Glossary
How to use this glossary
- Terms are grouped by category.
- Categories are ordered alphabetically.
- Within each category, terms are ordered alphabetically.
Analytics & Reporting
A/B test — A comparison experiment (e.g., prompt version A vs B) to measure which performs better on outcomes like automation rate or CSAT.
Analytics — The metrics and reporting layer used to measure how well DG automations perform (quality, volume, efficiency, business impact).
Attribution — The method for deciding which flow/change caused an outcome (e.g., CSAT improvement after a release).
Automation rate — The % of conversations resolved end-to-end by DG without human takeover.
Containment rate — Similar to automation rate; the share of conversations contained without agent handoff.
Coverage — The portion of customer questions/intents that DG can handle (often measured by intent set completeness + knowledge completeness).
Dashboard — A saved analytics view that tracks key metrics (e.g., automation, deflection, CSAT, AHT).
Deflection — The reduction in human-handled contacts because customers are helped by automation/self-serve.
Failure reason — A labeled explanation for why automation did not complete successfully (e.g., missing data, low confidence, unavailable integration).
Handover rate — The % of conversations that DG routes to a human agent.
Intent accuracy — How often DG assigns the correct intent to a customer message (often tracked via audits/labels).
KPI (Key Performance Indicator) — A metric used to judge success (e.g., CSAT, ROI, automation rate).
Latency — How long an automated step takes (e.g., module action response time) which impacts customer experience.
Resolution rate — The % of conversations marked as solved (by DG or by a human), often combined with containment for quality.
SLA (Service Level Agreement) — A target/contracted service metric (e.g., deliver within 5 business days).
Time series — A trend chart of a metric over time (daily/weekly).
AI & LLM Concepts
AI agent — An AI system that can plan and take actions (e.g., call tools/flows) to complete tasks, not just generate text.
AI-generated response — A reply produced by a model (often guided by a prompt), potentially using retrieved context and variables.
Confidence score — A numeric estimate of how certain the model/classifier is (commonly used for intent detection).
Context window — The amount of conversation/knowledge the model can “see” at once when generating a response.
Embedding — A vector representation of text used for semantic search/retrieval (finding similar meaning, not exact keyword matches).
Few-shot example — A prompt pattern where you provide a few example inputs/outputs to steer the model’s behavior.
Guardrails — Rules/instructions that constrain outputs (e.g., no promises, no policy violations, escalate if uncertain).
Hallucination — When a model produces information that sounds plausible but is not grounded in provided data.
LLM (Large Language Model) — A model trained on large text data that can generate and transform language.
Model — The AI component that produces predictions or text (e.g., an intent classifier or an LLM).
Prompt — Instructions given to an LLM to control output style, structure, and content.
Prompt injection — A malicious or accidental instruction inside user text that tries to override system rules.
System prompt — High-priority instructions that define how the AI should behave across many situations.
Temperature — A generation setting that controls randomness/creativity; higher = more varied, lower = more deterministic.
Conversation & CX (Customer Experience)
AHT (Average Handle Time) — Average time a human agent spends handling a conversation/ticket.
CSAT (Customer Satisfaction Score) — A customer satisfaction metric, typically collected via a post-interaction survey.
Escalation — Moving a conversation from automation to a human agent due to uncertainty, policy, or customer request.
Handoff / Handover — The process of transferring a conversation from DG to a human agent.
Handover reason — The stated trigger for escalation (e.g., low confidence, “talk to a person”, missing information).
Resolution — A conversation outcome indicating the customer issue is solved.
Sentiment — A classification of the customer’s tone/emotion (e.g., positive/neutral/negative).
Tone of voice — The desired writing style/brand voice DG should use in replies.
Triage — Quick classification and routing of an incoming request (often via intent + rules).
Flow Building & Orchestration
Activity — A single step within a flow (e.g., decision, action call, knowledge retrieval, response).
Action — A type of activity that calls a module operation to read/write external data.
Branch — A path in a flow that runs when conditions are met (if/else logic).
Condition — A rule check used in a decision (e.g., “order_found = true”).
Decision — An activity that evaluates conditions and chooses the next branch.
Entry point — Where a flow starts (e.g., triggered by a specific intent, channel, or event).
Error handling — Steps that run when something fails (fallback messaging, retries, handover).
Fallback — A safe response or alternate path when automation cannot proceed confidently.
