RAG vs. traditional search: when enterprises actually need it

Traditional enterprise search — the search bar in your intranet, wiki, or document store — is a document finder. You type keywords, it returns a ranked list of files, and the last step is always on you: open the document, scan it, extract the answer.

Retrieval-augmented generation (RAG) collapses that last step. Instead of returning ten documents that might contain the answer, a RAG system retrieves the relevant passages and has a language model compose a direct answer, with citations back to the sources. The difference sounds small. In practice it changes who can use your knowledge base, and how often they do.

So when do you actually need RAG? The strongest signal is question shape. If your teams ask things like "what's our parental leave policy for contractors in year one?" — questions whose answers are assembled from clauses across documents — keyword search makes them do the assembly. RAG does it for them. If your users mostly need to locate a specific known file ("the Q3 board deck"), classic search is faster, cheaper, and easier to maintain.

The second signal is scale of repetition. RAG systems earn their keep when the same categories of questions are asked hundreds of times a month — onboarding questions, policy lookups, product specs, support macros. Every repeated question is a place where minutes of searching become seconds of answering.

What the vendor demos skip: RAG inherits every data-quality and permissions problem you already have. If two policy documents contradict each other, a search engine shows both and lets a human judge; a naive RAG system may confidently answer from the stale one. And if your document store has loose permissions, RAG will cheerfully surface content people could technically access but never found before. Permission-aware retrieval and citation-first answers aren't nice-to-haves — they're the difference between a co-pilot and a liability.

Our rule of thumb: start with your top 20 recurring question categories. If most are lookup-by-title, invest in better classic search. If most are answer-assembly questions over living documents, a scoped RAG engine — permissions-aware, citations mandatory, evaluated against a golden question set — will pay for itself quickly.