Glossary

Vendors in this market use these words as though everyone agrees what they mean. Several of them carry real weight in a sales conversation, and two of them are routinely used to make a number look better than it is. This glossary defines each one plainly, says why it matters to a store, and points out where the common usage is misleading.

  • Retrieval-augmented generation (RAG)

    Retrieval-augmented generation, usually shortened to RAG, is a way of making a language model answer from specific documents instead of from whatever it absorbed during training. When a question arrives, the system first retrieves the passages most likely to contain the answer, then asks the model to generate a reply using only those passages.

  • Chunking

    Chunking is the step in a retrieval-augmented system where documents are split into passages, or chunks, small enough to be retrieved individually. It happens once, when content is added, and it quietly sets the ceiling on how good every later answer can be.

  • Embedding

    An embedding is a list of numbers that represents the meaning of a piece of text. Passages that mean similar things get similar lists of numbers, even when they share no words at all. That is what lets an AI assistant search by meaning instead of by matching words.

  • Similarity threshold

    The similarity threshold is the minimum score a retrieved passage must reach before an AI assistant is allowed to use it in an answer. Everything above the line can be used; everything below it is dropped. It is the single setting that decides whether an assistant answers a question or declines it.

  • Deflection rate

    Deflection rate is the share of conversations that an automated assistant handled without passing them to a person. If 1,000 people start a chat and 700 never reach a human, the deflection rate is 70 per cent. It is one of the most quoted numbers in customer service software, and one of the easiest to flatter.

  • Resolution rate

    Resolution rate is meant to be the share of conversations in which the customer's problem was actually solved. It sounds like the right thing to measure, and it is, which is why it matters so much how each vendor defines "solved". Two vendors quoting the same percentage can be measuring completely different things.

The words that describe how an answer is made

  • Retrieval-augmented generation (RAG) is answering from documents fetched at the moment of the question, rather than from what a model memorised in training.
  • Chunking is splitting documents into passages small enough to retrieve one at a time, and it quietly decides how good the answers can be.
  • An embedding is a list of numbers that represents the meaning of a passage, so a search can find text by what it means rather than by the words it uses.
  • The similarity threshold is the score a passage must reach before the assistant is allowed to use it. It decides whether the assistant answers or declines.

The words that describe how well it is doing

  • Deflection rate counts the conversations that did not reach a person. It says nothing about whether the shopper got what they needed, which is why it flatters.
  • Resolution rate is meant to count conversations where the shopper's problem was solved. How a vendor measures “solved” matters more than the percentage.

The words that describe what a shopper sees

  • Grounded answer is an answer built from passages in your own content that cleared the threshold. ChatWidget marks these with “Sourced from this store”.
  • Handover is passing a conversation from the assistant to a person on your team, with its history, so the shopper does not have to start again.
  • Verified answer is an answer you have written yourself for a question that must always be answered the same way. It is used ahead of anything the assistant found.

How to use this glossary

Each term in the list above has its own page with a longer explanation, a worked example and the questions to ask a vendor who uses the word. Start with retrieval-augmented generation if the others are new to you.

Why a shopping assistant needs these words

When an answer is wrong, the fix depends on which stage failed: the passage was never chunked properly, never retrieved, or retrieved and scored below the line. The words above are the names of those stages. Once you know them, the trace beside each answer in your console reads like a diagnosis rather than a table of numbers.

See the words on a real answer

A trace is where retrieval, threshold and grounding stop being vocabulary and become the line between a product that was used and one that was dropped.

See the trace panel Read the guides