How we work
Four stages. Each one is worth having on its own.
They overlap, and most engagements do not need all four. We would rather you understood the shape of the work before you commit to it than discover it in month four.
The engagement
VisionIQ, Platform, Migration, ManagedIQ.
Fixed fee for the defined stages and a monthly service for the continuing one. No time and materials except for genuine outliers, and we will say so up front if we think you are one.
VisionIQ
Week one
Agree the business question and the decision behind it, map the data and platform environment it touches, prioritise the use cases, and build a working proof of value on your own data. Roughly seven working days of our effort, once we have access.
- Current-state map
- Governance criteria
- Now, Next and Near backlog
- Working proof of value
Platform
Runs alongside
The production foundation. Fabric workspaces and OneLake patterns, Foundry projects, semantic models, identity, security, deployment automation, observability and cost controls, established inside your own tenant rather than ours.
- Governed platform foundation
- Deployment and operating patterns
- Cost and access controls
Migration
Only if you have something to move
Recovering and moving the data, reports, pipelines, procedures and semantic logic that still matter, with parity preserved where the business or a regulator requires it and modernization where it creates value. Plenty of engagements never need this stage.
- Migrated and validated workloads
- Preserved business logic
- Documented gaps and disagreements
ManagedIQ
Continuous
A monthly service. New data sources, new use cases, new agents and new reports, worked against a backlog you can see, with a record of what improved and why. This is where capability actually accumulates.
- Continuous enablement
- Performance and cost optimisation
- Compounding organisational context
The method
Speed comes from changing how the work is done.
This is the part most firms leave vague, and the vagueness is exactly what makes people suspicious. They are right to be. Here is the mechanism.
During delivery, AI does not analyse your data. That distinction is the whole thing. Large models are genuinely good at reading code and genuinely capable of being confidently wrong, and those two facts do not coexist safely in a regulated reporting system.
So we use AI to build the analysis harness, which is deterministic code that reads your system. The harness does the analysis. It is re-runnable, so the same inputs produce the same answers every time, and we can run it again against your full dataset when we have it. Nothing is hallucinated into existence.
What that buys is scale. A person reading every calculation, every feed and every report in a fifteen-year-old system takes months and misses things. A harness reads all of it, repeatably, in days.
The tooling we deliver with does not ship to you. If you want grounded copilots or agents running in your own estate, that is something we build deliberately, with its own governance, its own identity model and its own monitoring. It is a decision you make on purpose, not one that arrives as a side effect of the platform work.
Set the vision for your platform.
Bring one business question, or one system that has to move. We will map what it touches and tell you what a working proof of value could show in the first week.