Glossary
Large language model (LLM)
A large language model (LLM) is a neural network trained on a huge volume of text. It understands a question written or spoken in everyday words and writes an answer in plain language. In a contact center, language models power voice and chat AI agents, the agent-facing AI assistant, and automatic conversation summaries.
Why it mattersIt powers AI agents and agent suggestions – they handle conversations that used to require a person.
Also called: LLM, Large language model, Language model, Generative AI
How it works
How it works
The model takes a prompt and, step by step, picks the most likely next words of its answer. What it says depends on what goes into the prompt. That's why, in a contact center, the model gets not just the customer's question but everything it needs to answer correctly.
- Instructions: role, tone, and what it may and may not say
- Conversation history and customer data
- Knowledge base articles relevant to the question – found by RAG
- Tools: API requests to your systems – check an order, create a ticket, book an appointment
Use cases
Where it's used in a contact center
- AI agents – voice and chat
- Suggestions and draft replies for agents
- Call and chat summaries, inquiry topic tagging
- Quality review of every conversation and sentiment analysis
Limitations
Limitations and how to handle them
- A model can confidently make up a price or a policy – have it answer only from the approved knowledge base and say "I don't know" when the answer isn't there
- Test answer quality in your customers' language on your real questions, not on demo examples
- Decide up front which customer data can go to an external service and which must stay with you
- The bigger the model, the more it costs and the slower it answers – match the model to the task
- Log the answers and spot-check them regularly
In CONTACT.WORK
How it works in CONTACT.WORK
Agents and the assistant connect to the model through the AI connector, so you can switch models without rebuilding the agents. The model can run in the cloud or on your own hardware. The same model writes short conversation summaries, which appear in reports.
On your own servers, a GPU or a Mac mini is needed for both speech recognition and synthesis and for the language model. For up to 10 agents, the GPU goes into a shared server; from 11 agents, into a dedicated one – but never into the telephony server. The model and speech services are billed separately unless they run on your hardware.
More
Where to use it
Backlinks
Where it appears on the site 29
Solutions 5
AI agents 3
Services 3
Integrations 11
Pricing 1
Comparisons 2
Glossary