Knowledge base · 34 entries

The glossary of AI for customer service

Terms, acronyms and concepts — explained the way someone who has to ship an AI assistant would use them, not how someone selling a slogan would. Each entry links to examples and products.

Curated by the helpcode teamUpdated monthly

A B C D E F G H I J K L M N O P Q R S T U V W X Y Z

A — agent · attribution · automation gap

Agent

AI system that executes multi-step actions, not just answers a question.

Attribution

Tracking the source (document, paragraph) of an AI answer.

Automation gap

The space between tickets AI can close and tickets that need a human.

C — chunking · context window · contextual retrieval

Chunking

Splitting documents into retrievable segments. A critical RAG strategy.

Context window

The maximum text a model can read at once. Measured in tokens.

Contextual retrieval

Adding document context to each chunk before embedding it.

D — data residency · deflection · dpa

Data residency

The physical location where customer data is stored and processed.

Deflection

The share of customer tickets resolved without a human agent.

DPA

Data Processing Agreement — the legal contract between controller and processor.

E — embedding · eu ai act · evaluation

Embedding

A numeric vector that represents the meaning of a text fragment.

EU AI Act

EU regulation classifying AI systems by risk and setting compliance duties per class.

Evaluation

Measuring an AI system’s answer quality on a curated set of questions before shipping.

F — fine-tuning · first-contact resolution

Fine-tuning

Further training a base model on a small task-specific dataset.

First-contact resolution

The share of issues solved on the first interaction — no callbacks, no escalations.

G — gdpr

GDPR

EU regulation on personal data — applies to every AI system handling EU residents.

H — hallucination · hybrid retrieval

Hallucination

A plausible answer that is not supported by the data. Mitigable with attribution.

Hybrid retrieval

Semantic search + keyword search, combined and reranked. The helpcode default.

I — indexing

Indexing

Building the data structures that make retrieval fast at query time.

K — knowledge base

Knowledge base

A curated, retrievable collection of authoritative documents for an AI assistant.

L — l1 tickets

L1 tickets

Tier-1 support: repetitive, scriptable, well-documented. The deflection sweet spot.

M — mcp server · multi-tenant

MCP server

Anthropic’s Model Context Protocol — exposes tools and data to AI models.

Multi-tenant

Architecture where customers share infrastructure but stay isolated at the data layer.

O — observability · ocr

Observability

The practice of recording prompts, retrieval, model calls and citations so a system is debuggable.

OCR

Optical Character Recognition — extracting machine-readable text from images and scans.

P — prompt injection

Prompt injection

A security attack where an attacker hides instructions inside data the model reads.

R — rag · reranker

RAG

Retrieval-Augmented Generation. The pattern under helpcode Knowledge Base and Troubleshooting.

Reranker

A secondary model that reorders retrieval results for relevance.

S — session handoff · sub-processor

Session handoff

Passing a conversation from an AI to a human agent with full context.

Sub-processor

A third party your processor uses to process your data.

T — token · tool use

Token

The unit of text language models consume and produce. Roughly 4 characters in English.

Tool use

A model calling external functions / APIs as part of its reasoning.

V — vector database · voice ai

Vector database

A database optimised for similarity search on high-dimensional vectors.

Voice AI

AI that handles speech-in / speech-out conversations — typing not required.

W — webhook

Webhook

An HTTP callback your system sends when an event happens.