AI for Cambodian businesses is most useful when it solves a specific task, not when an organisation adopts AI simply because it is becoming popular. For Cambodian SMEs, hotels, NGOs and professional organisations, the practical starting point is usually prompting an existing AI tool for manual work, then moving to APIs, specialised systems or agents only when the simpler approach is no longer sufficient.
HubSpot’s AI for Business Builders resource presents four levels of working with large language models: prompt engineering, prompt engineering with an API, fine-tuning and building a dedicated large language model. It positions these as progressively more technical approaches, with the first level suitable for many general manual tasks. Written by: Kim
Start with a business problem, not an AI tool
A common mistake is choosing an AI tool first and then searching for something for it to do. The better approach is to identify work that is repetitive, time-consuming or difficult to manage consistently.
For a Cambodian organisation, useful starting points could include:
- Drafting customer responses
- Turning meeting notes into structured summaries
- Creating first drafts of marketing copy
- Analysing customer feedback
- Preparing job descriptions
- Organising internal information
- Summarising documents
- Analysing business data
HubSpot specifically identifies marketing, customer experience, human resources and operations as areas where prompt engineering can support manual tasks.
The important question is not whether AI can perform a task. It is whether using AI improves a measurable result without creating more checking work than it saves.
When prompt engineering is enough for Cambodian SMEs
Prompt engineering is the simplest level described in HubSpot’s framework. It involves giving an AI model clear instructions and enough context to produce a useful output. HubSpot recommends this approach for manual tasks that do not require automation or highly specialised domain knowledge.
This is often the most sensible starting point for a small Cambodian business because it requires less technical development than building an integrated AI system.
A hotel, for example, could use AI to create first drafts of responses to common guest questions. Staff could then check the information before sending it.
A retailer could use AI to turn approved product information into different descriptions for English and Khmer audiences. Staff would still need to verify prices, availability and product claims.
An NGO could use AI to organise meeting notes or create a first draft of a report from verified programme information. Sensitive information should only be entered into an AI system when the organisation’s data and privacy requirements allow it.
Why better prompts produce more useful business outputs
AI does not automatically understand the exact business result you want. HubSpot explains that effective prompting requires clear and concise directions and enough information to reduce ambiguity.
A useful business prompt should normally specify:
- Task: what the AI needs to do
- Goal: what the output should achieve
- Audience: who will use or read it
- Context: information the AI needs
- Format: how the result should be presented
- Constraints: what it must avoid or preserve
For example, instead of asking AI to write a hotel advertisement, provide the actual hotel information, intended audience, offer, booking action, language and required format. This gives the model a defined task rather than a vague request.
Use examples when consistency matters
HubSpot describes few-shot prompting as giving an AI model examples so it can produce outputs that follow the desired format or style. Its example includes using previous meeting summaries to teach the model a consistent structure.
This can be particularly useful when a Cambodian organisation repeatedly produces the same type of document.
Examples include:
- Customer service replies
- Property listing descriptions
- Meeting summaries
- Social media captions
- Sales follow-up messages
- Programme updates
- Internal reports
Give the AI several examples that represent the standard you actually want. Then check whether new outputs follow the same structure before using the process at scale.
When Cambodian organisations should consider an API
An API becomes relevant when people repeatedly perform the same AI task and manual prompting starts creating unnecessary work. HubSpot identifies larger datasets, domain-specific knowledge and specific process flows as situations where API-based use can become appropriate.
For example, imagine a Cambodian company receives hundreds of enquiries through its website. Staff could manually paste each enquiry into an AI tool, but that does not scale well.
An API could connect the AI capability to the company’s existing system so that a defined process happens automatically, subject to appropriate controls and human review.
Potential applications include:
- Classifying incoming enquiries
- Extracting customer requirements
- Routing enquiries to the correct department
- Creating structured summaries
- Generating first drafts of responses
The decision should be based on volume, repetition and the cost of manual work. A small organisation with 20 relevant enquiries a month may not need an API. An organisation processing hundreds or thousands of similar records may have a stronger reason to investigate one.
When fine-tuning is worth investigating
Fine-tuning is substantially more specialised than writing better prompts. HubSpot describes it as adapting a pre-trained model for a particular industry or task, and identifies large quantities of domain-specific data and sufficient budget as important considerations.
That makes fine-tuning a poor starting point for most small organisations. Having a collection of documents does not automatically mean fine-tuning is necessary.
