How to Build a Personal AI Assistant for Your Business
A personal assistant who knows your business, your preferences, your clients, and your priorities — available 24 hours a day, never sick, and infinitely patient with repeated questions. Building a customised AI assistant for your business is now achievable without a technical team. This guide shows you how.
Beyond Generic ChatGPT
The difference between using Claude.ai directly and having a customised business AI assistant is context. Claude.ai starts every conversation without knowing anything about your business, your clients, your processes, or your preferences. A customised business assistant is pre-loaded with all of that — your company knowledge base, your client profiles, your standard operating procedures, your product or service information, and your communication preferences.
The customised assistant answers business-specific questions that a generic AI cannot: what is the status of the Acme project, what are our standard payment terms, how do we handle client scope changes, what was the decision we made on the pricing review last month? It drafts communications in your brand voice, generates analysis from your specific business data, and provides recommendations based on your actual situation rather than hypothetical examples.
What to Build
Layer 1: The knowledge base
Everything the assistant needs to know about your business: your company overview and positioning, your services or products with detailed descriptions, your standard processes (how you onboard clients, how you handle complaints, how you price projects), your team structure and roles, your key clients and relationship context (anonymised if appropriate for privacy), your pricing and commercial terms, your most common FAQ from clients and prospects, and any internal policies that affect how decisions are made. This knowledge base is stored in a Bubble.io database and passed to Claude as context in every assistant interaction. The assistant that knows your business is exponentially more useful than the one that does not.
Layer 2: The assistant interface
A Bubble.io application provides the assistant interface: a chat window where you ask questions and the assistant responds with full context of your business. The key design elements: the conversation history is maintained so follow-up questions work naturally (you can ask then what should I do next without repeating the context), the relevant sections of the knowledge base are retrieved and included in each prompt based on the query topic (not the entire knowledge base for every query — just the relevant sections), and the assistant can trigger actions (create a GoHighLevel task, draft an email for review, retrieve a specific client record) rather than just providing information.
Layer 3: System integrations
The most powerful version of the business assistant connects to your live business data: the GoHighLevel CRM (so the assistant can answer what is the status of leads from this week or which clients have overdue follow-ups), the Xero accounting data (so it can answer what is outstanding on the Smith invoice or what was our revenue last month), and the project management tool (so it can answer where does the Henderson project stand). Make.com provides the integration layer: when the assistant receives a query that requires live data, it calls Make.com, which retrieves the relevant data from the connected system and passes it back to Claude for the response.
The Practical Steps
Write the knowledge base
The most important and most underestimated step: writing the knowledge base that the assistant will use. This is not a technical task — it is a documentation task. Write in plain English: your company overview (who you are, what you do, who you serve), your services (detailed descriptions with scope inclusions and exclusions), your processes (step-by-step for the most common workflows), your commercial terms (pricing, payment terms, change request process), your team (roles and responsibilities), and your key client context (the relevant background on each active client relationship). This documentation should be thorough — the assistant cannot answer questions about information that is not in the knowledge base.
Build the Bubble.io assistant application
Create the Bubble.io application: a ConversationMessage data type (role, content, timestamp, conversation ID), a KnowledgeBaseSection data type (topic, content, priority), and a Conversation data type (user, created date, last message date). Build the chat interface: a Repeating Group showing all messages in the conversation, a text input for the user’s query, and a send button that triggers the workflow. The workflow: create the user message record, retrieve relevant knowledge base sections based on keyword matching, call the Claude API with the conversation history and relevant knowledge base sections in the system prompt, create the assistant message record, refresh the repeating group.
Configure the system prompt
The system prompt defines the assistant’s behaviour: You are the personal business assistant for [founder name] at [company name]. Your role is to help with any business question, task, or decision. Respond based on the company knowledge and context provided below. When answering: (1) be specific and direct — give clear recommendations rather than vague options when you have enough information, (2) reference specific company information when relevant — not general advice, (3) flag when a question requires information you do not have and suggest where to find it, (4) for draft communications, match the communication style specified in the brand voice guide below. Company knowledge: [retrieve relevant sections from the Bubble.io knowledge base database and paste here]. Brand voice: [paste your brand voice guide]. This system prompt, maintained and updated as the business evolves, is the intelligence layer of the entire assistant.
Connect to live data
Build the Make.com webhooks that the assistant triggers for live data queries. When the assistant detects that the user is asking about live CRM data (what is the status of…, how many leads…, which clients are…), it calls the Make.com webhook with the query, Make.com retrieves the relevant GoHighLevel data, and returns it as structured text that Claude incorporates into the response. The same pattern works for Xero data (financial queries), project management data (project status queries), and any other connected system. The assistant that combines static knowledge base context with live system data provides answers that no generic AI tool can match.
How do I keep the knowledge base current?
Build the knowledge base update habit: whenever a significant process changes, a new service is added, or a key commercial term is updated, update the relevant knowledge base section in Bubble.io immediately. Assign one person as the knowledge base owner — responsible for ensuring the content stays accurate. A monthly knowledge base review (30 minutes — review each section for accuracy and completeness) catches the drift that accumulates between updates. An outdated knowledge base produces outdated assistant responses — the discipline of maintenance is what makes the tool remain valuable over time.
Can multiple team members use the same business assistant?
Yes — build user authentication into the Bubble.io application and associate each conversation with the logged-in user. Each team member gets their own conversation history, and the system prompt can be customised based on role (the account manager’s assistant emphasises client relationship context; the finance person’s assistant emphasises financial data). The shared knowledge base ensures consistent information across the team; the personalised interface ensures each person interacts with the assistant in the way most relevant to their role.
Want a Custom Business AI Assistant Built?
SA Solutions builds personalised Bubble.io AI assistant applications — knowledge base design, assistant interface, system prompt engineering, and live system integrations.
