Method
From business problem to adopted AI solution
The same five steps apply in every domain. What changes is the vocabulary, the data and the risks.
Discover
Map stakeholders, run pain-point workshops, and build a longlist of AI ideas tied to revenue, cost or efficiency. Output: stakeholder map, as-is process, opportunity list.
Prioritise
Score every idea on business value, data readiness, feasibility, risk and adoption ease, so the first pilot is chosen on evidence. Output: weighted scoring matrix and a one-page business case.
Define
Write requirements engineers can build from: user stories, acceptance criteria, data requirements, and AI-specific needs such as accuracy targets, evaluation and human-in-the-loop rules. Output: requirements package, risk register.
Prototype
Test the riskiest assumption quickly with data analysis, an LLM workflow or a small demo, before anyone commits to a full build. Output: working prototype and evaluation results.
Adopt
Plan the change: training, rollout, measures of success, and a feedback loop so the solution keeps improving. Output: adoption plan and KPI dashboard.
Technical edge
Data
SQL, Excel, Power BI. Python for data analysis (currently learning, practised through the projects on this site).
AI
Prompt design, RAG concepts, LLM workflow design, evaluation planning, responsible-AI checklists.
Systems
Requirements for enterprise platforms (including IBM Maximo), APIs and integrations, process and data modelling (BPMN, UML).