Composable Commerce
Composable Commerce, With Shopify Still Underneath It
A headless PWA that carried a brand from $15M to $150M in annual revenue. Sub-second load times. 11+ years exclusive to Shopify.
Modular architecture optimized for speed, performant at scale, and designed so your marketing team can still ship a page without filing a ticket. We keep Shopify as the back end — its checkout, its primitives, its admin, and its roadmap.
Experience
18+
Years of Unlocking Growth
Scale
$3.14B+
in GMV Migrated to Shopify
Partnership
1 of 5
Founding NA Shopify Platinum Partners
Precision
94.7%
Implementation Predictability
Clients
8
Unicorns & Counting
Headless PWA architecture supporting six years of growth from $15M to $150M in annual revenue, at sub-second load times, across 47 countries on one domain.
When composable is the right answer — and when it isn’t
It’s likely right when
- Multiple storefronts share one commerce back end. Rothy’s ran this shape across Canada, the UK, and Australia alongside 31 retail locations.
- Search and filtering hit real limits. Vuori’s build included Algolia with optimized sync pipelines and custom handling of Shopify’s API limits.
- The content experience genuinely exceeds the theme layer. Four Sigmatic’s education-heavy content engine and AG1’s headless CMS migration are both this case.
- Markets need materially different experiences — different merchandising logic, content structures, and journeys per region.
- Non-commerce surfaces need to transact — apps, kiosks, in-store displays, or partner properties sharing the same commerce back end.
It’s likely wrong when
- The driver is performance. Good Ranchers left headless and got mobile product pages 67% faster with 45% lower total cost of ownership. Native is usually sufficient and always cheaper.
- The driver is “modern architecture.” Not a business case, and it won’t survive scrutiny from finance.
- There’s no internal owner. Thesis invested in headless without the engineering depth to sustain it, and left.
- The team ships frequently through the theme. Good Ranchers’ headless stack had put campaigns and promotions into the engineering queue.
Composable work we’ve run
AG1
A headless PWA that carried $15M to $150M in annual revenue at sub-second load times, across 47 countries on one domain
Explore the AG1 success storyVuori
Headless Shopify with checkout rebuilt on Checkout Extensibility, plus a VIP portal for 80,000+ members
Explore the Vuori success storyRothy’s
$0 to $220M and a $1B+ exit, across three international markets and 31 retail locations
Explore the Rothy’s success storyFour Sigmatic
A PWA that cut load times below a second and dropped bounce rate 90%
Explore the Four Sigmatic success storyGood Ranchers
Off headless in eight weeks — 45% lower TCO, 67% faster mobile PDPs, and marketing off the ticket queue
Explore the Good Ranchers success storyAnatta was fundamental in helping us make great technology decisions.
Ben SmithCOO, AG1
Anatta is experienced across all major platforms, with a broader eCommerce business perspective than the other agencies we’ve worked with.
Stephen HawthornwaiteCo-Founder, Rothy’s
Composable projects fail in predictable ways
The architecture is sound and the failure modes are well documented. Nearly all of them come from the same root decision: treating composable as an opportunity to replace Shopify rather than to extend it.
Architecture risk
Decoupling for performance that was available anyway
Slow Shopify stores are typically slow because of app load, unoptimized media, and theme architecture, none of which decoupling fixes. Four Sigmatic’s PWA cut load times below a second; Good Ranchers came off headless and saw pages load 67% faster. Same goal, opposite architectures, both correct for their situation. → Enterprise Shopify Development
Losing the platform roadmap
A build that treats Shopify as a commodity API stops receiving what Shopify ships. AG1’s and Vuori’s headless builds both kept Shopify underneath, which is why both kept receiving platform capability.
Rebuilding checkout
Checkout Extensibility now handles most of the customization that used to require replacing checkout. Vuori’s Project Alpine reduced a three-page checkout to one, on the extensibility framework rather than a replacement.
Service sprawl
The right number of services is the smallest number that meets the requirement. → Integrations · Technology Stack
Organizational risk
The marketing team loses autonomy
Good Ranchers’ headless stack had put campaigns and promotions in the engineering queue. Before a build is finished we test whether merchandising can launch a campaign page without engineering.
No in-house capability to own it
Thesis invested in a headless storefront without the engineering depth to sustain it. Vuori went the other way, structuring the engagement so Anatta’s engineers bolted onto theirs.
Chosen as a preference rather than a business case
If the business case is “modern architecture” or “flexibility,” it isn’t finished.
Three things we’ve concluded about composable architecture
These come from building on this platform exclusively for eleven years, in both directions.
Two brands went headless and two came back, and all four were right
Four Sigmatic and AG1 moved to headless PWAs and got sub-second load times and years of growth. Good Ranchers and Thesis left headless architectures and got faster, cheaper, and easier to operate. The architecture is a consequence of the situation.
