actually do in 2027 — and, just as usefully, which ones you are allowed to ignore.
The honest triage
| Trend | For an SMB | Effort |
|---|---|---|
| Structured product data for agents | ACT | Days |
| Product pages readable by AI | ACT | 2–4 weeks |
| Retail media | SKIP | – |
| Wallet coverage at checkout | ACT | Days |
| Landed cost & duties | IF YOU IMPORT | 2–3 weeks |
| Returns economics | IF >15% | 3–6 weeks |
| B2B self-service | IF YOU SELL TRADE | A quarter |
| Social commerce | WATCH | Test budget |
| Composable architecture | SKIP | – |
| Server-side tracking | ACT | Days |
Four items are unconditional, three are conditional on your business model, and three you can leave alone. That is a very different article from the one above, and it is the one most small businesses need.
What you are allowed to ignore
Retail media. Unless you are already spending meaningfully on Amazon Ads or Walmart Connect, nothing in trend 03 requires action from you. The conversation about launching your own retail media network is for retailers with traffic to sell. If you are buying traffic rather than selling it, this is somebody else’s problem.
Replatforming. Composable architecture is the most oversold item on the list for businesses your size. If Shopify or WooCommerce currently does what you need, the correct 2027 decision is almost always to stay and spend the money on product data and page speed instead. A replatform is not always the highest-return path. When the underlying problem is conversion friction rather than platform integrations, targeted optimizations may deliver value faster and at a fraction of the cost.
Autonomous checkout. You do not need to integrate a commerce protocol in 2027. What you need is the structured product data those protocols read — and that same data improves your Google Shopping feed and your organic search at the same time. Do the underlying work; skip the integration.
Three worked examples
Two-person apparel brand
Shopify · ~$800k/yr · US plus some international · 28% return rate · agency on retainer
- 1. Returns first. At 28% on apparel, this is the largest recoverable number in the business. Per-SKU size guides with real garment measurements, fit feedback captured at return, and a returnless threshold for anything under $25.
- 2. Duties at the cart. International orders need duty shown before the shipping step, via a landed-cost provider such as Zonos or Passport.
- 3. Meta Conversions API. Server-side events, plus a two-question post-purchase survey in Klaviyo asking how they found you.
Local specialty retailer
- 1. Product data and rendering. Complete specs, dimensions and compatibility in plain HTML, with valid Product schema. Check nothing important loads only after JavaScript.
- 2. Wallets. Apple Pay, Google Pay and PayPal enabled and tested on a real phone. This is a low-effort settings change that can remove avoidable checkout friction.
- 3. Ignore almost everything else. No replatform, no retail media, no protocol work. Cross-border does not apply. Spend the remaining time on page speed.
Regional B2B distributor
- 1. Reorder flow. Logged-in customers seeing their contract pricing and repeating a previous order without calling a rep. Highest-volume, lowest-risk portal feature there is.
- 2. Get the specs out of the PDFs. Your buyers research with AI assistants now, and content locked behind a contact form is largely inaccessible to the crawlers that power AI search and discovery.
- 3. Keep the human path obvious. Rep-free is a preference for routine orders, not for configured or high-value ones.
A realistic year
None of the “act” items above require a platform change, a new vendor category, or a hire. Sequenced across four quarters, a small business year looks roughly like this: fix product data and structured markup; fix checkout payment coverage and, if relevant, landed cost; fix measurement with server-side events and a post-purchase survey; then spend the last quarter on whichever of returns or B2B self-service your model actually exposes you to.
The exact effort will vary by stack and team, but the work can be sequenced across the year in manageable phases, most of it focused on foundations rather than anything labelled as a 2027 trend. The awkward truth of a forecast article is that the correct response to most of it, for most small businesses, is to do unglamorous things properly.
13. One thing compounds. The rest are maintenance.
Planning your 2027 ecommerce roadmap?
Anala builds and modernizes ecommerce for small and mid-sized businesses — catalogue and structured data, checkout and payments, headless surfaces, and the measurement layer underneath. Bring us your current stack and constraints; we will tell you which of these ten actually apply to you and which you can safely ignore.
14. Frequently asked questions
How big will global ecommerce be in 2027?
