Ask anyone on your team how they spent their week and "looking for things" will not be on the list. It is still one of the biggest line items.
The search tax
In its 2012 report on the social economy, the McKinsey Global Institute estimated that interaction workers spend nearly 20% of their workweek looking for internal information or tracking down colleagues who can help. That is roughly one day a week, per person, spent finding answers that already exist somewhere in the company.
The tools have changed since 2012, but the problem has arguably grown. Knowledge is now spread across more places than ever: shared drives, chat threads, ticketing systems, CRMs, PDFs and the heads of a few long-tenured people.
Why traditional search does not fix it
Most internal search tools match keywords. That works when you know the exact name of the document. It fails when you have a question, like "what is our refund policy for annual plans signed before the price change?", whose answer is a paragraph buried in a 40-page policy document.
What an internal copilot does differently
An internal copilot answers questions in plain language, using your own documents as the source of truth. A well-built one:
- Reads across systems, not just one shared drive.
- Cites its sources, linking to the exact document and section so people can verify the answer.
- Respects permissions, so people only see answers drawn from documents they are allowed to access.
- Says when it does not know, instead of guessing.
That last point matters most. A copilot that confidently invents an answer is worse than no copilot at all.
Where to start
The MIT NANDA GenAI Divide research found that the strongest returns from generative AI came from back-office work rather than flashy customer-facing projects. Internal knowledge is a good example: the users are your own team, the content is already yours and the value shows up in hours saved every week.
Good first candidates share three traits:
- The questions repeat. Policy, process and product questions that come up every week.
- The answers are written down. Somewhere, in some document, even if it is hard to find.
- Mistakes are recoverable. Start where a wrong answer is an inconvenience, not a liability, and keep a human in the loop for anything consequential.
Measuring the payoff
Before you build, take a baseline: how many internal questions does a team field each week, and how long does each take to resolve? After launch, track the same numbers along with how often people click through to the cited source. Rising usage with a high click-through rate is the signal that people trust the answers and are checking the work, which is exactly what you want.
