Skip to content

Architecture | Scale

Platform selection, system design, integration architecture and scale planning. The work that happens before the build, and determines most of what the build will cost.

How engagements work

In one paragraph

A written document: the decision, the options considered, what each would cost to build and to run, the failure modes of each, and a recommendation with the reasoning attached so it can be argued with. Typically covering platform selection, how the commerce platform and the surrounding systems divide responsibility, where product and customer data is owned, the integration pattern between systems, and what the architecture does at three times current volume. The deliverable is the reasoning, not a diagram.

What this is

Most expensive commerce problems are architecture decisions made quickly two years earlier: the platform chosen because a vendor was persuasive, the integration built point-to-point because it was the fastest route, the data model that assumed one warehouse.

None of those is visible when it is made. They surface as a second brand nobody can launch, a catalogue that cannot be extended, an ERP integration that breaks whenever volume moves. The work here is making those decisions deliberately, with the trade-offs written down while they are still cheap to change.

What this page is not

This is design and decision work, and the deliverable is a document rather than a running system. If you have already decided and need it built, the platform pages cover the implementation.

What usually brings this in

  • Three vendors have proposed three different platforms.

    Nobody internally can evaluate the technical claims, and the decision is drifting toward whoever presented best.

  • Launching a second brand or market means rebuilding.

    The architecture assumed one of everything, and the cost of the assumption arrives all at once.

  • Every system talks to every other system directly.

    Point-to-point integrations grow with the square of the system count, and no one can change anything safely.

  • Nobody can say what happens at three times the traffic.

    Capacity is managed by over-provisioning and hoping, which is expensive and still fails on the day it matters.

What gets built

Platform selection with the reasoning written down
Requirements against candidates, cost to build and to run, and an explicit statement of what each option would be bad at. A recommendation you can argue with is worth more than one you have to trust.
System and data ownership design
Which system owns product, price, stock and customer, and what the others are allowed to assume. Most integration pain is really an ownership question nobody answered.
Integration architecture
Event-driven, queued or synchronous, chosen per flow against what the business actually tolerates when one side is down.
Multi-store and multi-market structure
Designed so the second brand, market or currency is configuration and content rather than a fork.
Scale and failure planning
Where the system breaks first, what it does when it breaks, and what it costs to move that ceiling. Answered with numbers.
Technical due diligence
An independent read on a codebase, a proposal or an estimate, for a buyer who cannot evaluate it internally.

Technologies

Platforms assessed
  • Magento 2
  • Adobe Commerce
  • Shopify Plus
  • WooCommerce
  • Headless
Integration
  • REST
  • GraphQL
  • Webhooks
  • RabbitMQ
  • ETL
Scale
  • Varnish
  • Redis
  • Elasticsearch
  • CDN
  • Read replicas
Infrastructure
  • AWS
  • Docker
  • CI/CD
  • Observability

How delivery runs

  1. The actual question

    Often not the one in the brief. "Which platform" is usually "what does this business need the platform to do in three years", and the second question has a defensible answer.

  2. Read what exists

    The codebase, the integrations, the data, the release process, and the numbers. Opinions here are worth little without reading the thing first.

  3. Options, with costs attached

    Two or three real options, what each costs to build and to run, and what each would be bad at. An option with no stated downside has not been examined.

  4. A recommendation that can be argued with

    One recommendation, the reasoning, and the conditions under which it would change. Written so someone who disagrees can say precisely where.

  5. A sequence, if it is wanted

    What to do first, what can wait, and what must not start until something else is finished. Phasing is where most good architecture decisions are lost.

Questions

How long does an architecture engagement take?
A platform selection or a due diligence read is usually one to three weeks. A full architecture for a replatform is longer and is better phased, because decisions made after reading the code are considerably better than decisions made before.
Will you recommend a platform you do not work on?
Yes, and it happens. A recommendation that always lands on the platforms I build for is a sales pitch with a document attached, and it is worth nothing to you.
Can you review a proposal from another vendor?
Yes, and it is one of the highest-value things on this page. An independent read on an estimate you cannot evaluate internally costs a fraction of the project and occasionally saves the whole of it.
Do you stay involved after the architecture is agreed?
Where it is useful. An architecture nobody owns during the build tends to be followed for about six weeks. That can be a review cadence rather than a full engagement.

Describe what is in front of you and you get a written scope and a figure, usually within a few days of a short call. No charge for either.

All services