Flow — The automation blueprint that defines how DG handles a conversation scenario end-to-end.
Flow version — A saved iteration of a flow, useful for releases and rollbacks.
Loop — A flow pattern where a set of steps repeats until a condition is met.
Orchestration — Coordinating multiple activities, decisions, and actions to complete a task.
Retry — Re-attempting an action after a transient failure (timeouts, temporary API issues).
Routing — Selecting which flow or branch should run (often based on intent, conditions, or channel).
Sub-flow — A reusable flow called from another flow to avoid duplicating logic.
Timeout — When an external call takes too long and the step fails.
Intent & NLU (Natural Language Understanding)
Entity — A structured piece of information extracted from text (e.g., order number, email address, product name).
Intent — The customer’s goal/purpose behind a message (e.g., “Track order”, “Return item”).
Intent detection — The process of classifying a message into an intent (often with a confidence score).
Intent set — The list/library of intents configured for a tenant.
Out-of-scope — A request DG is not configured to handle with existing intents/flows.
Utterance — An example customer message used to train/test intent detection.
Knowledge & Content
Article — A unit of knowledge content (FAQ, policy, troubleshooting guide).
Chunking — Splitting a long document into smaller parts to improve retrieval quality.
Citation / source — A reference to the knowledge used to answer (helps trust and auditing, even if not shown to the customer).
FAQ (Frequently Asked Questions) — A common Q&A style knowledge format.
FAQ retrieval — Finding the most relevant FAQ/article(s) for a question (often via semantic search).
Knowledge Base (KB) — The repository of factual content DG can use to answer questions.
Knowledge gap — A missing or outdated piece of information that prevents accurate automation.
Knowledge retrieval — Pulling relevant KB content into the context used for a response.
Relevance score — A measure of how well a knowledge item matches a query.
Semantic search — Search by meaning (embeddings) rather than exact keywords.
Synonym — Alternate phrasing that should map to the same concept/intent/article.
Taxonomy — A structured classification system for knowledge (categories/tags).
Modules & Integrations
API (Application Programming Interface) — The interface DG uses to request data/actions from another system.
Authentication — Proving identity to an external system (API keys, OAuth tokens, etc.).
Connector — A pre-built integration to a third-party tool (often synonymous with module).
Credentials — Secrets used for authentication (tokens/keys); should be stored securely.
Integration — A configured connection between DG and an external system.
Module — DG’s integration with an external system, exposing actions that can be used in flows.
Rate limit — A cap on how many API calls can be made within a time window.
Webhook — An event callback from another system to DG (system pushes data to you).
Operations, Quality & Troubleshooting
Audit — A review of conversations/decisions to assess correctness (intent labels, answer quality).
Bug — An incorrect behavior in flow logic, configuration, or integrations.
Edge case — A rare scenario that breaks the “typical” path (missing fields, unusual phrasing).
Incident — A production issue impacting customers (outages, integration failures).
Monitoring — Observing system health and performance (errors, latency, volume).
Regression — A previously working behavior that breaks after a change.
Root cause — The underlying reason a failure occurred.
Runbook — A step-by-step troubleshooting guide for common operational problems.
Testing — Validating changes before rollout (unit tests, staging, sample conversations).
Personalization & Variables
Dynamic content — Response text that changes based on data (order status, name, delivery date).
Placeholder — A slot in a template that gets filled with a variable value.
Personalization — Tailoring responses using customer/order context.
Variable — A named value passed between steps (e.g., order_id, tracking_url).
Templates & Responses
Chat activity — A response activity that uses AI prompting to generate a message (more flexible than templates).
Completion — The generated model output for a prompt.
Message template — A structured response with placeholders for variables.
Response step — The activity that produces the customer-facing message.
Structured response — A response constrained to a format (bullets, fields, JSON, etc.) to keep outputs consistent.
Template — Pre-written response text with placeholders used for consistent, on-brand replies.
Value & Business Impact
Cost savings — Reduced support cost due to automation/deflection.
Efficiency — Doing the same work with less time/effort (often improved by automation).
ROI (Return on Investment) — The business return from DG vs the cost of implementing/running it.
Payback period — How long it takes for benefits (savings or revenue) to cover investment.
Revenue impact — Increased sales or reduced churn driven by better support outcomes.
Updated 6 months ago