1 · Should we replatform?

Probably not. The numbers on what headless and open source builds actually cost, and the narrow cases where they earn it.

2 · Does design still sell?

Yes, but it stopped closing. Shoppers verify everything now, and a new kind of shopper reads your data instead of your design.

3 · Is agentic shopping real?

Real, measured, and growing fast. AI-referred retail traffic grew 393% year over year in Q1 and now converts better than every paid channel.

4 · What should we do in 90 days?

A structured-data sprint, not a rebuild. Five steps, none of which require touching your platform.

Audience: owners, operators, and digital leaders of any store doing real online revenue.

The headline

• $7.4 trillion global e-commerce this year, growing 8% YoY
• On pace for $8 trillion by 2027
• Online takes 21.8% of all retail
• 2.86 billion online shoppers
• ~28 million stores, ~2,100 new ones launching daily

The three facts that matter more

• Acquisition is expensive, loyalty is thin. CAC up 40 to 60% in two years, now $68 to $84. 72% of shoppers switched brands this year. Returning customers still drive ~60% of revenue.
• Direct is normal. US DTC hit $239 billion, 19.2% of retail e-commerce. 52% of shoppers browse internationally.
• A new traffic source appeared. And it converts. Next slide.

Open source

• License is free; the responsibility is the price
• Security patches, plugin conflicts, performance tuning: all yours
• Past ~100k orders/year, scaling is real infrastructure work
• Magento-to-composable migrations rate complexity 8 of 10: a rebuild wearing a migration costume

Staying put has a cost too

• Monolith switching costs grow ~15 to 20% per year
• The $200k replatform of today is $280k in two years

React templates and headless

• Beautiful day one; quarterly npm audits and framework breaking changes after
• Platform apps (reviews, upsells, loyalty) inject scripts that fail headless and must be rebuilt as components
• The most common failure is organizational: no in-house React capacity, so every copy change becomes an agency ticket

When headless earns it

• Multi-brand / multi-region on one backend
• Storefront is the differentiator
• $50M+ GMV and a real engineering bench

What agents do not see

• Your color palette
• Your hero video
• Your typography
• Your animations

What agents read

• Structured data and schema markup
• Your product feed
• Your reviews, with recency
• Your shipping, returns, and warranty policies
• Ten competitors, in milliseconds

The measurable gap

Adobe's AI Content Visibility Checker scores US retail homepages at 75% machine-readable on average.

Product pages score 66%.

Roughly a third of the average product page is invisible to the agent deciding whether to recommend it, and the product page is the page that closes the sale.

The consequence

A store that wins the human eye and fails the machine read loses sales it never knew it was considered for.

The numbers

• $20.9 billion AI-driven retail spend estimated for 2026, ~4x 2025
• 7.3% of US e-commerce revenue through AI-assisted transactions projected by Q4; above 12% in high-SKU categories
• McKinsey: $3 to 5 trillion agent-mediated by 2030. Bain: 15 to 25% of online retail by decade end
• 62% of consumers research with AI; only 23 to 30% trust an agent to spend their money

The protocol question, resolved

• April 24: Amazon, Meta, Microsoft, Salesforce, Stripe joined the Universal Commerce Protocol Tech Council, alongside Google, Shopify, Etsy, Target, Wayfair
• OpenAI retired Instant Checkout in March, pivoted to retailer apps inside ChatGPT
• The working model: discover in the chat, transact on the merchant's site. Walmart's in-chat checkout converted at one-third the rate of click-outs
• ChatGPT's share of gen-AI traffic fell from 87% to ~57% in fourteen months. No single channel is a safe solo bet

Do not rebuild for one protocol

Production agentic checkout (ChatGPT retailer apps, Amazon Buy for Me, card-network agent programs) is concentrated among giant retailers. The direct-API door OpenAI opened in 2025 closed within six months.

Do not buy bespoke agent checkout

Merchants present across multiple open surfaces report roughly 40% more agentic traffic than single-protocol peers. The durable surface is the open one: a storefront machines can read, verify, and trust.

1 · Pixels are table stakes

Replatforming your pixels costs six figures and buys little. The narrow headless cases are real; most stores are not in them.

2 · Data is the contest

Shoppers verify, and now their agents verify faster. Accurate, structured, consistent product data is the new brand surface.

3 · The sprint beats the rebuild

Making your existing store legible to machines costs a structured-data sprint and buys entry to the fastest-growing traffic source in retail.

4 · The test is checkable

The brands that win the agentic era will be the ones whose claims a machine can check.