Where the Hours Hide: AI in Operations

Ask a team where its time goes and the answers name the visible work — the orders processed, the tickets closed, the reports filed. Time a real process end to end, though, and a second, shadow workload appears, usually larger than anyone guessed. It is not laziness and it is rarely the steps anyone lists. The hours hide in the seams: the handoffs, the waiting, the re-typing of information that already exists somewhere, the hunting for the current version of a thing. This is why AI in operations so often disappoints when it is aimed at the visible work, and so often pays when it is aimed at the seams. The seam work is voluminous, rule-bound, and nobody’s favorite — close to a description of what the tools are for.

The usual hiding places

They recur often enough to make a checklist. Re-keying: the same facts typed into a second system, a spreadsheet, an email, because the first system cannot hand them over. Waiting: work sitting idle between steps — in an inbox, in a queue, in a person’s memory — which no stopwatch on the steps themselves will ever catch. Searching: people hunting for documents, versions, and answers that exist but cannot be found, then re-creating them when the hunt fails. Rework: errors discovered three steps downstream, returned upstream at full fare. And translation: information reformatted between every pair of hands it passes through. Walk one process with a notebook and count these honestly, in your own operation, with your own numbers — the total usually startles the people who run the process daily, because seam work is invisible to everyone doing a small piece of it.

Fix the process first — then aim the tool at the seam

The order of operations matters more than the tool choice, and the tools themselves should stay unglamorous and vendor-neutral in your thinking: no product is the answer to a process nobody has mapped. Some seams should not be automated at all — they should be eliminated, because the handoff exists for a reason that expired years ago, or the report is formatted for a reader who retired. Automating an unnecessary step makes it faster and immortal. The seams that survive that audit are the honest candidates: the summarization of long threads into a brief, the drafting of the routine record from the raw notes, the first pass at sorting and routing what arrives. And one fence does not move: a human owns every output. AI in operations drafts, gathers, and first-passes; a named person checks what matters and answers for what ships. The hour saved by skipping the check gets repaid, with interest, the first time the tool is confidently wrong in front of a customer.

Prove it on one process, in your own numbers

The persuasive case for AI in operations is never a vendor’s benchmark; it is one process you own, timed before, timed after. Pick the seam with the honest count behind it, make the change small enough to survive a bad week, and measure the same way twice. Sometimes the finding will be that the tool was not the answer — the fix was deleting a handoff — and that finding is a success of the method, not a failure of the pilot. Run this way, operations work stops being a technology rollout and becomes what it should have been all along: a guided tour of where your hours actually live, conducted by the people who do the work, with the tool carrying the seams and the humans carrying the judgment.

The hours are findable. They are just not where the job descriptions say — they are between the jobs, waiting to be counted.

← All Insights