EMARKETER forecasts worldwide retail ecommerce sales of approximately $7.38 trillion in 2027, rising to about $7.89 trillion in 2028 when online reaches roughly 22.5% of total retail spending. For context, 2026 is forecast at $6.88 trillion, or 21.1% of all retail. Growth is steady but decelerating — roughly 7% annually, down from the double-digit rates of the early 2020s. Note that other forecasters using different category definitions publish materially higher figures; Oberlo’s series, for example, puts 2027 at $7.57 trillion. Use one source consistently rather than mixing series.
What is agentic commerce, and does it matter for a small ecommerce business?
Agentic commerce is when an autonomous AI agent handles part or all of a purchase, interpreting what the buyer wants, comparing merchants, and in some cases completing checkout, rather than a human browsing a website. In 2026, it is still small: EMARKETER’s July 2026 forecast puts US retail ecommerce sales originating on AI platforms at approximately $19.74 billion, about 4% below its previous forecast. For a small business in 2027, the practical implication is not building agent integrations. It is making sure your product data is structured, complete and machine-readable so agents can accurately represent and recommend you. That work also improves conventional SEO and can help reduce returns, so it pays for itself regardless of how quickly agentic commerce grows.
Will AI search reduce my ecommerce traffic?
It will change its composition more than its volume, at least through 2027. Adobe Analytics, measuring more than a trillion visits to US retail sites, found AI-referred traffic growing 393% year over year in Q1 2026 while converting 42% better than non-AI traffic in March 2026 — a reversal from converting roughly 38% worse a year earlier. The visitors arrive pre-qualified because comparison happened inside the assistant. But absolute volume remains small: Contentsquare reported AI-referred traffic at around 0.2% of total traffic in Q4 2026. Plan for fewer, better-qualified sessions from AI, and continue to defend conventional organic search, which still carries the majority of discovery.
What is the de minimis exemption?
What does the end of de minimis mean for a small online store?
If you import goods or ship internationally, every parcel now carries duty — and the change is permanent. The US suspended its $800 exemption for imports from China in May 2025 and for all countries from 29 August 2025, and Section 70531(b) of the One Big Beautiful Bill Act terminates the exemption outright by statute on 1 July 2027. US Customs and Border Protection codified an indefinite suspension in its own regulations in June 2026, independent of the executive orders. The EU abolished its €150 exemption on 1 July 2026, replacing it with a flat €3 customs duty per item category, and the UK plans to remove customs-duty relief for low-value imports of £135 or less by October 2028 at the latest. Practical consequences: landed cost per unit rose, low-value imported SKUs may no longer be viable standalone, and under EU and UK frameworks the merchant is now the legal customs debtor. Show duties in the cart rather than surprising customers at delivery.
Is retail media still worth the spend in 2027?
For most brands, yes — but with tighter measurement and lower expectations of growth. WARC Media forecasts global retail media investment reaching $223.4 billion in 2027, growing 11.5% overall but only 9.8% excluding Amazon, the slowest rate since WARC began tracking the sector. Amazon accounted for roughly three-quarters of US retail media spend in 2025, so “retail media performance” for most advertisers really means Amazon performance plus a long tail. Retail media converts existing demand efficiently and underperforms at brand building. Run holdout tests at least twice yearly rather than relying on network-reported closed-loop attribution, and watch whether rising ad load is degrading the quality of the placements you buy.
What is a realistic ecommerce return rate benchmark for 2027?
Benchmark by category and channel mix rather than against a blended average. The one figure here with a named institutional source is NRF and Happy Returns’ 2025 Retail Returns Landscape, which put the US ecommerce return rate at roughly 19.3% against a 15.8% blended retail rate. Coresight Research found a 23.4% average return rate for US online apparel, while ICSC research found a 5% return rate for in-store purchases. Electronics and beauty benchmarks vary by source. The more actionable metric is cost per return, and there is no authoritative flat-dollar public benchmark for it. Deloitte reports that returns can cost merchants approximately 15–30% of the returned item’s value. Calculate your own, including support labour and unsellable write-offs, and set a returnless threshold per SKU tier from that number.
What is composable commerce, and how is it different from headless?
Most ecommerce platforms are monolithic — one system handles catalogue, search, cart, checkout, content and admin together. Headless splits that in two, detaching the customer-facing shopfront from the order-processing engine so the two communicate over APIs; you can rebuild the front end without touching the back end. Composable goes further, breaking the entire stack into separate specialist services — search from one vendor, checkout from another, product information from a third — assembled through APIs. MACH is the industry acronym for the underlying principles: Microservices, API-first, Cloud-native, Headless. Headless is one architectural change; composable is a whole-stack strategy with correspondingly higher integration cost.
