From a business question to a testable use case.
01
Strategy and readiness
Review business objectives, AI experience, data foundations, governance, security and organisational readiness.
02
Planning and adoption
Work through skills, the adoption plan, governance, cloud cost management and the first AI workload.
03
Architecture and deployment
Design and review the architecture, security and Azure landing zone needed for the agreed proof of concept.
04
Proof of concept
Implement and test the chosen use case against the business question and agreed success measures.
Bring the questions you need to resolve.
This engagement can help an organisation exploring AI for the first time, extending an existing Copilot investment or connecting business data to an AI use case. The assessment defines which data, people, access permissions and technical dependencies the proof of concept requires.
01
Give the first experiment a useful boundary
The published AI Jumpstart starts with strategy and readiness, then connects a chosen use case with architecture, security and a proof of concept. A useful first experiment has a business owner, representative work and a decision it will inform. It should be possible to explain why a correct result matters and what happens when the system is wrong.
A document-search assistant, information-extraction workflow and agent that updates a business record require different evidence and permissions. Begin with that distinction. The experiment should expose which parts of the task can be supported by existing Microsoft capabilities and which would require integration or custom development.
02
Review organisational readiness alongside data readiness
Prepare examples of the work, the sources used today and the people qualified to judge the result. Identify the owners of those sources and how often the information changes. A technically accessible document is not automatically authoritative, current or suitable for the intended audience.
Microsoft’s generative AI evaluation guidance provides a basis for examining quality and risk. For an agent, also consider the choices and actions in the workflow. The customer should define acceptable behaviour, review ambiguous examples and identify actions that need human approval. Evaluation should include failures and limits as well as successful demonstrations.
03
Treat the proof as a decision point
A proof of concept should leave the customer with test evidence, known limitations and the requirements of the next stage. A convincing presentation alone is not sufficient evidence for production use. Record the expected task volume, selected services, access model and operational responsibilities before estimating expansion.
The engagement scope should identify assessment, preparation, architecture and proof work explicitly. Production integration, monitoring, support and later improvement need their own commitments. The customer remains responsible for business acceptance and ownership of the process; Braintree’s delivery responsibilities should be clear in the proposed scope.
04
Decisions a useful proof can inform
Requirement
What to examine
Decision to make
Task and authority
The requested output and whether the system may change records.
Answer, draft, recommend or act, with human review where required.
Evidence quality
Representative sources, expected outputs and meaningful failure cases.
The test set and acceptance criteria used to judge usefulness.
Production readiness
Integration, cost, access, monitoring and an operating owner.
Whether to expand, revise or stop the experiment.
EVIDENCE & DOCUMENTATION
See the work in context.
BEFORE YOU DECIDE


