How to reduce AI escalations

Every escalation is the AI saying 'I do not have what I need to handle this confidently.' Reducing escalations means giving the AI the information it needs to handle more conversations on its own. This is not about making the AI more aggressive, it is about making it more knowledgeable so it can genuinely help your guests without human intervention.

Start by analyzing your current escalations. Pull the last 30 days of escalated conversations and read through them. You will quickly see patterns: the same question asked in different ways, topics the AI consistently defers on, language the AI misunderstands. These patterns are your knowledge base backlog.

Step 1: Identify your top escalation reasons

In Rezi's Reports > AI Performance, filter for escalated conversations and look at the AI's stated reason for escalating each one. Group them into categories. Common top escalation reasons for new Rezi accounts: check-in and access questions (door codes, parking, wifi), availability and pricing questions, policy questions (pet policy, quiet hours, smoking), and context-specific requests (the guest wants to discuss something specific that has not been covered in the knowledge base).

Step 2: Add knowledge base articles for each top reason

For each escalation category, write a knowledge base entry in the property's AI knowledge base. An entry does not need to be long, 2-3 sentences that clearly answer the question the AI was failing on. For check-in questions: add the exact door code source, parking instructions, and wifi details. For pet policy: add the specific rule (whether pets are allowed, any size limit, any pet fee). Specificity matters more than length.

Step 3: Test the AI after adding each article

After adding a knowledge base article, use the Test AI feature to ask the question that was previously escalating. Confirm the AI now gives a correct, complete answer. If it still escalates or gives a partial answer, the article may need to be reworded or expanded. Test with the exact phrasing a guest would use, not the formal version you wrote.

Step 4: Review monthly

Escalation patterns change as your operations change. New policies, new guest demographics, seasonal issues (air conditioning in summer, heat in winter) create new escalation spikes. A monthly 20-minute review of escalation patterns and a corresponding knowledge base update keeps your AI performance improving continuously rather than plateauing.

What not to do

Do not try to lower escalation rates by making the AI's escalation threshold harder to trigger. An AI that escalates less because it was configured to be more confident, rather than because it is more knowledgeable, gives worse answers. The goal is a confident AI that is right, not one that is simply less cautious.