TUGI's AI Agent

TUGI is our proprietary tool that fine-tunes frontier models to perform specific tasks with precision. 

Instead of relying on models like ChatGPT or Gemini out of the box, TUGI adjusts their behavior to meet our needs and quantifies performance to ensure alignment. It combines the power of cutting-edge language models with our custom machine learning approach. This hybrid setup allows us to leverage the latest AI advancements while tailoring models to our own metrics. 

TUGI starts with a domain-specific goal, defines rules, and breaks down the task into manageable steps. It then uses examples to guide the model’s self-evaluation and iterative improvement. The model adapts its reasoning to the task requirements through this process. 

Our AI agent is specifically trained for the iGaming industry, making it knowledgeable about the sector’s terminology, workflows, and customer interactions. This ensures responses are not only accurate but also contextually relevant to the unique needs of gaming operators and players.

The AI Agent can handle a wide range of queries. Its responses are based on several data sources:

• iGaming Expertise
  • The AI Agent is trained with comprehensive iGaming knowledge and terminology, allowing it to answer general industry-related questions.
• Brand-specific Policies
  • It uses the brand's policy documents, workflows, and instructions stored in TUGI to provide answers tailored to your brand. These need to be added and written in text format in TUGI > AI Settings.
• JWT Data from the Back Office
  • The AI Agent can read real-time values from the back office via JWT, such as:
    • transaction history
    • game rounds
    • player segment
    • active or inactive bonuses
    • account limits etc.

The result is a fast, efficient, and cost-effective solution for solving complex, real-world problems.

You must be an Admin in order to configure AI Agent details. AI Settings can be configured by Configura > AI Settings > select the brand. 

Activating the AI Agent

Before using the AI Agent, it must be activated on the brand level:

• Go to Configuration
• Go to AI Settings and select the brand.
• Locate the AI Settings section.
  • AI Agent chat channel: Enables / disables the AI Agent in the chat channel
  • AI Agent email channel: Enables / disables the AI Agent in the email channel
• Once activated, the AI Agent can start assisting with conversations according to your defined rules and preferences.

Configuring the AI Agent

• AI Service Prioritization:
  • Route tickets directly to human agent: ON/OFF
  • This setting takes into consideration if there are human agents online in the Agent Workplace. If there are, the tickets are routed directly to human agents first. If there aren't any human agents online, the AI Agent answers all incoming tickets.
    • Note: AI Agent needs to be enabled for this feature to work.
• AI Agent Settings: Set AI Agent's alias name by language, which is shown to the players
  • By default, the AI Agent is called “AI Agent”, but feel free call yours whatever!
  • You can configure the AI Agent's name based on languages.
  • To remove languages, click on the trashcan picture on the right side of the language

• AI Knowledge Hub: The policies and information the AI Agent bases their answers on. Read more in next tab, “Adding AI policies”.

Setting AI Agent Profile

AI Agent profile can be configured from Configuration > Users > AI Agent. 

You can recognise the AI Agent from the users list by user name “AI Agent”. AI Agent's email address is your brand id and ending with tugi.ai.

1. User roles
   1. Set user role: Agent
2. User languages
   1. Set User languages to the languages, that you wish the AI Agent serves the players with
      1. Set by pressing  “Set languages”
   2. If your brand has any prohibited languages, do not set them as a language the AI Agent serves in.
3. Agent departments
   1. Set your AI Agent in the default department of your brand.
      1. Toggle: “AI Default Dept.” 
      2. Toggle “Is Assigned” box to active.
      3. Set by pressing “Assign departments”
4. Expiration date
   1. If you wish to deactivate the AI Agent on a certain date, choose an expiration date from the calendar.

Setting AI Policies

After activation, you can configure AI Agent policies specific to the brand in Configure > Brands. These policies help guide how the AI responds to customers—what it can say, when it should escalate to a human, and how it should sound. Each policy can contain up to 1 040 000 characters. You can add as many policies as you wish. Every time you update the policy, it's updated in real-time to the AI Agent.

Well-written policies help the AI:

• Give correct answers 24/7
• Match your brand tone
• Know when to ask for help from a human agent

For example, you can define:

• Terms & conditions
• Bonus terms & conditions
• Privacy policy
• Refund policy
• Responsible Gambling 
• Campaigns
• Knowledge base

Each brand has its own unique AI Agent configuration, ensuring that the AI behavior aligns with your brand identity and operational needs.

How to write AI Agent Policies?

You can add new policies and edit existing ones in Configura > AI Settings > “AI knowledge hub”.

