This Article was originally published on HR.com.

PCL Construction is a family of companies that build massive projects like bridges, office towers, and factories. Its ability to be efficient and profitable ultimately depends on whether its employees can get the right information at the right time.

Understanding this imperative, its executives poured huge amounts of information, ranging from policies and technical reports to training classes and workshops, into “PCL Connects,” a company intranet. The idea was sound. But the execution went off the rails – to the point where some workers started to refer to search on the company intranet as a “random document generator.”  Eventually PCL Connects was re-engineered and works well today.

HR professionals may well wonder why a project like the one at PCL construction is their problem. But if internal communication is part of your job – as it is for many HR departments – then intranet systems will be crucial to it. Too often, HR doesn’t get involved in these decisions, and the results leave employees less productive and unable to find the information they need. To contribute in a meaningful way, you’ll need to understand some of the issues involved in effective knowledge management.

The experience at PCL Construction is emblematic of the importance of knowledge management to company operations. In the absence of effective information retrieval, companies become sclerotic and everything slows down. But sadly, most companies’ approach to this problem has been haphazard. At any given moment, an employee may be searching within a corporate wiki, a shared set of directories, their email, or a Slack-type employee conversational system. They may be learning bits of information in workshops, company meetings, or conversations with tenured employees who seem to know everything – at least until they retire. HR is often intimately involved in some of these systems and shut out of others. But an information environment full of poorly designed, fragmented, overlapping, and disconnected systems can’t solve the problem. Adding more tools is easy – but more tools just make things more complex.

In fact, when knowledge workers complain about “information overload,” what they are really experiencing is “filter failure.” They’d be happy with the right answer, but poor information hygiene instead bathes them in irrelevant information that saps productivity and increases costs.

Fixing this problem requires first understanding the breadth and diversity of information held at the average large company. There are both unstructured knowledge (like internal blogs) and rigidly structured systems (like a managed collection of digital assets). There are low-value, unfiltered content collections (like Slack posts) and highly curated and vetted content repositories (like approved methodologies and best practices). Finally, there are both narrowly focused content (like engineering troubleshoot guides) and broadly applicable content (like company vision statements). Tossing this all into a big bin without properly engineering access to it is likely to generate a “random document generator” like the original system at PCL Construction.

The solution to the knowledge retrieval problem is a careful and deliberate system of tagging, to indicate which is the structured, high-value content most applicable to a given problem. But tagging everything is a mammoth, endless task.

This is where AI can make all the difference. Using a process called “text analytics” and starting from a carefully tagged set of training data, a machine can identify patterns in documents. The resulting “auto-classifier” creates models to represent the concepts contained in the documents, based on a master classification structure called an “ontology.” For example, when documents contain an organization’s legal name, the term “policy,” a list of employee instructions and links, and an author designated as an HR staffer, text analytics can confidently classify it as an HR policy.  

Naturally, machines sometimes make errors in classification. But if you include a human who is spot-checking these classifications, you can improve the algorithm. And the improved algorithm can then quickly and accurately tag huge collections of documents and make them accessible in ways that are most helpful to employees seeking answers. 

Organizations can fine-time performance further by having machines observe how employees react to the information and answers they receive. Do they find it useful? Keep searching because it fell short? Give up and create their own answer? Call a support line? Analytics on the intranet pages reveals where the algorithm is surfacing high value content, and where it needs improvement.

By making large and continuing improvements in the classification and retrieval of intranet content, AI can take a company’s productivity to far higher levels. When HR is involved with these improvements, it can make a meaningful and measurable contribution to the company’s bottom line.