Nothing gets built until the decisions are made.

The Transformation Blueprint is how Kilvora designs Microsoft Fabric platforms. Ten architectural decision areas, with major decisions documented through evidence, rationale and expected consequences before implementation begins.

Architecture & Platform Design, made tangible. The Blueprint is the method behind Kilvora's Architecture & Platform Design capability. It opens an engagement before implementation starts, whether delivery is embedded or fixed scope.
Built against the design, not around it.
01

The principle

Fabric platforms fail on decisions, not on technology.

Almost every failing Fabric estate I am called into is failing for the same reason. Not because the tooling was wrong, and rarely because the engineering was poor. It is failing because nobody decided what was being built before people started building it.

Pipelines end up running under a developer's account. Transformation gets pushed into the raw layer. Workspaces appear without owners. Capacity is selected through a licence conversation rather than workload evidence. These are not simply build mistakes. They are decisions that were never made, and delivery proceeded around the gap.

These are not separate failures. They are one architectural failure, surfacing in different places.

The asymmetry is what makes the design worth doing properly. Deciding the identity model before the first pipeline runs costs a conversation. Retrofitting it across a live estate with security already in place is expensive rework. The same is true of layer design, governance, capacity and deployment.

So the design comes first, and it is written down. That is the method.

02

When it applies

Three situations, one method.

The design work is consistent. What changes is how much of the estate already exists, and how much of it should survive the design.

Starting point

Nothing built yet

Fabric is agreed, licences may already be bought, and nobody has decided what the platform should look like. The Blueprint establishes the design that implementation is then measured against.

Starting point

Build already underway

An internal team or delivery partner has started, but coherence is the concern. The method reviews what exists, keeps what is sound, and designs forward from there.

Starting point

Assessment completed

A Readiness Assessment has recommended proceeding. The design phase turns that recommendation into an architecture and reuses the relevant discovery rather than repeating it.

03

The design standard

What the method holds itself to.

Design work is easy to describe and harder to evidence. These are the standards that must be visible in the resulting architecture and working documents.

Every design produces

  • Clear technical and business reasoning for every major architectural decision
  • An architecture shaped around the estate, workloads, operating model and constraints
  • Workspace, domain, governance and ownership boundaries agreed before uncontrolled growth begins
  • A capacity recommendation derived from workload evidence rather than licence assumptions
  • Security, deployment and operational controls designed before they become retrofit work
  • An implementation sequence that explains what is built first and why

What it never becomes

  • A generic reference architecture with the client name added
  • A default medallion pattern applied without considering the workload
  • Recommendations without enough reasoning to challenge or revisit them
  • A Fabric feature showcase presented as architecture
  • A slide deck used once and then separated from delivery
  • A design created from a template before the estate has been understood
04

What gets decided

The ten architectural decision areas.

Every Fabric platform must resolve all ten areas, whether or not anyone notices. The method is to work through them deliberately, in an order that prevents one choice from quietly constraining another.

01

Workspace and domain topology

How the Fabric estate is organised across business domains, environments and shared platform services: workspace boundaries, domain hierarchy, ownership, isolation and promotion paths.

02

Lakehouse strategy and layer design

Whether a medallion approach is appropriate, how Lakehouses and layers are structured, what each layer is responsible for, and where raw, standardised and serving data separate.

03

Semantic layer strategy

Direct Lake against Import, model scope and ownership, refresh strategy, certification and the standards that keep the semantic layer maintainable.

04

Capacity and SKU planning

The SKU the workload evidence supports, capacity placement across domains and environments, pause and resume, and a modelled cost trajectory rather than a launch price.

05

Security architecture

Identity model, workspace RBAC, row-level security design, sensitivity labelling, PII handling and how compliance obligations are met.

06

Governance and ownership model

Named domain and workspace ownership, workspace approval, certification, naming standards, change control and the operating model that holds the platform together.

07

Deployment and environment strategy

DEV, TEST and PROD workspace structure, Git branching, deployment pipelines, configuration management and how a failed release is rolled back.

08

Monitoring and observability

Run logging, failure alerting, data quality visibility, capacity tracking, per-workspace attribution and who is told when something breaks.

09

Implementation sequencing

The order the platform is built in: domain and workspace foundations first, security before data, governance before self-service, controls before pipelines and pilot before rollout.

10

Integration and dependency architecture

How Fabric meets the rest of the estate: source systems, retained Azure services, the existing Power BI estate, identity and ITSM processes, and third-party tooling.

05

The record

Every decision is written to be questioned later.

