Asked or watched: two ways to find out how your team works.

Picture a rush order landing at four on a Thursday afternoon. The newest estimator opens one capability, and it prepares the quote the way your best estimator would: the right premium, the press schedule checked, and the customer whose files always need extra prepress work flagged before anyone promises a date. The estimator reviews it, adjusts it and sends it in ten minutes. Every estimator works to the standard of the best one. That is what it looks like to 10X the ROI of your existing AI spend.

To build that, you first have to know how the work runs, including the parts nobody has written down. A company has two ways to find out. It can install software that records and measures what people do at their computers, or it can ask the people who do the work. Each produces a picture of the work, and the two pictures differ.

Two ways to find out.

• Watching.Observation software: an application on each computer, or connections to the systems people use, that log which applications are open, what is clicked and typed, how long each step takes and in what order. It turns those logs into process maps, time by task and a ranked list of where the hours go.
• Asking.A conversation with the person who does the work, about real cases: what started this one, what they opened first, what they decided and why, and what would have changed their mind. It produces the rules, the exceptions and the standard, written in their words and corrected by them.

What watching does well.

Observation software is good at three things people are bad at reporting.

• Volume.It counts every instance, on every desktop, with nobody estimating. Ask a team how many rush quotes it prepares a week and you get a guess. The log gives you a number.
• Timing.It shows where the minutes go, including the waiting, switching and searching that people stop noticing.
• Steps nobody mentions.Copying a job number from an email into the estimating system, then into a spreadsheet, then back again. The expert has done it so often it no longer feels like a step. The recording shows it every time.

For hundreds of people doing the same screen work, those numbers are useful, and they cost nobody an hour of conversation.

What watching cannot see.

A recording shows what happened on the screen. What makes the work good mostly happens somewhere else.

• Judgment.The log shows the estimator added a premium. It cannot show why, or why she left it off yesterday’s job, which looked identical.
• Exceptions.The cases that matter most are rare. The specialty paper, the customer in a dispute, the job that needs the owner’s sign-off may never occur in the weeks recorded, and when one does, the handling often happens on the phone or across the desk.
• The standard.Software records the usual way, averaged across everyone. What you want to build is the way your best person does it, which shows up in the data as an exception, not the pattern.

Then there is trust. People who know they are being recorded may work differently and say less. The colleagues whose know-how you most need are the ones you will ask to shape what gets built, and recording them first can tell them they are a cost to be measured. In a smaller company a role may have a single holder, and then measuring the role means measuring a person, whatever the dashboard calls it.

One piece of work, both ways.

Here is what a recording shows, and what one hour with the person who does the work produces.

Illustrative example with invented data. Pricing a rush order at an invented 50-person commercial printer.

Asking misses the counts and the steps people have stopped noticing. Both gaps close without a recording. The estimating system already holds the number of quotes: the work’s own records count the work. Ask the estimator to talk you through one live quote, start to finish, and the retyping appears.

The six questions.

That hour runs on six questions, each put to real cases, ideally three: one routine, one awkward and one the person would handle differently now. The full interview guide is in The questions that turn expert judgment into a workflow your whole team can run.

Illustrative example with invented data. The same invented printer. The questions put to its senior estimator.

Where Fully Powered stands.

Asked, not watched. We ask the people who do the work, and we never record or measure them. A role in your plan is a kind of job, such as estimator or credit controller, never a named person, and nothing we build ranks one colleague against another.

We ask because the thing worth building is your best person’s standard, and only that person can tell you what it is. Asking also does something a recording cannot: it makes the expert the author. Their rules become the team’s method, their awkward cases become the tests, and they correct the written version before anything is built. Build your AI plan with the people who do the work shows how to keep them involved after the first conversation, and Your SOP describes the happy path. Where are the exceptions? covers work that already has a written procedure.

Start from the outcome, map the work that crosses teams, land each capability on the roles that do it, prove it on real work, then roll it out. Asking is how each step learns what good looks like. Your plan decides which work to ask about first; A plan before agents sets out how.

The finished capability runs in the AI your team already uses: ChatGPT, Claude, Copilot or Gemini. Your platforms, your pace, your people, and your organization keeps control of how the work gets done.

Bring the piece of work your team waits on one person for, and the role of the person who does it best.