2.3.2. Build, Buy, or Extend Decisions for AI Components

Every AI component in the solution requires a sourcing decision. The architect must evaluate whether to build custom, buy a third-party solution, or extend an existing Microsoft capability. The exam frames this as "Analyze whether to build, buy, or extend AI components for business solutions."

Decision Framework:

Evaluation Criteria:

When to Build:

• The requirement is truly unique to the organization
• Competitive advantage depends on proprietary AI capabilities
• No existing solution provides adequate functionality even with extension
• The organization has the talent and infrastructure to sustain custom AI

When to Buy/Use Prebuilt:

• The business process aligns with a standard pattern (customer service, finance reconciliation, demand forecasting)
• Time-to-value is critical
• The organization lacks specialized AI development talent
• Maintenance and updates should be the vendor's responsibility

When to Extend:

• A prebuilt solution covers 60-80% of requirements
• The gaps can be filled through configuration, plugins, or connectors
• The extension points are well-documented and supported
• The total cost of extension is lower than building from scratch

Third-party components and regulated domains: for a third-party component, integration complexity with your Microsoft stack (Copilot Studio, D365, Foundry connectors) is the strongest long-term viability factor — difficult integration means custom code at every touchpoint and ongoing maintenance — ahead of market share or price. In a regulated domain (for example, pharma/GxP), buy a specialized service when it meets the compliance requirements and supplies validation documentation; build only if no compliant service exists.

Role of the AI CoE: it prevents redundant custom development by maintaining a reuse library of existing AI components, an inventory of Microsoft capabilities, approved technology lists and mandatory evaluation gates, so teams check for an existing capability before custom work is approved. Without that evaluation framework, organizations discover afterwards that Microsoft already offered equivalents of what they built.

Exam Trap: The exam frequently presents scenarios where the "build custom" option sounds impressive but is unnecessary. A common pattern: the question describes a standard customer service use case and offers "build a custom NLP model" as a distractor, when "configure Copilot Studio with business-specific topics" is the correct answer. Always check whether the requirement can be met through configuration or extension before recommending custom development.

Reflection Question: A retail company wants AI-powered product recommendations. They have a small data science team (3 people) and want the solution live in 8 weeks. They're evaluating: (a) building a custom recommendation model, (b) using D365 Commerce AI features, or (c) integrating a third-party recommendation API. Which do you recommend and why?