Welcome to our dictionary - it contains the glossary of terms used within the Unless system and dashboard. This article provides clear definitions for the key concepts, features, and entities you'll encounter as an AI Manager to help you navigate, configure, and optimize your AI-powered customer experiences.
Glossary
Account
A unique customer organization profile in Unless.com used to configure AI interactions, customer experiences, manage users, and oversee AI-driven interactions across channels.
Action button
An interactive UI element in an Unless component that initiates a configured task or flow.
Agent
An AI-driven software entity within Unless that automates customer interaction, provides assistance, and executes tasks.
Agent actions
Operations performed by an Unless AI agent on behalf of an end user, such as sending messages, retrieving data, or triggering workflows.
Answer
Content, guidance, or solutions provided by the Unless AI assistant in response to a question or query.
Audiences
Customer segments defined in Unless based on behavioral, demographic, or engagement data for personalization, access and authorization.
Allowed language
Languages enabled for AI interaction within an Unless account configuration. By default, it's all languages, but allowed languages limit these to the possibly limited set that your organization can handle.
Base language
The primary reference language for AI training data in an Unless account - the language in which most of your data is written.
Browser extension
The technology that is used to distribute the Team Assistant into the browsers of its audience.
Close
An action to dismiss, terminate, or exit a module, content panel, or interaction.
Compliance controls
Built-in Unless safeguards ensuring adherence to privacy laws, data policies, and regulatory standards.
Component
An interface module that can be deployed into any customer's user interface, sometimes containing AI features. Functionality varies from onboarding to sales or customer support. Components may be launched by the AI if it sees fit.
Consent management
A process in Unless for obtaining, storing, and managing user permissions for data handling.
Control answer
A predefined response used as a benchmark for validating AI accuracy.
Control group
A baseline audience segment in Unless used for A/B testing and performance comparison.
Control question
A specific question used to verify the accuracy and consistency of AI-generated responses, associated with an independent control answer.
Conversation
A structured exchange between an end user and the Unless AI, tracked for analytics and improvement.
Cookie whitelist
An authorized set of custom cookies that will be passed on to the Unless environment for the creation of audiences.
CRM connector
An Unless integration that synchronizes customer relationship management data to enhance user profiles that may be used for audience segmentation and personalization of AI responses.
Custom rule
A user-defined condition within Unless that dictates AI behavior.
Dashboard
The interface in Unless.com that visualizes live customer engagement metrics, AI performance data, and personalization outcomes for real-time monitoring and optimization, and where the AI Managers can manage your AI and its behaviour.
Data access request
A formal request tracked in Unless for a user to view their stored personal data.
Data correction request
A process in Unless to update or amend incorrect personal data records.
Data erasure request
A process in Unless to delete all personal data for a requesting user in compliance with laws.
Data management
Dashboard section for the organization and handling of training data, personal information, and agent actions within Unless.
Downvote
A user action signaling dissatisfaction or inaccuracy in an AI-generated response.
DPO
Data Protection Officer - the role assigned in Unless to oversee data protection strategy and compliance.
Draft version
An unpublished iteration of content, rules, or configurations still under review in Unless.
End user
A visitor who directly interacts with Unless-powered AI experiences, benefiting from personalized content, automated support, or tailored workflows.
Escalation
Routing a customer inquiry from AI handling to human support when higher expertise or intervention is required.
Fallback language
The default language used by Unless when a preferred language is unavailable.
Follow-up question
A supplementary inquiry prompted by the Unless AI to clarify or expand on a previous answer.
Generated answer
An AI-created response from Unless based on training data, context, and live inputs.
Indexed item
A content, document, or entry cataloged by Unless’s AI search index for fast, relevant retrieval.
Insights
Dashboard section for actionable findings generated by Unless analytics, highlighting trends, performance metrics, and optimization opportunities for customer success.
Integration
The process of connecting Unless to other systems (CRM, CMS, or MCP servers, etc.) to unify data flows and automate actions.
Limited Agency
LLM-based systems with excessive permissions may take unintended actions due to inadequate user oversight. To prevent this, UNLESS enforces user approval for new action categories.
Limited reliance on LLMs
Dependence on LLMs for critical decisions can lead to misinformation or insecure outputs without human review. To prevent this, UNLESS enforces a disclaimer in all AI assistants, and limits functional purpose of LLMs.
MCP server
Model Context Protocol servers are like a bridge between the AI and any product API. The AI can leverage an MCP server by offering the "MCP tools" as Agent Actions that the end user can choose to execute. Basically, it helps the AI to execute actions on behalf of the end user.
Model Denial of Service (DoS)
Attacks could cause resource-heavy LLM operations, degrading service quality or increasing operational costs. To prevent this, UNLESS limits the context window, and enforces API rate limits.
Optimizer
An Unless tool or algorithm that refines content using gap detection, deduplication, and rewriting.