Most architecture documents record what was chosen. Kilvora records the evidence, rationale and consequences behind major decisions so they can be questioned and revisited later. Where a genuine trade-off exists, the material alternatives are also captured.

ADR-002 · CAPACITY & SKU PLANNING Accepted

Production runs on F64. Development and test share an F8 with overnight pause.

Context

The SKU in place was selected during a licensing conversation, before any workload existed. Nobody had modelled consumption.

Evidence

Workload modelling puts peak consumption at 52 CUs during the morning batch window, driven by four ingestion pipelines running concurrently against overlapping sources.

Rejected

F32, which covers average load at lower cost but throttles during the morning window. That delay pushes the semantic refresh beyond the hour people rely on the reports.

Consequences

Higher committed spend than F32. Headroom for the second domain without another SKU change. Dev and test pause outside working hours, recovering part of the difference.

Review

Reassess once the second domain is in production, or if the batch window changes.

One of typically 12 to 20 decision records produced across the ten decision areas.
06

What exists at the end

Seven working documents the build runs on.

Not a slide deck presented once and filed. These are working documents used throughout delivery and updated when a decision changes.

A

Enterprise Fabric Blueprint

Executive summary, maturity baseline, decision records, target architecture narrative, governance, security and sequencing rationale.

B

Architecture diagrams

Target state, workspace and domain topology, logical business data flow, security architecture and deployment model, each written for the audience that needs it.

C

Governance framework

Workspace standards, certification model, naming conventions, change control and ownership, written so it can be adopted operationally.

D

Capacity recommendation

Workload inventory, peak-load analysis, SKU recommendation with clear rationale, environment configuration and cost trajectory. Material alternatives are compared where they affect cost, risk or performance.

E

Risk and dependency register

Architectural risks and inter-domain dependencies, each with impact, trigger, owner and mitigation.

F

30/60/90 day roadmap

Foundation, pilot, validation and expansion, expressed as an architectural sequence rather than a generic project plan.

G

Executive presentation

Current state, target state, the decisions that matter most, governance, cost, roadmap and top risks, written for the people funding the build.

07

How it runs

Four stages before implementation begins.

Nothing is drafted before discovery is complete. Every decision is traceable to evidence from the estate, not to a preference brought in on day one.

Stage 1 — Discovery

Understand the estate as it is

Executive alignment, technical deep dive, domain sessions, security and compliance, BI and platform operations. Evidence from the environment is reviewed alongside what people say.

Stage 2 — Design

Make the ten decisions

Each major decision is taken against the evidence, with its rationale and consequences stated. Competing options are compared only where the trade-off materially affects cost, risk, scalability or operations. Capacity and governance models are built from what discovery found, never from a template.

Stage 3 — Challenge

Your architects try to break it

The design is put in front of the people who know the estate and challenged before anything is finalised. Decisions that survive that session are the ones worth building on.

Stage 4 — Commit

The design becomes the delivery baseline

Executive presentation, technical walkthrough and open risks stated plainly. Any departure from the design becomes a decision made consciously and on the record.

08

Delivery models

Then the platform gets built.

The design phase opens the engagement. What follows is the same architectural baseline carried into delivery rather than a separate document handed over and forgotten.

Day rate

Embedded architecture and engineering

I join your team as the architect and, where appropriate, as the engineer. The design phase runs first, then the platform is built against it as scope evolves.

Fixed price

Fixed scope platform delivery

A bounded piece of platform is delivered to an agreed definition of done. The design phase establishes the scope before price is fixed, which makes the commitment evidence-based rather than optimistic.

09

Common questions

Before you reach out.

How does this relate to the Readiness Assessment?

The Readiness Assessment determines whether Microsoft Fabric is the right direction and which workloads should move. The Transformation Blueprint begins once that decision has been made and defines how the platform should be designed before implementation starts.

We already have an architecture. Is this still relevant?

If your architecture is documented, agreed across teams and supported by clear technical reasoning, it may not be. If key decisions exist only in people's heads or across disconnected documents, the method creates one defensible baseline for delivery.

Is every organisation expected to work through all ten decisions?

The ten decision areas provide the framework. Where an area is genuinely not relevant, that is recorded explicitly as a constraint rather than left undefined or skipped by omission.

What happens when delivery changes the original design?

The design is a baseline, not a prison. When evidence changes or a new constraint appears, the relevant decision is reopened, the trade-off is documented and the architecture is updated deliberately.

Start with the right architectural decisions.

Tell me what you are planning to build, what already exists and where the uncertainty sits. A short conversation establishes which delivery model fits and what needs to be decided first.

Start a conversation