Using CAF to Guide Cloud Work
Case study
Outcome
- Cut legacy system costs by 75% by moving Azure VMs to Databricks and enabling ML model deployment without downtime.
- Set up and tracked costs for a Databricks sandbox used for more than 100 training sessions and proofs of concept.
- Supported greenfield Azure Databricks infrastructure using Infrastructure as Code (Bicep) and hub-and-spoke architecture aligned with the Azure Cloud Adoption Framework.
- Delivered Databricks database migrations and Genie and LLM work for clients.
- Contributed to CAF, WAF and CCoE discussions with client leadership.
Context
In consulting, a technical recommendation can easily sound like personal opinion, particularly when it involves an expensive platform change or cloud migration.
I used Microsoft’s Cloud Adoption Framework (CAF) to ground those conversations in recognised guidance. It gave clients and consultants a shared way to discuss the current platform, the risks in it and the practical next steps.
CAF’s anti-patterns were particularly useful. Instead of criticising a decision, I could describe the pattern we were seeing, explain the risk or cost it introduced and point to the recommended alternative.
Challenge
At 4most, many client relationships began with Chief Risk Officers. To discuss wider cloud strategy and implementation, we also needed to reach the people responsible for the client’s technology estate. CAF gave us a natural route through the Cloud Centre of Excellence (CCoE), where one existed.
Approach
Where a CCoE existed, it brought cloud maturity, governance, data platforms and implementation into the same conversation. Where it did not, we could discuss whether creating one would help. Asking to speak with the CCoE also framed the conversation around governance and alignment, rather than simply asking for access to the CTO.
These conversations helped us build relationships beyond the initial risk engagement and work with the wider technology organisation.
One client initially asked for generative AI in its platform. After working through the requirement, we found that the real need was text-to-SQL: querying data in plain language without writing code. We built a purpose-designed table and used Databricks Genie to provide that interface. It took days to deliver and cost a fraction of the proposed GenAI implementation. Challenging the original brief helped turn a one-off piece of work into a longer relationship.
Results
Client relationships
- Built relationships with stakeholders outside the original risk engagements
- Created warm introductions to other 4most business units
Cloud governance
- Contributed to CAF, Well-Architected Framework and CCoE discussions with client leadership
- Reviewed cloud architecture and made recommendations
- Discussed security, compliance, cost management and governance