Now imagine how much time it would take to retrieve information in an enterprise environment, when the answer you need could be buried in a document repository, locked away in a business application, or simply “known by someone” on another team.

Well, researchers who surveyed organisations across Europe estimated that employees lose the equivalent of 4.8 working weeks each year just trying to find the information they need across different tools.

In this article, we’ll look at how enterprise AI search can help employees find the information they need, even when they don’t know which system to look in or exactly what to search for, and how much time that can save when finding an answer would otherwise mean searching through several places.

What is knowledge discovery, and why do enterprises need it?

Knowledge discovery is the process of examining an organisation’s existing information to find relevant knowledge, understand relationships between different pieces of information, and identify information that was not already obvious or explicitly searched for.

To translate this, it helps employees access existing knowledge they may not know about or know where to find, so they can use it to solve problems and make decisions without starting their research from scratch.

When existing knowledge is easy to find across a business, employees can:

- find scattered information

- resume previous research

- find solutions to problems other teams have already faced

- understand data or decisions without having to reconstruct the reasoning behind them

In the best-case scenario, when a business does a good job of managing its knowledge, less time and effort will go into solving the same problems twice, leaving more of both for new work.

Why is scattered and siloed enterprise knowledge a growing challenge?

First, let’s focus on the source of the problem. Everyday work produces a steady stream of information.

But even when an organisation invests time and resources the time to document and store its project work, customer records, policies, and other information employees need, finding the right information can become harder as the company grows and different teams store their work in different systems (or sometimes just in “their heads”).

The direct consequence further down the line is that employees can become less productive and more likely to rely on outdated or inaccurate information.

Just look at Atlassian’s State of Teams research, which found that Fortune 500 companies waste 2.4 billion hours every year searching for information, with executives and teams spending around a quarter of their workweek on it.

For an enterprise, all of that ultimately means increased operating costs and an impact on overall business performance.

Can AI improve enterprise knowledge discovery?

The limitation of current knowledge discovery systems (Wikis or intranet platforms) is that, despite being good at finding relevant information, they typically return documents or search results that employees still have to read and work through themselves.

According to iManage Knowledge Work Benchmark Report employees spend between 30 minutes and two hours searching.

And here is why 89% of organisations expect cost savings once they implement AI.

AI brings capabilities that allow employees to:

• Search by meaning (semantic retrieval and intent recognition): Ask a question in their own words and find relevant information even when the company’s documents use different terminology.
  
  

• Connect knowledge across sources: Ask one question and get relevant information from different company systems, as long as those sources are connected and accessible, instead of searching each one separately.
  
  

• Investigate further (agentic retrieval): Ask questions that require more than one search and let the system follow up on what it finds when additional information is needed.
  
  

• Handle multi-part questions (agentic reasoning): Ask a question that requires information from several places and get an answer that combines what the system finds, without having to search for each piece separately.
  
  

• Get answers from company knowledge (RAG): Get a direct answer grounded in retrieved company information instead of having to open, read, and compare the search results themselves.

Case Study: Speeding up research for Norway’s leading financial media company

Finansavisen is a leading Norwegian financial media company that covers business, financial markets, investing, and the people shaping Norway’s economy. For more than 30 years, it has kept readers informed about developments in Norwegian business and the financial markets that affect companies and investors.

The challenge

Finansavisen’s archive contains reporting that stretches back to 2002, much of which can still be relevant to stories its journalists are working on today.

Finansavisen’s archive contains reporting that stretches back to 2002, much of which can still be relevant to stories its journalists are working on today.

So, if a journalist wants to write a new article about an investor’s view of a particular company, previous coverage can add context and show readers how that view has changed over time, and perhaps why.

Finding one quote is easy enough. But what if the journalist needs analyst or investor statements from different points in time, along with the data reported at the time? They would have to locate the relevant articles and read through them to bring those pieces into the new story.

With conventional search, finding those articles often depends on knowing the right terms to look for. If the journalist does not know exactly what to search for, or needs to connect information across several articles, they can spend hours searching and reading before they even start writing the new piece.

As Finansavisen’s team and our forward-deployed engineers worked through the problem together, together they began to shape the process where an AI layer on top of the existing system could reduce the time journalists spent searching through previous coverage.

