For six years, I ran the intranet program at H&R Block. There was one particular leader who had the most to gain from an improved intranet search experience — he oversaw tax preparation software used by nearly 80,000 frontline workers across 11,000 locations.
The help desk was the first to know when the digital workplace failed to produce a valid search result. He'd been trying to solve it for years.
When I came to him with a proposal, I could see the skepticism in his eyes. I asked for 90 days and one simple thing — a three-question feedback form on the search results page:
What terms did you search for?
What were you hoping to find?
Which employee group are you in?
No new technology. No platform investment. Just a mechanism to capture quick feedback.
By the end of tax season, the form had shown us exactly where search was failing:
- Tax procedure content — the biggest problem area — was overhauled by a small team of subject matter experts, cut from 500 pages to 300 and moved somewhere the search engine could actually see.
- For common searches whose answers lived in other systems, we added wayfinding — so employees could actually get there.
- Key phrases were updated to match the terms employees were actually searching.
The result came faster than anyone expected.
Help desk calls related to one major content migration dropped 19.5% the following tax season — over $400,000 in savings. Our intranet search ranking improved from 16th to 4th globally out of 256 organizations in the Worldwide Intranet Challenge.
What if we ran the experiment today, in the age of AI?
Traditional search failure is clear to the end user at the moment it happens. When employees encounter poor results, they know something went wrong and try again.
AI search is different. It synthesizes whatever it can find and delivers a fluent, well-formed answer — whether that answer comes from your current policy or one retired in 2019, whether it reconciles two conflicting documents or fills a content gap with plausible inference.
But when AI search surfaces a wrong answer, the employee has no way to know. They might not find out until they've already acted on it — followed an outdated process or made a decision based on guidance that was retired two years ago.
By then, the connection back to search is gone. Trust erodes quietly. You won't hear about it — not directly, anyway. They just stop coming back.