Before considering it, ask:
- Do we have enough high-quality domain-specific data?
- Is prompting or retrieval already insufficient?
- Is the task repeated often enough to justify development?
- Can we measure whether the specialised model is better?
- Can we afford development, testing, maintenance and evaluation?
If the answer to these questions is mostly no, start with a simpler approach.
Why building your own AI model is usually the wrong starting point
HubSpot describes building a dedicated large language model as expensive and time-consuming, and notes that this approach is more relevant to large organisations seeking highly specific capabilities.
For most Cambodian SMEs, hotels and NGOs, building a foundation model would be disproportionate to the problem. The organisation is more likely to benefit from applying existing AI models to a clearly defined workflow.
There is a major difference between using AI and building an AI model. Most organisations need the first one.
How different Cambodian sectors can apply AI
The best application depends on the organisation’s work and the information it already has.
| Organisation | Potential AI task | Measure |
|---|---|---|
| Hotels | Draft responses to common guest enquiries | Response time and booking enquiries |
| Retailers | Create first drafts of product descriptions | Production time and content accuracy |
| Property firms | Structure enquiries and listing information | Processing time and qualified leads |
| NGOs | Summarise meetings and organise programme information | Staff time saved and review accuracy |
| Schools | Create first drafts of learning or administrative materials | Preparation time and quality checks |
| Professional services | Draft reports, proposals or follow-up messages | Preparation time and revision rate |
These are use cases to investigate, not claims that AI will automatically improve results in these sectors. Each organisation should test the application against its own data and workflow.
AI should support Khmer and English workflows differently
Cambodian organisations often communicate with more than one audience. Local customers, foreign residents, tourists, international buyers and donors may require different information even when they are interested in the same organisation.
AI can help create or adapt language versions, but translation should not be treated as a purely mechanical task when accuracy affects a commercial or organisational decision.
For Khmer-language communication, check terminology, names, prices, locations and culturally specific wording. For English-language communication aimed at international customers or donors, check whether the message provides the context that audience needs to make a decision.
The objective is not to produce more translated content. The objective is to make the correct information understandable to the intended audience.
How to measure whether AI is actually helping
AI adoption should have a baseline. Without one, an organisation cannot tell whether the new process is better than the old one.
Choose one measurable result before introducing the AI process:
- Time: How long does the task take before and after AI?
- Quality: How often does the output require significant correction?
- Capacity: How many tasks can the team complete within the same working period?
- Cost: What does the AI process cost compared with the existing process?
- Business result: Does the change improve enquiries, bookings, sales, service response or another relevant outcome?
For example, if a business uses AI to draft customer replies, measure the time saved and the percentage of drafts requiring major edits. If quality falls, the time saving may not justify the process.
What Cambodian businesses should not automate first
Do not start with decisions where an incorrect output could create significant financial, legal, employment, safety or reputational consequences.
Human review is especially important when AI is handling customer commitments, financial information, contracts, sensitive personal information, employment decisions or claims that require evidence.
AI can prepare a draft or organise information. That does not make its output automatically correct.
A practical AI decision framework for Cambodian organisations
Use the simplest level that solves the problem.
| Situation | Approach to investigate | Why |
|---|---|---|
| Occasional manual task | Prompt engineering | Low technical effort |
| Repeated workflow with high volume | API integration | Reduces repetitive manual steps |
| Large specialised dataset and defined task | Fine-tuning or specialised implementation | May improve performance for a specific use case |
| Highly specialised large-scale AI requirement | Custom model development | Only justified by substantial requirements and resources |
Moving to a more advanced level should happen because the simpler approach has reached a measurable limitation, not because the technology sounds more sophisticated.
How Cambodian organisations can start with AI
Choose one repetitive task that currently consumes staff time. Record how long it takes and what a good result looks like. Test an AI-assisted version using approved information, then compare the time, quality and outcome.
If prompting solves the problem, there may be no reason to build anything more complicated. If the process is high-volume and repetitive, investigate an API. If the organisation has substantial specialised data and a clear performance problem that simpler approaches cannot solve, then consider more advanced options.
HubSpot’s AI for Business Builders resource provides a useful framework for thinking about these different levels of AI use. Its central practical lesson is that businesses do not need to begin with the most technically advanced option.
For Cambodian SMEs, hotels, NGOs and other organisations, the strongest AI investment is the one that solves a defined problem, can be checked by people and produces a measurable improvement. Save and download for later.