Should I replatform to headless or composable commerce in 2027?
Probably not as a full-stack project. Research from Shopify and IDC shows a much more mixed architecture landscape: 45% of surveyed enterprises use a composable front end with a full-stack back end, while 27% use a fully headless or modular architecture. That supports a more incremental approach rather than assuming every organization should move toward fully composable commerce.
How do I measure marketing performance when AI assistants sit between me and the customer?
Accept that user-level attribution will keep degrading and build a portfolio of imperfect methods instead of chasing one perfect one. Start with server-side conversion sending — Meta’s Conversions API and Google’s Enhanced Conversions — which recovers events that browser pixels miss; this is the highest-leverage single fix. Add a post-purchase survey asking how customers found you, which can help uncover AI-assisted discovery and word-of-mouth that traditional referral and click-based attribution misses. Use media mix modelling on aggregated data for budget allocation, since it needs no cookies or device IDs. Then run quarterly incrementality holdouts on your largest channel to settle disputes the models cannot. Report directional confidence rather than false precision.
Which ecommerce trend should a small business act on first?
Product data quality, for three compounding reasons. Structured, complete, accurate product data determines whether AI assistants and shopping agents can find and correctly describe your products; it improves conventional organic search at the same time; and accurate sizing, materials and compatibility information can reduce avoidable returns, which are one of the fastest-rising ecommerce cost lines. It requires no replatform, no new vendor, and no bet on how fast agentic commerce grows. After that, in order: landed-cost transparency at checkout if you sell cross-border, wallet coverage per market, and server-side conversion tracking.
15. Sources
All figures current as of August 2026. Forecast figures are the publisher’s, not ours; where forecasters disagree materially we have said so in the text.
- EMARKETER, worldwide retail ecommerce forecast (February 2025 series), reported via Shopify — shopify.com/blog/global-ecommerce-sales
- Oberlo, Global Ecommerce Sales Growth 2022–2028 — oberlo.com/statistics/global-ecommerce-sales-growth
- ResearchAndMarkets, AI Shopping Agents and Agentic Commerce 2026: Adoption Trends and Execution Limits (May and August 2026 releases)
- commercetools, The agentic commerce radar: Spring 2026 update, 22 April 2026, and Agentic Commerce Stats: Enterprise Guide, 2026 — commercetools.com
- Forrester, The State of Agentic Commerce, Q2 2026 — forrester.com
- Adobe Digital Insights, Q2 2026 AI Traffic Report (published 16 April 2026, covering calendar Q1 2026) — business.adobe.com
- Digital Commerce 360, “Adobe: AI-referred traffic to retail sites doubles in a year”, 17 June 2026 — digitalcommerce360.com
- Contentsquare digital experience benchmark, 2025 (99 billion sessions), reported via industry analysis
- WARC Media, The Future of Commerce Media 2026 — reported via MarketingReport and bestmediainfo, August 2026
- WARC Media retail media forecast, November 2025 — retailmediaage.co.uk
- EMARKETER, “Retailers hold AI commerce advantage over ChatGPT,” July 23, 2026 — updated forecast of $19.74 billion in AI-platform-driven US retail ecommerce sales in 2026, approximately 4% below its December 2025 projection. — emarketer.com
- EMARKETER, US retail media ad spending share by platform, 2025–26 estimates — emarketer.com
- Worldpay (now part of Global Payments), Global Payments Report 2026, 63,000+ consumers across 42 markets — worldpay.com
- Worldpay, Global Payments Report 2024 digital wallet projections to 2027 — corporate.worldpay.com
- US Customs and Border Protection, “Indefinite Suspension of the De Minimis Exemption for Merchandise Arriving Through All Modes Other Than the International Postal Network”, interim final rule, 91 FR 37789, 24 June 2026 — federalregister.gov
- One Big Beautiful Bill Act, Public Law 119-21, Section 70531(b) — terminates the de minimis exemption effective 1 July 2027
- Learning Resources, Inc. v. Trump, 607 U.S. __ (2026), decided 20 February 2026; Executive Orders 14388 and 14389 of the same date
- Euromonitor International, “The definitive end of the de minimis tariff exemption” — euromonitor.com
- ShipperHQ, “De Minimis Is Ending in the EU”, June 2026 — shipperhq.com
- DELIVER Asia 2026, session with David Meads, Field CTO, Zonos — deliver.events
- National Retail Federation and Happy Returns, 2025 Retail Returns Landscape, October 2025
- Forrester, 2022 B2B E-Commerce Forecast, US — forrester.com
- Elogic Commerce, B2B ecommerce trends and composable commerce analysis, 2026 — elogic.co
- Gartner, B2B buyer survey (646 buyers, fielded August–September 2025, published 9 March 2026): 67% prefer a rep-free experience for at least part of their purchase. An earlier Gartner survey of 632 buyers, fielded August–September 2024, found 61% preferred an overall rep-free experience — gartner.com
- Gartner, “Gartner Says By 2030 that 75% of B2B Buyers Will Prefer Sales Experiences that Prioritize Human Interaction Over AI”, press release, 25 August 2025 — gartner.com
- McKinsey & Company, B2B Pulse — buyer channel counts across the purchase journey
- Baymard Institute, cart abandonment meta-analysis (70.2% average across ~50 studies) — baymard.com
- Shopify and IDC, enterprise commerce architecture research, survey of 1,000+ enterprise organizations — 45% use a composable front end with a full-stack back end; 27% use a fully headless or modular architecture.
