Our approach
The AI CROPS Framework
Cloud, Resilience, Operations, Performance, Security and Sustainability: six pillars we use to assess, design and run every system we work on. It is our engineering methodology, not a product you buy.
Six pillars
What we look at, and what we put in place
- 01
C — Cloud
Architecture that fits the workload and spends money deliberately.
Questions we ask
- Is each workload on the right service and the right size?
- Can every environment be recreated from code?
- Does every team see the cost of what it runs?
What we put in place
- Landing zones and account structure
- Infrastructure as code
- Tagging, budgets and cost-per-service reporting
- 02
R — Resilience
Systems that degrade gracefully and recover predictably.
Questions we ask
- What happens when a zone, dependency or model provider fails?
- Are backups tested by restoring them?
- Are recovery objectives defined and met?
What we put in place
- Multi-zone design and failure-mode analysis
- Backup, restore and disaster-recovery drills
- Fallbacks for AI and third-party dependencies
- 03
O — Operations
Changes and incidents handled through automation, not heroics.
Questions we ask
- Is every change reviewed and deployed through a pipeline?
- Do runbooks exist for the alerts that page people?
- Is toil measured and reduced?
What we put in place
- CI/CD and GitOps
- Runbook and incident automation with approval gates
- Post-incident reviews
- 04
P — Performance
Latency, throughput and model quality measured against user-facing objectives.
Questions we ask
- Are there SLOs that reflect what users experience?
- Is capacity planned or reactive?
- Is AI quality measured on every release?
What we put in place
- SLOs and load testing
- Autoscaling and capacity planning
- Evaluation suites for ML and LLM features
- 05
S — Security
Least privilege, defence in depth and evidence that controls work.
Questions we ask
- Who and what can access production and sensitive data?
- Are misconfigurations caught before deployment?
- Can you prove your controls to an auditor?
What we put in place
- Least-privilege IAM and secrets management
- Policy as code and posture management
- Automated compliance evidence; AI guardrails
- 06
S — Sustainability
Efficient use of compute, storage and accelerators — good for cost and for emissions.
Questions we ask
- Are idle and oversized resources removed?
- Are AI workloads sized to the task?
- Is the carbon footprint of the platform visible?
What we put in place
- Rightsizing and scheduling of non-production
- Model selection and caching to reduce inference compute
- Provider carbon-footprint reporting
How it is used
Applied at every stage of delivery
In assessments
Each pillar is scored with evidence, giving a baseline and a prioritized list of gaps.
In architecture reviews
Designs are checked pillar by pillar before build, and trade-offs are recorded in decision records.
In operations
Recurring reviews track the same pillars after go-live, so improvements continue rather than stopping at launch.
Want a framework review of your platform?
Tell us what you are building or fixing. An engineer — not a sales script — will review it and suggest a sensible first step.





