AI CAPABILITY • DEPLOYMENT
Implementation & Integration
What to deploy, how to connect it
In 30 Seconds
After strategy, governance, and data decisions comes a practical question: what do we actually deploy? Options range from browser-based AI to fully integrated enterprise systems – and the right choice depends on your specific context.
Most organisations don't need custom infrastructure. They need the right combination of existing tools, properly integrated and governed.
Where we help: Navigating choices, avoiding over-engineering, and ensuring what you deploy actually gets used. The goal is capability, not complexity.
The Implementation Spectrum
From simplest to most sophisticated – each level has its place.
Browser-Based
Lowest barrier
What: ChatGPT, Claude.ai, Gemini web interfaces
Integration: None – standalone tools
Data control: Usage policy-based; data may be used for training unless enterprise tier
Best for: Getting started, individual productivity, low-sensitivity tasks
Workspace-Integrated
Common enterprise path
What: Microsoft 365 + Copilot, Google Workspace + Gemini
Integration: Built into existing productivity tools
Data control: Enterprise agreements, data stays within tenant
Best for: Broad productivity gains, organisations already on these platforms
IDE & Developer Tools
Technical teams
What: VS Code + Copilot/Codex, Cursor + Claude, JetBrains AI
Integration: Connected to repositories, project context
Data control: Configurable; can exclude sensitive files, use enterprise tiers
Best for: Development teams, technical work, code-heavy organisations
API & Custom Integration
Flexible, more effort
What: OpenAI API, Anthropic API, MCP servers, custom applications
Integration: Full control – connect to any system, build custom workflows
Data control: Code-controlled; explicit decisions about what data goes where
Best for: Custom applications, domain-specific tools, sophisticated integrations
Private & On-Premise
Maximum control
What: Azure OpenAI, AWS Bedrock, self-hosted models (Llama, Mistral)
Integration: Enterprise infrastructure, private networks
Data control: Complete – air-gapped options, data never leaves your environment
Best for: Regulated industries, sensitive data, specific compliance requirements
Key Decision Factors
Five questions that drive implementation choices
1. Data Sensitivity
What's the most sensitive data AI will touch? Client confidential? Personal data? Regulated information? This drives your security model more than anything else.
2. Existing Platforms
Microsoft shop or Google shop? What development tools do you use? Working with existing platforms is usually faster and cheaper than adding new ones.
3. Use Case Breadth
General productivity across the org? Specific technical work? Domain-specific applications? Broad use cases favour platforms; specific needs may justify custom builds.
4. Technical Capability
What can your organisation actually manage? API integration requires developers. Custom infrastructure needs DevOps. Be realistic about what you can sustain.
5. Speed to Value
How quickly do you need results? Browser-based AI works today. Workspace integration takes weeks. Custom builds take months. Starting simple and iterating usually beats building the “perfect” system that never ships.
Common Patterns
What we typically see working for different organisation types
Starter Pattern
Small teams, limited AI experience, budget-conscious
Setup: Browser-based AI (Claude Pro, ChatGPT Plus) + clear usage policy + shared prompt library. Low cost, fast start, builds fluency before bigger investment.
Enterprise Productivity Pattern
Established organisations, broad productivity focus, existing Microsoft/Google investment
Setup: Microsoft 365 + Copilot OR Google Workspace + Gemini across the org. Governance embedded in existing admin controls. Training on effective use.
Technical Team Pattern
Development teams, technical consulting, code-heavy work
Setup: IDE integration (Cursor + Claude, VS Code + Copilot) connected to internal repos + MCP servers for tool connectivity + clear policy on what code can be shared.
Advanced Integration Pattern
Organisations with custom workflows, domain-specific needs, technical capability
Setup: API access (OpenAI, Anthropic) + custom applications + MCP for system integration + knowledge management systems feeding context. Higher investment, higher customisation.
Regulated/Sensitive Pattern
Financial services, healthcare, legal, government, highly sensitive data
Setup: Azure OpenAI or AWS Bedrock (private endpoints) + data classification tiers + strict access controls + audit logging. May include air-gapped options for most sensitive work.
Security Models
Different approaches to AI data security – choose based on your risk profile
| Model | Data Handling | Best For |
|---|---|---|
| Public APIs + Policy | Clear guidelines on what can/can't be shared with AI | Non-sensitive work, general productivity |
| Enterprise Agreements | Data processing agreements, no training on your data | Business data, internal documents |
| Private Endpoints | Data stays in your cloud tenant, dedicated resources | Sensitive data, compliance requirements |
| Air-Gapped / On-Premise | No external connectivity, self-hosted models | Highly regulated, classified information |
Hybrid approaches are common: Public AI for general productivity, enterprise tier for business documents, private endpoints for client data. The key is clear classification – everyone knows which tier applies to what.
The Integration Layer
How AI connects to your existing systems
MCP (Model Context Protocol)
Anthropic's open standard for connecting AI to external tools and data. File systems, databases, APIs, web browsers – MCP provides a consistent interface.
Growing ecosystem of pre-built connectors. Increasingly supported across tools.
Direct API Integration
Custom code connecting AI APIs to your systems. Full flexibility but requires development effort. Common for internal tools and custom workflows.
OpenAI, Anthropic, Google all provide robust APIs with SDKs.
Platform Connectors
Built-in integrations within platforms. Copilot accesses Microsoft Graph, Gemini connects to Google Workspace data. Zero-code but limited to platform boundaries.
Fastest path for organisations already invested in these ecosystems.
Automation Platforms
Zapier, Make, Power Automate – connecting AI to workflows without code. Good for business users, limited for complex logic.
Bridge between no-code and custom development.
How We Help
Navigating choices, avoiding over-engineering, ensuring adoption
Implementation Assessment
Review your current state, use cases, and constraints. Map the decision factors. Recommend an implementation approach that fits your actual context – not a generic enterprise architecture.
Platform Selection Support
Navigate the vendor landscape. Evaluate options against your specific requirements. We're vendor-neutral – the goal is finding what works for you, not selling a particular stack.
Integration Design
Design how AI connects to your systems. MCP server setup, API integration patterns, data flow architecture. Practical implementation that your team can maintain.
Pilot & Rollout
Start with a focused pilot. Learn what works. Then scale. We help design pilots that generate real learning and rollout approaches that drive adoption.
Related Capabilities
Implementation decisions connect to the layers before and after.
Data & Knowledge Foundation
Understanding your data landscape drives implementation choices.
Learn more →Context Engineering
Once deployed, build the knowledge systems that make AI useful.
Learn more →Strategy & Governance
Implementation happens within strategic direction and governance guardrails.
Learn more →Start With What Works
The best implementation is the one that actually gets used. If you're navigating AI deployment choices – what to buy, how to integrate, where to start – we can help you find the right path for your context.