The solution: Agent for faster article production

The first step was to vectorize Finansavisen’s archive, so the search no longer had to depend only on the exact words used in an article.

A journalist could instead describe what they were researching in natural language.

The system would search across the archive and retrieve the passages and data most relevant to the question, even when they came from different articles.

So rather than getting a list of potentially relevant articles and opening each one to find a quote, figure, or previous statement, the journalist could go straight to the parts of the archive that addressed the question.

Soon after, we added another feature to the system, with the goal of letting journalists investigate broader questions across the archive. 

In short, through the same chat interface, a journalist could give the assistant some context about the article they were writing and ask it to investigate a particular angle.

Say they are writing about current market conditions and want to compare them with a particular period ten years ago. The assistant could pull reporting from both periods and draft a response based on the relevant passages.

In the end, the journalist still makes the call on whether the comparison makes sense and whether it is even relevant to the article. The idea was to free them from piecing the research together through a series of separate searches.

Now, a process that could previously take hours can be completed in a fraction of that time.

FAQ

Where should we start if we want to use AI with our company knowledge?

Start with a specific piece of work where finding or using information takes too much time. Look at what employees are trying to accomplish, what information they need to do it, and where that information currently lives.

That gives us a starting point to build around. We can identify the data the AI needs, connect the relevant systems, and test the result against questions employees actually encounter instead of starting with a company-wide AI project.

Do we need to clean and organize all of our company data first?

No. You do not need to clean up years of company data before you can start.

We start with the sources the system needs for the job at hand and check those for anything that could interfere with the search. Say employees need to search contracts and internal policies. 

We would focus on those documents first and check whether outdated copies could show up alongside current ones, whether important details such as dates are missing, or whether the text can be extracted properly.

We fix those issues as part of the project rather than asking you to clean and reorganize everything first.

Can AI search across information stored in different systems?

Yes, as long as those systems can be connected. Your information does not need to be moved into one database before AI can search across it.

Say your contracts are stored in one system, internal documentation in another, and customer information somewhere else. Those sources can be brought into the same search experience, so an employee can ask one question instead of searching each system separately.

How we do that depends on the systems involved. Some sources can be indexed, while others are better queried directly. The choice depends on factors such as how often the data changes, how the source makes that data available, and which access controls need to be preserved.

Can AI do more than find documents?

Yes. Finding the right documents is only one part of what it can do. AI can also work with information from several sources to answer a question that no single document answers on its own.

For example, you might want to compare this year’s figures with the previous three years, see how an internal policy has changed over time, or compare what different teams have said about the same issue. Instead of first figuring out which documents contain each piece, you can start with the question itself and let the system bring the relevant information together.

How do we keep company information secure and respect existing permissions?

If an employee isn’t allowed to access a document in the original system, they shouldn’t be able to retrieve its contents through AI either.

We preserve those access restrictions when we connect the source and enforce them when someone searches or asks a question. The language model doesn’t get to decide who is allowed to see what.

Permissions are only one part of the security picture. Before anything goes live, we also establish what company data can be sent to the AI provider, where that data is processed, what gets logged, and how long any submitted data is retained.

How do we know whether we can trust the answers?

A convincing answer isn’t necessarily a correct one, so we test more than the final response. We check whether search retrieved the documents and passages that actually answer the question, and whether the AI’s response is supported by those sources.

Employees don’t have to take the AI’s word for it either. They can open the sources behind an answer and check the original document or passage themselves.

And the testing doesn’t stop once the system goes live. We keep looking for questions where search misses important documents or the AI makes claims its sources don’t support. That gives us something concrete to fix, whether the problem lies in retrieval or in how the answer is generated.

Do we need to replace our existing systems to add AI?

Usually, no. If your contracts are in SharePoint, customer records are in a CRM, and other company data sits in internal databases, there’s no reason to move everything just because you want to add AI.

The better question is whether those systems give us a reliable way to access the information inside them. Some already have APIs or connectors we can use. Documents from others may need to be indexed before employees can search them with AI.

Older systems can make this harder, especially if they don’t have a good way to connect with other software. That may mean some integration work, but it doesn’t mean ripping out the systems your company already relies on.

Exploring AI for enterprise search and knowledge discovery? Talk to Profico about how it can work with your existing knowledge, systems, and workflows.