Enterprise Search

Large companies accumulate enormous amounts of files over the years: contracts, presentations, emails, meeting notes, spreadsheets with sales figures. This content is scattered across various systems, often in multiple countries and multiple languages. Enterprise search is a search function that scans precisely this internal inventory rather than the public internet. An employee types in a question and gets results from their own company’s records. The decisive difference from a normal web search: every user may only find what they are actually authorized to see. So the search must not only know where something is located, but also who is allowed to see it.

What lost documents cost a company

In surveys, employees regularly report spending several hours per week searching for information. With thousands of employees, this adds up to considerable costs. It becomes even more expensive when someone can’t find a document and therefore does the work a second time. Proposals get written twice, analyses get created twice.

There is also a legal aspect. Companies must be able to prove, during audits or legal proceedings, which records they hold on a given matter. Without a functioning internal search, this becomes a Sisyphean task. The General Data Protection Regulation also requires that a company be able to state, upon request, what data it has stored about a person.

Since language models became popular, the topic has gained additional importance. Anyone wanting to build an internal company chatbot first needs a reliable search over their own data. The chatbot is only as good as the documents it finds. That’s why many companies are now investing in systems they neglected for years.

From the index to the permission check

First, a program collects the content. It connects to the company’s individual systems, such as the file server, the email program, or the project software. It extracts the text from every document, including PDFs and presentations. These texts end up in an index, essentially a giant keyword directory. The index is continuously updated whenever files change.

Two methods come together in the search itself today. Classic keyword search finds documents in which the typed words appear literally. Semantic search instead compares meaning: it converts texts into long sequences of numbers, where texts with similar meaning receive similar numbers. This way you find a text about “notice periods” even if you searched for “terminate contract.” The combination of both methods is called hybrid search.

The most elaborate part is the permission check. For every document, the system also stores which people and groups are allowed to access it. Before results are displayed, these permissions are checked against the user. A common mistake in practice: the permissions in the index are outdated, and someone suddenly sees salary lists. Such mishaps are a main reason why projects of this kind fail.

From SharePoint to the company chatbot

The best-known products in this space are Microsoft Search in Office 365, Google Cloud Search, and Elasticsearch, which many companies run themselves. Newer providers like Glean or Perplexity Enterprise advertise that they deliver answers in full sentences instead of just lists of results. Salesforce and SAP have also built their own search functions into their software.

In business news, the term usually appears in connection with AI assistants for companies. The underlying technology is often called RAG, short for retrieval-augmented generation. In this setup, the system first searches for matching documents and only then has a language model formulate the answer. Enterprise search is the search component in this architecture, without which the rest cannot function.

A common misconception is that such a search is simply a smaller version of Google. Google benefits from the fact that millions of pages link to one another, allowing it to infer what is important. Company documents hardly link to each other at all. That’s why the ranking of results must rely on other signals, such as the age of a file, the user’s department, or how often a document has been opened.