# AI integration > We add AI to the systems you already run, in the specific steps where it saves effort, and keep a person in charge of anything that reaches a customer. Source: https://www.fyadvisers.com/services/ai-integration/ Author: FYAdvisers Last updated: 2026-10-09 AI integration means adding a language model to one specific step in a workflow your business already runs: reading a free-text enquiry, pulling details out of a document, or preparing a first draft. FYAdvisers builds these integrations for small and mid-sized businesses in Singapore. We add AI only where the input varies too much for plain rules, and we design every integration so a person reviews the output before it reaches a customer. This is part of our [IT consultancy](/services/it-consultancy/) work. ## Where AI helps - **Understanding enquiries.** People describe what they need in their own words. AI can summarise the enquiry and suggest the right product, service or person. - **Reading documents.** Tenders, contracts, forms and invoices contain details buried in long text. AI can pull out requirements, dates and amounts into a structured list for checking. - **Preparing drafts.** Replies, proposals and summaries can be drafted from the information already in your systems, then edited by your team. - **Searching your own information.** Staff can ask questions across conversation history, documents or records instead of scrolling through them. - **Translation.** Messages and documents can be translated for review, which helps teams working in more than one language. ## Examples from our work At [Jeff Learning Center](/work/jeff-learning-center/), an enquiry form collects student details and AI suggests the best-matching course. The suggestion goes only to the owner, and his team contacts the lead. For [Chromatic Media](/work/chromatic-media/), AI reads uploaded tender files and prepares a compliance matrix and proposal drafts for the team to check before submitting. For [an individual PropNex agent](/work/propnex-agent/), AI extracts the people, deal details and outstanding items from WhatsApp conversations, and prepares reply drafts and English to Chinese translations that the agent reviews and sends. ## Where AI should not decide Language models can state wrong information confidently. We design around that. - **Prices, availability and commitments** are confirmed by your team, never by the model. - **Advice with legal, financial or medical weight** stays with qualified people. - **Anything sent to a customer** is a draft until a person approves it, unless you have decided the risk is low. - **Predictable decisions** use rules, which are cheaper, faster and easier to test. ## Data and privacy We agree what data each integration sends to an AI provider, what is stored, who can see it and how long it is kept. We use providers' business APIs, and we remove personal details from examples during testing where we can. If your enquiries or documents contain personal data, PDPC's guide to the [data protection obligations](https://www.pdpc.gov.sg/overview-of-pdpa/the-legislation/personal-data-protection-act/data-protection-obligations) is the place to start. PDPC has also published advisory guidelines on using personal data in AI recommendation and decision systems. ## How a project runs 1. **Pick one step.** We find a task where people read, sort or draft, and agree what a good output looks like. 2. **Test with real examples.** We run the AI on your past cases and compare the results with what your team actually did. 3. **Build it into the workflow.** The AI step is connected to your tools through a [workflow automation](/services/workflow-automation/) or a small [custom application](/services/custom-software-development/), with a clear review point. 4. **Monitor and improve.** We log what the AI suggested and refine the instructions as your team finds edge cases. ## What to prepare before we talk - The task where people spend time reading, sorting or drafting. - A handful of real examples, with personal details removed. - What a good result looks like, and who should check it. - Any rules about where your data can be processed. [Get in touch](/#contact) and we will tell you whether AI is worth adding, and where. ## Frequently asked questions ### What is AI integration? AI integration means adding an AI model to a specific step in an existing workflow or application, such as reading an enquiry, extracting details from a document or preparing a draft. It is different from giving staff a general chatbot to use on their own. ### Which AI models do you use? We choose the model per task, based on accuracy, cost, speed and where the data is processed. We use well-known commercial models through their business APIs and explain the choice before building. ### Is our data used to train AI models? It should not be. We use business API terms where providers state that inputs are not used for training by default, and we agree with you what data is sent, what is stored and for how long. ### Do you build customer-facing chatbots? Rarely by default. We usually use AI behind the scenes to sort, summarise and draft, with a person replying. An automatic reply can be added where it is low risk, but prices, commitments and advice stay with your team. ### How do you check that the AI is accurate? We test with your real examples before launch, keep a person reviewing the output, and log what the AI suggested so mistakes can be spotted and the instructions improved.