AI Tools that Handle Customer Enquiries within Business Rules

Build the Rules Before You Turn the Tool On

An AI enquiry tool only works within business rules if those rules are written down in a form the tool can actually use. Start with five categories: services you provide, locations you cover, hours you reply, information required before quoting, and issues that must be escalated to staff.

For example, a cleaning company might allow the assistant to confirm that it covers SW and SE London postcodes, explain that tenancy cleans need the number of bedrooms and bathrooms before a quote, and offer booking slots between 8am and 6pm. The same tool should refuse to promise an exact price for a hoarder clean, because that needs photos and staff review. Those boundaries turn vague automation into something dependable.

1. List the questions you are happy for the tool to answer without approval.
2. List the questions it may answer only after collecting missing details.
3. List the subjects it must not answer, such as legal, medical or bespoke pricing questions.
4. Define the words that trigger an urgent handover, such as complaint, refund, cancellation or safety issue.

Worked Example: Physio Clinic Reception Flow

A physiotherapy clinic receives the same first-contact questions every day: appointment availability, pricing for standard sessions, parking, insurance and whether the clinic treats a certain injury. A rule-based AI assistant can answer most of that quickly, but only if the rules are precise.

It can say that new patients need a forty-five-minute assessment, that evening slots usually book first, and that major insurers are accepted if the client brings an authorisation number. If the patient types chest pain, numbness, or a request for urgent diagnosis, the tool should stop short of advice and tell the person to seek appropriate medical help or contact the clinic directly. That protects the business and the customer.

The value comes from consistency. Every enquirer gets the same opening information, while the clinic team only handles the messages that need judgement. The result is faster replies without taking unnecessary risks.

Common Mistakes and a Quick Checklist

• Using broad phrases such as we normally reply quickly instead of setting a real response window.
• Allowing the tool to guess at bespoke pricing.
• Forgetting to update rules when opening hours, staff cover or service areas change.
• Not testing edge cases, such as out-of-hours complaints or requests from outside your service area.

Before launch, test ten real enquiries from your inbox. If the tool cannot answer them safely, your rule set is not ready yet. Also check that staff can see when the AI has handed something over and what information it already captured; otherwise the customer ends up repeating themselves.

First-Month Review Points

In the first month, review more than reply speed. Check whether the tool is collecting the right details before offering next steps, whether out-of-area messages are being closed cleanly, and whether staff are still rewriting a large share of replies. One useful exercise is to sample twenty conversations and mark each one as correct, incomplete or unsafe. The patterns usually reveal where rules are too loose. You may find that the tool needs a better price boundary, a stricter escalation trigger, or a clearer prompt for missing information such as postcode or service type. Those small corrections are what turn a rule-based assistant into something dependable over time.

A Simple Approval Rule

Many small businesses improve results by adding one plain approval rule: the AI may answer standard operational questions, but any reply that changes price, scope or liability must be reviewed by a person. That single distinction prevents a large share of avoidable mistakes.

Keep the Rules Versioned

Date each rules update so staff know which version is live. That stops old assumptions from creeping back into customer replies.