Build, Buy, or Extend: The AI Decision Framework
Learn the framework for deciding when to buy AI off-the-shelf, extend Copilot with plugins and agents, or build custom AI solutions from scratch.
Three paths to AI
Think of AI like a kitchen. You can eat out (buy), customise a meal kit (extend), or cook from scratch (build).
Buy means using Copilot for M365 as-is. It works out of the box — no setup, no customisation. Like ordering from a restaurant: fast, reliable, but you eat what’s on the menu.
Extend means adding custom capabilities to Copilot — plugins, connectors, and agents that make it smarter for YOUR organisation. Like a meal kit: you get the ingredients and recipe, then add your own twist.
Build means creating a completely custom AI solution using Foundry or Azure OpenAI. Like cooking from scratch: total control, but you need skills, time, and ingredients.
The decision framework
| Factor | Buy (Copilot as-is) | Extend (Copilot + plugins) | Build (Custom AI) |
|---|---|---|---|
| Time to deploy | Days to weeks | Weeks to months | Months to years |
| Cost | Licence only ($30/user/mo) | Licence + development | Significant dev + infrastructure |
| Control | Limited to built-in features | Moderate — custom within the platform | Full control over everything |
| Maintenance | Microsoft handles updates | Shared — Microsoft platform + your customisations | You own all maintenance |
| Technical skill needed | IT admin | Power users or developers | AI/ML engineers and developers |
| Best for | Standard M365 productivity | Organisation-specific workflows | Unique, competitive-advantage AI |
| Risk level | Low | Low to medium | Medium to high |
When to Buy
Choose “buy” when:
- The use case is standard M365 productivity (drafting, summarising, analysing)
- Speed matters more than customisation
- You want the lowest risk and fastest ROI
- Your team lacks technical resources for customisation
When to Extend
Choose “extend” when:
- Copilot needs to access data outside Microsoft 365
- Users need Copilot to perform organisation-specific tasks
- You want to stay on the Copilot platform but add custom intelligence
- Your team has power users or developers who can build extensions
When to Build
Choose “build” when:
- No existing product meets the requirement
- The AI application is a competitive differentiator
- You need full control over the model, data pipeline, and user experience
- You have the engineering team and budget to develop and maintain it
Exam tip: The exam favours the simplest sufficient approach
When the exam presents a scenario, apply this priority:
- Can you Buy? If Copilot handles it out of the box, that’s the answer.
- Can you Extend? If Copilot needs a plugin or connector, that’s next.
- Must you Build? Only if neither Buy nor Extend works.
The exam penalises over-engineering. If a scenario describes a common productivity need, “build a custom solution in Foundry” is almost always the wrong answer.
The Copilot extensibility framework
When you choose to “extend,” Microsoft provides a structured framework:
| Extension Type | What It Does | Who Builds It | Example |
|---|---|---|---|
| Declarative agents | Custom agents defined by instructions, knowledge, and skills | Power users or devs | An agent that answers questions about company policies using SharePoint docs |
| API plugins | Connect Copilot to external APIs and services | Developers | A plugin that lets Copilot query Salesforce CRM data |
| Graph connectors | Bring external data into the Microsoft 365 index | IT admins or devs | Connect ServiceNow tickets so they appear in Copilot search results |
| Custom engine agents | Agents powered by custom AI models (via Foundry) that surface in Copilot | AI engineers | A specialised industry compliance agent using a fine-tuned model |
| Message extensions | Interactive cards and forms within Copilot conversations | Developers | A booking extension that displays available meeting rooms |
How extensions layer on top of Copilot
Think of it as a stack:
| Layer | What It Provides |
|---|---|
| Copilot for M365 (base) | Built-in productivity in Word, Excel, Teams, Outlook |
| + Graph connectors | External data becomes searchable alongside M365 data |
| + Declarative agents | Custom agents with specific knowledge and instructions |
| + API plugins | Real-time actions in external systems |
| + Custom engine agents | Completely custom AI models surfaced within Copilot |
🏗️ Ravi’s decision tree for TechVantage
Ravi, CTO of TechVantage Solutions, has five AI needs. He applies the framework:
1. Help developers write documentation faster
- Decision: Buy — Copilot in Word handles this natively
- Reasoning: Standard productivity use case, no customisation needed
2. Let sales reps query the CRM from within Teams
- Decision: Extend — API plugin connecting Copilot to Salesforce
- Reasoning: Copilot can’t access Salesforce by default, but a plugin solves this
3. Make IT knowledge base searchable in Copilot
- Decision: Extend — Graph connector for the Confluence wiki
- Reasoning: Bring external docs into the M365 search index
4. Build an AI assistant that analyses code quality
- Decision: Build — Custom application using Foundry
- Reasoning: Highly specialised, competitive differentiator, no off-the-shelf solution
5. Create an onboarding agent that guides new hires through their first week
- Decision: Extend — Declarative agent in Copilot Studio
- Reasoning: Custom knowledge + structured workflow, but doesn’t need a custom model
Notice the pattern in Ravi's decisions
Ravi used all three strategies:
- 2 Buy decisions for standard productivity (documentation)
- 3 Extend decisions for organisation-specific needs (CRM, knowledge base, onboarding)
- 1 Build decision for a unique, competitive-advantage application (code quality AI)
This is the typical enterprise pattern. Most needs are met by Buy or Extend. Building is reserved for truly unique requirements.
Ravi's IT team at TechVantage wants Copilot to answer questions using data stored in their ServiceNow instance. What is the best approach?
Ravi needs AI to help developers write documentation faster. They already use Word. What approach should he take?
Next up: Microsoft Foundry: Your AI Platform — understand when and why to use Foundry for custom AI development.