Policies Overview

Policies are business rules that tell the AI how to handle specific situations. While the knowledge base provides information for answering questions, policies provide instructions for behaviour — what the AI should do, must avoid, or when it should escalate to a human.

How the AI uses policies

Before the AI drafts a response, it checks for applicable policies based on the conversation topic and context. If a matching policy exists, the AI follows its instructions. This means you can enforce business rules consistently across every customer interaction without relying on individual agent memory or training.

Policy structure

Each policy has the following parts:

• Topic — The category this rule applies to (e.g. “returns”, “pricing”, “shipping”)
• Rule text — The specific instruction the AI must follow
• Conditions (optional) — Criteria that determine when the policy applies

Examples of policies

Here are some real-world examples to illustrate what policies look like:

Topic	Rule
Returns	Return requests with a value above EUR 2,000 require human approval — escalate to a human agent
Returns	Always require an RMA number before processing a return
Pricing	Never disclose wholesale pricing to end customers
Shipping	For orders over 50 kg, inform the customer that delivery requires a loading dock or forklift
Support	If the customer mentions legal action, immediately escalate to a human agent
Credit	Do not offer credit extensions without manager approval

Conditions

Policies can include conditions that control when they apply. A condition narrows the scope of a rule so it only triggers in certain situations.

For example, the rule “Escalate return requests to a human agent” might have a condition “Order value exceeds EUR 2,000”. Without the condition, every return request would be escalated. With it, only high-value returns get flagged.

The Policies list view

The Policies page displays all your rules in a list. Two tabs are available at the top:

• All rules — Shows every policy regardless of status
• Pending review — Shows policies proposed by the learning loop that need your approval

Policies vs. knowledge base

A common question is when to use a policy versus a knowledge base entry. The distinction is straightforward:

• Knowledge base — Answers the question “What information should the AI know?”
• Policies — Answers the question “What rules should the AI follow?”

If a customer asks “What is your return window?” — that is a knowledge base entry. If you want the AI to always escalate returns above a certain value — that is a policy.