- Supermetrics, Facebook Ads: New historical limitations, attribution window and metric removals — Meta 7-day and 28-day view-through attribution window deprecation effective 12 January 2026 — docs.supermetrics.com
- Jon Loomer, Meta Ads Attribution 2026 — current Meta attribution-window wording including engage-through — jonloomer.com
- Google Analytics Help, Get started with attribution — retirement of first-click, linear, time-decay and position-based attribution models in November 2023 — support.google.com
- IAPP, U.S. State Privacy Legislation Tracker — comprehensive consumer privacy laws in effect by 2026 — iapp.org
- Indiana Attorney General, Indiana Consumer Data Protection Act guidance — in.gov
- Kentucky Attorney General, Kentucky Consumer Data Protection Act — ag.ky.gov
- Rhode Island General Assembly, Data Transparency and Privacy Protection Act — rilegislature.gov
- Google Ads Help, Enhanced Conversions — conversion measurement using first-party signals — support.google.com
- Stripe, Testing the conversion impact of 50+ global payment methods — measured conversion impact of offering relevant payment methods — stripe.com
- 1WorldSync, Product Content Benchmark — relationship between inaccurate product information and avoidable returns — 1worldsync.com
- ECR Retail Loss, Buy Online Return in Store — returns as a growing retail cost pressure — ecrloss.com
- Coresight Research, Shifting the Size and Fit Paradigm — U.S. online apparel return-rate research — s3.alvanon.com
- ICSC, consumer returns research — in-store versus online return rates — icsc.com
- Deloitte, retail industry outlook — returns can cost approximately 15–30% of the returned item’s value — deloitte.com
- Shopify, Returnless Refunds — growing use of returnless refunds — shopify.com
- Shopify, composable commerce guidance — integration complexity and organizational/technical maturity considerations — shopify.com
- EMARKETER, “TikTok Shop Makes Up Nearly 20% of Social Commerce in 2025,” December 9, 2025 — US social commerce sales of $87.02 billion in 2025, 18% projected growth in 2026, and TikTok Shop sales forecasts.
- Grand View Research, social commerce market — Asia-Pacific share of global social-commerce revenue — grandviewresearch.com
- European Commission, temporary €3 customs duty guidance for low-value imports — ec.europa.eu
- European Commission, low-value ecommerce import facts and figures — ec.europa.eu
- UK Government, Reforming the customs treatment of low-value imports into the UK — removal of low-value customs-duty relief by October 2028 at the latest — gov.uk
- U.S. Supreme Court, Learning Resources, Inc. v. Trump, decided 20 February 2026 — supremecourt.gov
- White House, Executive Order 14388, continuing the suspension of duty-free de minimis treatment for all countries — whitehouse.gov
- Federal Register, Executive Order 14389 — termination of relevant IEEPA duties — federalregister.gov
- Google Search Central, technical requirements and indexable file types — public accessibility is required for crawling, and PDFs are indexable when accessible — developers.google.com
- OpenAI Help Center, Publishers and Developers FAQ — public content can appear in ChatGPT Search when OAI-SearchBot is allowed access — help.openai.com