
Pillar 3: Workloads
Develop and deploy workloads that meaningfully impact the organisation’s mission.
Executive Snapshot
We’re not concerned with one app; we’re interested in one thousand workloads.
Balance Incremental, Extensible and Differential AI to hedge risk and maximise ROI.
Empower Power Users and Communities of Practice to extend AI themselves.
Why the Workloads Pillar Matters
Workload is not a throw-away word; it is the front- and back-end combination required to produce an AI-driven response or action.
Strategy & Vision set aspiration; Ecosystem Architecture built foundations. Workloads answer the question: how will we use AI to solve real-world challenges?
Workload Prioritization
Our goal is a prioritised roadmap of workloads to modernise with AI or build anew—core driver of ROI.
Techniques to rank workloads
Alignment to executive vision
Legacy location / technology sunset
Usage telemetry (MAU/WAU, data volume)
Security or compliance risk
Target technologies investment
Qualitative Assessment: compare impact vs. complexity → Quick Wins, Big Wins, Nice-to-Haves, Wastes of Time
Incremental AI
Incremental AI applies intelligence to tasks a human already performs
- speed, efficiency, scale.
How to source Incremental workloads
Turn on AI you already own – deploy Microsoft Copilots with high impact.
Ask your people – pain-points and wish-lists.
Rationalise workloads – bake AI into every app rationalisation exercise.
Mine processes – use Process Advisor to surface automation candidates.
Benchmark peers – monitor competitor adoption.
Extensible AI
Extensible AI occupies the broad middle range where Incremental workloads are extended with your data and context.
Example spectrum → Deploy Copilot for M365 → Custom Agent over Blob → Full RAG agent over OneLake
Architect’s mantra: Seek first to extend, then build bespoke.
Key extensibility tools: Graph Connectors, Teams Message Extensions, API Plugins, Copilot Studio Agents.
DiFFerential AI
Differential AI encapsulates moon-shot workloads that humans alone could not achieve - your secret sauce.
Hallmarks: creative thinking, mixed data sets, time-sensitive computation, higher implementation risk.
Example: DeepMind AlphaFold predicting protein structures beyond human timescales.
Rule of thumb: Bake Differential workloads into the roadmap early, but hedge with less complex bets.
Power Users
Power Users bridge general staff and technical teams, pushing AI tools beyond default use-cases and driving adoption.
Enablement checklist
Establish clear roles & Communities of Practice
Provide Copilot Studio and Azure AI Studio access
Train in prompt-engineering & data-security
Use rings-of-release and WAU metrics to track uptake
Success Checklist
✔ Roadmap ranks workloads by impact & complexity
✔ Portfolio balanced across Incremental, Extensible, Di8erential
✔ Extensibility patterns standardised (Graph Connectors, Agents)
✔ At least one Differential “moon-shot” in active discovery
✔ Power Users active; CoPs sharing best practice
FAQs
-
Incremental augments existing tasks; Extensible adapts vendor AI with your data; Differential delivers novel outcomes once impossible - see spectrum diagram.
-
Provide Copilot Studio, set clear roles, enforce data-security training and monitor adoption via rings-of-release before scaling.