How to Build an AI Agent That Knows Your Business

Most AI tools sound impressive in a demo. You ask a question, you get an answer. But drop a generic chatbot into a real trade business and it falls apart fast. It doesn’t know your pricing. It doesn’t know which technician handles commercial jobs. It doesn’t know that Mrs. Patterson at 14 Queen Street always wants the same plumber.

The difference between a generic chatbot and a genuinely useful AI agent is business knowledge. An agent that knows your SOPs, your rate cards, your service boundaries, and your customer history doesn’t just answer questions – it makes decisions that match how you actually run your business.

Here’s how to build one.

Why knowledge matters

A generic AI model knows a lot about the world, but nothing about your business. Ask it to book a job and it doesn’t know your service area, your availability windows, or whether the customer is on a maintenance contract. It guesses. And guesses mean mistakes, frustrated customers, and more cleanup work for your office staff.

A trained AI agent is different. It operates with the same context your best office manager has – the pricing structure, the escalation rules, the repeat customers, the scheduling preferences. It doesn’t guess because it doesn’t have to. The knowledge is there.

What to teach your agent

SOPs and procedures

Your standard operating procedures are the backbone of consistent service. Teach your agent how you handle new enquiries, how you triage emergencies versus routine maintenance, how you follow up after a job is completed, and how you handle complaints. When the agent knows your procedures, it follows them every time – no shortcuts, no forgotten steps.

Pricing and rate cards

Load your agent with your current pricing structure: standard hourly rates, after-hours premiums, materials markups, fixed-price packages, and any special pricing for contract customers. When a customer asks for a ballpark quote, the agent gives a number that matches your actual rates – not a fabricated figure.

Service area and availability

Define where you operate and when. Your agent should know your geographic boundaries, your standard business hours, your after-hours availability, and any blackout periods. When a customer in a suburb you don’t service asks for a booking, the agent politely declines rather than creating a job your team can’t fulfil.

Customer history from simPRO

This is where things get powerful. By connecting to simPRO, your agent has access to complete customer records – past jobs, open quotes, maintenance schedules, site notes, and preferred technicians. When a repeat customer gets in touch, the agent already has context. It knows the property, the history, and the relationship.

Job templates and workflows

If you have standard job types – hot water replacement, switchboard upgrade, annual AC service – build them as templates your agent can use. Each template includes the default job description, estimated duration, required qualifications, standard materials, and pricing. The agent creates consistent, complete jobs every time without your staff typing out the same details.

How knowledge works in Sprigr

Sprigr structures business knowledge into three layers that work together.

Shared company indexes

These are the knowledge bases that every agent in your organisation can access – SOPs, pricing, service area definitions, job templates, and company policies. Update a rate card once and every agent across your team uses the new pricing immediately. No retraining, no version confusion.

Private agent memory

Each agent also builds its own working memory from the interactions it handles. It remembers that a particular customer prefers SMS over email, or that a specific site requires a council permit before work begins. This memory is unique to each agent instance and grows more useful over time.

Integration sync

Live data from your connected systems – simPRO, Xero, Gmail – flows into the agent’s context automatically. When a technician’s schedule changes in simPRO, the agent knows. When an invoice is paid in Xero, the agent knows. The agent always works with current data, not a stale snapshot.

Persona configuration

Knowledge alone isn’t enough. Your agent also needs to communicate like your business.

Communication style

Set the tone – professional but friendly, formal for commercial clients, casual for residential. Define how the agent greets customers, how it signs off, and what language it avoids. If your business has a particular way of talking to customers, the agent should match it.

Industry terminology

Trade businesses have their own vocabulary. Your agent should understand and use the correct terms – “hot water unit” not “water heater,” “switchboard” not “breaker panel.” This builds trust with customers and avoids confusion with your technicians.

Escalation rules

Not everything should be handled autonomously. Define clear escalation paths: burst pipes go straight to the on-call manager, jobs over a certain dollar value need human approval, complaints are flagged for the business owner. The agent follows these rules without exception.

Guardrails: What your agent can and can’t do

A well-configured agent has clear boundaries. This is not a limitation – it’s what makes the agent trustworthy.

• Tool access policies. Control which systems the agent can read from and write to. Maybe it can create jobs in simPRO but can’t delete them. Maybe it can generate quotes but can’t send invoices without approval.
• Content restrictions. Define topics the agent should never engage with. It shouldn’t give legal advice, make guarantees about timelines, or discuss competitor pricing.
• Spending limits. If the agent can approve work or commit resources, set dollar thresholds. Anything above the limit gets escalated to a human.
• Scope boundaries. The agent handles what you tell it to handle. Everything else goes to your team. Start narrow and expand as you build confidence.

The feedback loop

Your agent gets better over time – but only if you close the loop.

When the agent makes a mistake, correct it. If it quotes the wrong rate, update the pricing index. If it schedules a job outside your service area, tighten the boundary definitions. If it misunderstands a customer request, refine the relevant SOP.

Every correction improves the knowledge base. The agent doesn’t repeat the same mistake because the underlying knowledge has been fixed. Over weeks and months, your agent becomes more accurate, more capable, and more trusted by your team.

Getting started

You don’t need to teach your agent everything on day one. Start with one department or one task – maybe handling after-hours enquiries or triaging incoming emails. Load the relevant SOPs, connect simPRO for customer context, set your guardrails, and let the agent run.

Watch the audit trail. See where the agent gets it right and where it needs more knowledge. Add to the knowledge base, tighten the rules, and expand to the next task. Within a few weeks, you’ll have an agent that genuinely understands your business – not a generic chatbot, but a tool that works the way your team does.