Outcome

A hybrid support model combining AI triage and human agents handled more than 170,000 chats between January and late March 2026. AI managed 48% of total volume, processing approximately 19.8 chats per active chatting hour against 10.1 for human agents. Human agents remained available for contacts requiring context, judgment, or a personal approach, including a dedicated faster routing path for identified VIP players.

Client Context

As support volumes grow, the question operators face is no longer whether to use AI or human agents. It is how to structure both effectively.

The operational challenge is that not every player wants the same support journey. Applying a single-path model, whether fully automated or fully human, creates either capacity constraints or friction depending on where volume lands.

KYZEN built a hybrid model designed to use each layer for what it does well, and to route contacts accordingly from the point of entry.

Challenges

Deploying AI in a support operation without a structured routing logic creates its own set of problems.

This created several constraints:

• In KYZEN’s experience across its operation, around 20% of players ask to speak to a human straight away, without wanting to go through an automated flow first. Forcing them through automation creates friction rather than efficiency.
• High-value VIP players typically have less tolerance for standard automated flows and their queries are more likely to require individual handling.
• Without a clear split between AI-appropriate contacts and human-required contacts, AI handles volume it is not suited for, and human agents absorb repetitive low-effort questions that do not require judgment.
• Repetitive, low-effort contacts occupying human agent capacity leave less availability for conversations requiring context, flexibility, or a less scripted approach.

This resulted in:

• A need for triage logic that could identify contact type at entry and route accordingly, rather than applying a single path to all incoming volume.
• A separate routing structure for VIP players to reach human support faster and with fewer steps.

The Approach

A hybrid model was built using AI as a first layer through triage and built-in journeys, with human agents handling the contacts where context, flexibility, or a personal approach mattered more.

Key elements included:

• AI Triage and Journey Layer: AI operated as the first layer, recognising intent, responding immediately to common questions, and directing conversations more efficiently. First response time through AI was close to 0 seconds.
• Human Agent Routing: Contacts requiring judgment, reassurance, or a less scripted approach were routed to human agents. Agents were relieved of repetitive low-effort queries, keeping their focus on the conversations that needed direct involvement.
• Direct Human Access: Players requesting human support immediately were not forced through an automated flow. The model accommodated that routing from entry.
• VIP Player Path: Identified VIP players were given a separate path to reach human support more quickly and with fewer steps, reflecting both their lower tolerance for standard automated flows and the higher likelihood their queries required individual handling.

Results

The hybrid model delivered measurable efficiency and volume capacity while preserving the human support layer where it mattered.

• More than 170,000 chats handled between January and late March 2026.
• AI handled approximately 48% of total chat volume. Human agents handled the remaining 52%.
• AI delivered an efficiency score of approximately 20 against 10 for human agents.
• AI processed approximately 19.8 chats per active chatting hour. Human agents handled approximately 10.1.
• Human agents were able to focus on contacts requiring judgment, context, or a personal approach rather than repetitive low-effort queries.
• VIP players reached human support faster through a dedicated routing path.

Operational Takeaway

The efficiency gap between AI and human agents in this model was not evidence that AI is better at support. It reflected the type of work each layer was given. AI handled structured, repetitive, lower-effort contacts. Human agents handled everything else. The split worked because the routing logic respected that distinction from the start.

The more important signal is directional. The line between what AI can handle and what requires a human is already moving. AI is increasingly able to pull information from back-office systems, use that context in real time, and handle more involved requests. What that boundary looks like today is not what it will look like in twelve months.

The operators who get the most from this model are the ones treating AI as a live operational layer that gets tested, updated, and expanded as capability develops, not as a static tool that gets implemented once and left alone.