The phrase “productivity tax” is the right one for this problem. Nobody in the organization had decided that agents should spend a working week each year looking for documents. It accumulated, one system at a time, until it was simply how the job worked.
Measuring the invisible
The hard part of this engagement was not the search technology. It was establishing a credible baseline for something nobody had ever measured. Time lost to retrieval does not appear in any system of record; it has to be observed, sampled and modelled.
Precision over recall
In a regulated context, the failure modes are asymmetric. An agent who finds nothing asks a colleague. An agent who confidently finds the wrong version of a compliance document creates a problem that surfaces months later. The solution weighted precision and document provenance accordingly, even where that meant returning fewer results.
The measurement that mattered
Search volume went up, which is the metric most search projects report. It is also close to meaningless — more searching can mean the tool is failing. The number worth defending was recovered hours per agent, and that is what the program was held to.