Decouple incrementally, one surface at a time
A single big-bang cutover concentrates all the risk on one date. Taking one surface at a time produces the same destination with recoverable steps and real data at each stage.
The assessment is priced so that “no” is a real answer
Our first phase can conclude with a recommendation not to proceed, and it has, more than once. It’s scoped as its own engagement so that outcome isn’t a commercial problem for anyone in the room.
The work
Architecture assessment and business case
What’s actually impossible on the native architecture, what decoupling would cost across three years, and our honest recommendation — including when it’s not to decouple.
Front-end architecture
Hydrogen and Oxygen, or a custom framework where requirements warrant it, architected for performance and customizability.
Content modeling and editing experience
The content architecture and authoring interface that determine whether your team can operate the site — scoped as a requirement, not a phase two that doesn’t arrive.
Storefront API and data layer
Query design, caching strategy, and rendering approach — the decisions that determine whether a decoupled front end is faster or slower than what it replaced.
Checkout and native capability retention
Native checkout with Checkout Extensibility, plus the platform capabilities worth keeping: markets, B2B, customer accounts, admin tooling.
Incremental migration path
A sequenced route from where you are to where you’re going, one surface at a time, with each step independently valuable and reversible.
What a composable engagement commits to
| Commitment | Acceptance criterion |
|---|---|
| Shopify stays the back end | Checkout, admin, and platform capabilities retained unless a documented requirement forces otherwise |
| Platform currency | You continue receiving what Shopify ships; deviations documented with their cost |
| Marketing autonomy | An agreed list of what your team can publish without engineering — specified before build and tested at handover |
| Performance | Measured against the native-architecture baseline, not against an unmeasured “before” |
| Service count | Justified per service, with the operating cost of each stated |
| Ownership | A named internal owner and a working handover, or a written plan to establish one |
| Reversibility | Each migration step independently deployable and independently reversible |
The third row is the one that tends to get cut under time pressure, and the one clients raise a year later.
How a composable engagement runs
Assessment and business case
Weeks 1–3
Native-architecture baseline, requirement analysis, three-year cost model, capability gap review, recommendation.
Architecture and content modeling
Weeks 2–6
Front-end architecture, data layer, content model, editing experience, service selection.
Incremental build
Ongoing
One surface at a time, each independently deployable, measured against baseline.
Handover and run
Continuous
Documentation, internal enablement, monitoring, and platform-update maintenance.
Timelines: A single decoupled surface can ship in weeks. A full storefront migration for a multi-brand operation is a multi-quarter program. The assessment tells you which one you’re looking at.
Anatta’s Agentic Operating System
Every build runs on one system.
Senior architects paired with an agentic framework that absorbs the commodity execution — component scaffolding, test coverage, documentation — so senior time goes to the architecture decisions that are expensive to reverse.
See how we workHow to tell whether you’re ready for composable
Work through these before talking to anyone, including us.
- Name what’s impossible today. Specifically. If the list is short and vague, the business case isn’t ready.
- Establish your native performance ceiling. Get a real measurement with app load, media, and scripts properly handled.
- Identify the owner. Who maintains a front-end application and gets paged for an incident?
- List what your team publishes without engineering today. That’s the autonomy you’re putting at risk.
- Count the services you’d add. Total the annual cost.
- Model three years, not one. Build plus maintenance plus dependency updates plus incidents.
- Decide what stays native. Checkout, admin, markets, customer accounts.
- Find your first surface. If you can’t identify one section to decouple first, you’re planning a big-bang migration.
The Composable Readiness Framework
The worksheet version, with the three-year cost model structure included. Name and email.
Frequently asked questions
Do we need to go headless?
Probably not. Many brands considering it are solving for performance, which is usually available on Shopify’s native architecture. Composable is right for specific requirements: multiple storefronts on one back end, search and filtering at scale, content experiences the theme layer can’t support, or markets needing different journeys.
Will we lose Shopify’s platform updates?
Not if Shopify stays your back end. What loses you the roadmap is treating Shopify as a replaceable commerce API.
What happens to our marketing team’s ability to ship?
It depends entirely on whether content architecture was treated as a requirement or an afterthought. We specify what your team must be able to publish without engineering before the build starts.
Hydrogen and Oxygen, or a custom front end?
Hydrogen and Oxygen for most cases. A custom framework makes sense where you have existing front-end infrastructure worth building on, or a requirement Hydrogen doesn’t serve.
Do we have to rebuild checkout?
No, and we’d argue against it. Checkout Extensibility covers most of what previously required replacing it.
What do we need in-house to own a composable stack?
Someone who can maintain a front-end application, manage deployments, and respond to incidents. Where the capability doesn’t exist and there’s no plan to create it, we’d recommend a different architecture.
Can we do this incrementally?
Yes, and we’d recommend it — one surface at a time, each independently deployable and reversible.
Tell us what you can’t build today.