Here are the most common categories, with examples:

AI Agent can help with self-service

If you include links to your most common scenarios, the AI Agent will automatically recommend players also links. 

Example in a policy:

How to cancel withdrawal: Players can cancel a withdrawal by going to menu > transactions and clicking cancel, if the withdrawal hasn't been approved yet or via link: http://tugicasino.com/transactions/

 

 

 

Our AI Agent assists by analyzing player messages and generating responses based on available context and policies.
To improve transparency, the AI Agent’s reasoning is visible internally on each ticket. This allows you to understand why the AI responded the way it did and what information it used when forming its response.

Where to Find AI Agent Reasoning

Ticket Inbox 

• Go to Ticket Inbox
• Open a ticket
• Navigate to AI Events
• Review the AI Agent reasoning shown for each AI-generated message

Agent Workplace

• Go to Agent Workplace
• Open a live ticket
• Select AI Events
• View AI Agent reasoning as the conversation progresses in real time

AI Agent response resources

The AI Agent does not respond randomly. Its reasoning is based on structured information available to it at the time of the message, including:

iGaming Knowledge (Tugi's AI)

The AI Agent bases its responses on our trained, underlying AI model.
This model includes pre-existing, general iGaming domain knowledge (such as common terminology and industry concepts). As a result, it understands concepts like slots, RTP, and sports betting without requiring these to be explicitly defined in AI policies.

AI Policies (Tugi)

The AI Agent references AI Policies configured in Tugi, which define:

• Policies
• Procedures
• These policies ensure the AI behaves consistently with your organization’s rules and support standards.

Player Profile Data (JWT)

When available, the AI Agent will also read player-specific information provided through JWT, such as:

• Account or profile attributes
• Status flags (e.g. verification, restrictions, eligibility)
• Relevant historical or contextual data
• This allows the AI to tailor responses to the individual player’s situation rather than providing generic answers.

The AI only uses the data that is made available to it and does not access information outside of configured policies or provided JWT context.

How to Read AI Agent Reasoning

AI Agent reasoning explains how the AI interpreted the message and which information influenced its response.

You may see reasoning that references:

• A specific AI policy being applied
• Player profile details used to personalize the response
• The intent or issue detected in the customer’s message
• Why a certain action, response, or escalation was chosen

Example

Customer message: “Why can’t I withdraw my winnings?”

AI reasoning:

• Identified withdrawal-related inquiry
• Checked applicable AI withdrawal policy
• Reviewed player profile status from JWT, if available
• Determined restriction applies and provided compliant explanation
• This shows why the AI gave that response, based on both policy and player context.

Reasoning is Internal Only

AI Agent reasoning appears only in Internal Notes and is never visible to players.

When to Review AI Agent Reasoning

Review AI reasoning to:

• Understand why the AI responded a certain way
• Confirm correct policy application
• Verify player data was interpreted correctly
• Investigate unexpected or incorrect responses
• Improve AI Policies or workflows in Tugi
• Train agents on AI-assisted decision-making

The AI Agent that serves players in chat and email is the same agent, and it follows the same AI policies.

If you have different procedures for chat and email interactions, make sure these are clearly defined in the policy documents.

However, the AI Agent’s conversation handling differs depending on the contact channel.

AI Agent flow in chat conversations

When a player opens a chat conversation:

Option 1: AI Agent resolves the issue

• The AI Agent greets the player.
• The AI Agent interacts with and advises the player.
• The AI Agent asks if there’s anything else they can help with (after 3 minutes).
• The AI Agent marks the conversation as Solved (after 15 minutes).
• A CSAT questionnaire is triggered.

Option 2: AI Agent transfers to a human agent

• The AI Agent greets the player.
• The AI Agent interacts with and advises the player.
• The AI Agent transfers the conversation to a human agent.
• The ticket moves to Unattended status in the human agent queue.

AI Agent flow in email conversations

When a player sends an email:

• The AI Agent replies with a greeting and either:
  • Answers the question directly (if it’s clear), or
  • Requests more information or clarification.
  • Asks how they can help, if the message is unclear.
• If the last message is not from the player, the AI Agent waits 4 hours for a response.
  • If the player replies within 4 hours, the AI Agent continues the conversation.
  • If no response arrives, the ticket moves to On-Hold status.
• If the matter is resolved, the AI Agent marks the ticket as Solved.
• If a player replies within 96 hours, the ticket returns to the Unattended queue.
• If there’s no player reply within 96 hours, the ticket automatically moves to Solved status.