Persona
A predefined profile in Unless representing a specific character of the AI, including tone of voice.
PII
Data such as names, emails, or identifiers that can uniquely identify a person, handled securely within Unless.
PII filtering
The detection and removal of identifiable personal data by Unless systems to meet compliance.
PII filtering whitelist
An approved list of data points exempt from filtering within Unless for specific operational needs.
PII obfuscation
Masking identifiable personal data in Unless to protect user privacy during processing.
Primary click
A user's main action click recorded within an Unless component, typically on key interface elements.
Privacy
The protection and respectful handling of user data within Unless’s operations.
Prompt injection
Attackers may manipulate the LLM through crafted inputs, potentially leading to data exfiltration, social engineering, or backend system exposure. To prevent this, UNLESS enforces privilege control, validates user input, and cleanses untrusted user input.
Published version
The final live iteration of content or configuration visible in Unless-powered experiences.
Quality assurance
Dashboard section to manage the process in Unless of reviewing and validating AI outputs and training levels.
Quality report
A generated Unless report summarizing quality checks, AI performance, and data accuracy.
Question
An inquiry submitted by a user or end customer to obtain information through Unless's AI assistant or search interface.
Query
A structured request for information sent to Unless's AI - may be equivalent to "question".
Report score
A numerical rating within Unless indicating the performance or quality of generated reports.
Resource
Any asset—document, article, dataset, or media—stored or linked within Unless for retrieval and user engagement.
Resource collection
A curated set of related resources within Unless, optimized for retrieval and contextual display.
Response
A single unit of output provided by the Unless AI—such as AI-generated answers, recommendations, or content—triggered by a user action or query.
Search
A user query that results in a list of relevant sources to be displayed.
Secondary click
An alternative user click recorded within an Unless component, often for additional options or context menus.
Secure output handling
Insecure output handling may occur when the outputs generated by an LLM are passed downstream without proper validation, leading to vulnerabilities like XSS or privilege escalation. To prevent this, UNLESS enforces strict validation for LLM responses, and applies LLM output sanitization.
Secure Plugin Design
LLM plugins can introduce vulnerabilities if they lack input validation or control over user actions, risking remote code execution. To prevent this, UNLESS enforces parameterized inputs, and applies plugin authentication.
Security
Protective measures in Unless guarding platform systems, integrations, and customer data against unauthorized use or threats.
Sensitive information protection
LLMs might inadvertently reveal confidential information, exposing sensitive data. To prevent this, UNLESS applies data sanitization filters, and limits external data access for the LLM.
Session
A tracked, continuous period of end-user interaction.
Source
The origin of a resource, dataset or piece of content ingested or referenced by the Unless platform.
Source connector
An integration module in Unless that retrieves data from a specified external system to be used for AI responses.
Suggested control question
A recommended accuracy-check question proposed by the Unless system to ensure proper AI calibration.
Supply Chain Security
Risks may be arising from third-party model or dataset dependencies, which may introduce outdated or compromised components. To prevent this, UNLESS uses a Content Security Policy (CSP).
Targeting condition
A rule in Unless specifying criteria for delivering tailored experiences to selected audiences.
Task
A discrete action or workflow proposed by the system to the AI Manager, such as improving a document based on negative feedback.
Team assistant
An AI-powered tool in Unless designed to coordinate tasks, share insights, and aid collaborative workflows of a customer support team or other employees of an organization.
Test mode
The type of A/B testing tactic that you can apply to a component.
Topic
A thematic category or subject label in Unless used to organize content, conversations, or AI knowledge. Topics are used to prevent hallucinations and AI mistakes.
Tracking
The monitoring of user interactions and behavior of end users across browser sessions.
Tracking whitelist
A curated list of approved pages where tracking is allowed or necessary.
Training data
Curated datasets used to define the scope of what Unless’s AI models can use to respond to queries.
Training data poisoning
Malicious actors may insert harmful or biased data into LLM training, which affects model behavior and output quality. To prevent this, UNLESS uses anomaly detection techniques.
Translation corpus
A stored dataset in Unless used to train translation functions for multilingual experiences, primarily used for terminology that refers to the customer's context, like product names and such that shouldn't be translated across languages.
Transparency
The clear and open disclosure of processes in Unless related to AI decision-making, personalization, and data use.
Upvote
A user action indicating approval or usefulness of an AI-generated response.
Use case
A defined scenario where Unless's AI capabilities are applied to achieve a specific business objective. Currently, a use case is a combination of a domain (like a website, or a product), and a target audience (either anonymous or identified).
User
The AI Manager - an individual with authenticated access to Unless, with defined roles and permissions to manage content, configurations, or analytics.
User experience
Dashboard section for configuration options to manage overall quality, ease of use, and satisfaction perceived by customers when interacting with Unless-powered solutions.
Variables
Data points that an AI agent needs to fulfill an end user's request.
View
The event when a component is fired in an interface, referring to having been seen by an end user.
