Copilot, Gemini, or Custom System? A Decision Framework for Generative AI Tools in Hong Kong Enterprises

There is a visible gap in how Hong Kong enterprises are using generative AI.

According to the Hong Kong Productivity Council's 2025 AI Workplace Readiness Survey, 88% of surveyed enterprise employees already use AI tools in daily work, and 92% of enterprises report plans to introduce AI into workflows. But among SMEs that have adopted AI, only around 32% use paid tools — the majority rely on free versions (HKPC SME Index, Q1 2026).

Adoption rates are high, but most usage stays at the individual level rather than reaching systematic deployment. One reason is that enterprises lack a clear understanding of the capability boundaries between different tools — they do not know when a subscription is sufficient and when a custom system is needed.

This article provides a decision framework, not a tool ranking.

The AI tool access environment in Hong Kong

Hong Kong enterprises face a specific access environment in tool selection. The OpenAI ChatGPT interface and the Anthropic Claude.ai interface are not currently open for direct registration by Hong Kong users. The underlying models from both companies remain accessible via enterprise cloud platforms — OpenAI models via Microsoft Azure OpenAI Service, and Claude models via the Anthropic API.

The practical effect is: consumer-facing versions are restricted; enterprise versions and the API layer continue to function normally.

The major platforms currently directly accessible in Hong Kong include: Microsoft Copilot (integrated into Microsoft 365), Google Gemini (officially opened to Hong Kong users from March 2026), Grok (xAI), and Chinese-developed language models including DeepSeek.

The capability boundaries of three tool categories

Generative AI tools can be divided into three functional categories, each with different suitable scenarios.

1. Microsoft 365 Copilot / Google Gemini for Workspace

These tools are deeply integrated into office suites. Copilot provides AI assistance within Word, Excel, Outlook, and Teams. Gemini integrates with Google Docs, Sheets, and Gmail.

Suitable scenarios: individual productivity improvement, document drafting, meeting summaries, email organisation. For enterprises already making heavy use of Microsoft 365 or Google Workspace, this is the lowest-friction starting point.

Capability boundaries: these tools are designed to assist existing office workflows, not to replace or reconstruct business processes. The depth of enterprise knowledge base integration is constrained by platform architecture. Memory and context cannot be migrated outside the platform. Usage data belongs to the platform, not the enterprise.

2. API-layer tools (via Azure OpenAI, Anthropic API, etc.)

Enterprises access underlying language models via cloud APIs and build applications on their own infrastructure. This layer supports a higher degree of customisation: specific prompt engineering can be designed, enterprise knowledge bases can be connected, and data flow can be controlled.

Suitable scenarios: enterprises that need to embed AI capability into existing business systems (ERP, CRM, document management); scenarios that require customising AI behaviour to match specific business logic.

Capability boundaries: requires technical capability to build and maintain the application layer. Compared to office suite integration, deployment cycles are longer and upfront investment is higher.

3. Custom AI systems

A complete custom system includes: business process analysis, RAG knowledge base design, custom memory architecture, system integration, and deployment. The entire system is built on the enterprise's own business logic and data assets.

Suitable scenarios: enterprises with large volumes of unstructured internal documents requiring AI processing; highly specialised business processes that off-the-shelf tools cannot accommodate; situations with strict requirements around data sovereignty or multilingual accuracy.

Capability boundaries: long build cycles; requires specialist engineering capability or an external technical partner. Six common functional gaps are detailed in Why Enterprises Need an AI Consultant, Not Just Claude or Poe.

Three key decision questions

Which tool category is right depends on the combination of answers to the following three questions.

Question 1: Are the business processes highly repetitive and predictable?

If the tasks the AI needs to handle have fixed input and output formats — extracting specific fields from PDFs, generating reports from templates, classifying customer queries — a fixed pipeline design has clear advantages in cost and auditability. If task boundaries are fuzzy or dynamic judgement is required, evaluate whether an agent architecture is appropriate. The relevant analysis is in Fine-Tuning, RAG, or Prompt Engineering? A Practical Decision Guide for Enterprise AI Architecture.

Question 2: Is access to internal enterprise documents a core requirement?

If the AI system needs to accurately reference internal enterprise documents — contracts, manuals, historical records — the knowledge base depth of office suite AI is typically insufficient. The Chinese-English mixed nature of Hong Kong enterprise documents creates additional requirements for RAG system design. The practical challenges in a bilingual environment are worth understanding before making a selection.

Question 3: Are data sovereignty and cross-border data flow key considerations?

If the AI system needs to process customer personal data or commercially sensitive documents, which service provider's servers the data is stored on, and which jurisdiction's laws govern it, are part of compliance responsibility. Both Copilot and Gemini store data in their respective enterprise cloud environments. Custom systems can be designed to operate on enterprise-owned infrastructure. Data flow considerations for Hong Kong enterprises handling confidential documents provides a specific evaluation framework.

Practical performance differences in Traditional Chinese

Hong Kong business communication switches frequently between Traditional Chinese, English, and Cantonese. This has a real impact on AI tool selection, but it is often overlooked.

Several points are worth noting: major AI platforms differ in their handling of Traditional Chinese versus Simplified Chinese — some models, when Traditional Chinese is not explicitly specified, tend to drift toward Simplified characters and vocabulary habits. Cantonese comprehension and generation capability varies across platforms. The accuracy of mixed Chinese-English queries differs based on platform architecture and training data distribution.

Before deploying any AI system to handle Traditional Chinese enterprise documents or customer communications, validation testing against actual business queries is a necessary step, not an optional one.

A practical starting point for decisions

The following framework can serve as a starting point for evaluation, not a final answer.

If the enterprise requirement is primarily individual productivity improvement, and Microsoft 365 or Google Workspace subscriptions are already in place: evaluate the AI features of the existing suite before deciding whether additional tools are needed.

If the enterprise needs AI to access internal knowledge bases or reconstruct business processes: the capability limits of office suite integration will become apparent quickly. API-layer or custom systems are worth including in the evaluation.

If the enterprise has specific requirements around data sovereignty, multilingual accuracy, or deep integration with existing systems: a full requirements analysis is recommended before comparing tool feature lists. The procurement framework in AI Consulting in Hong Kong: Six Questions Before You Buy is a useful reference.

Levi is an independent AI engineer based in Hong Kong, building production-grade LLM applications and RAG pipelines for enterprises in Hong Kong and the Greater Bay Area. Academic background includes a Master's in Business Analytics and AI from Vlerick Business School, Belgium. Production deployments include an insurance industry RAG comparison system and HKSoka (hksoka.com), a multi-layer memory AI platform. Past work has also been applied in client projects.

smartai.hk+ai.consulting@proton.me
linkedin.com/in/levi-innovation