At 9:10 on a Tuesday morning, the quoting e-mployee at a Beirut distribution company is paused. A vendor change has switched it off until the afternoon, perhaps longer. Customers are already waiting for prices. In the sales office, the two people who used to write those quotes by hand look at each other and ask the same thing: where is the price-exception sheet?
Neither knows. One thinks it lives in a shared folder that was reorganised in the spring. The other remembers a spreadsheet with a red tab, but not whose laptop it was on. It takes forty minutes to find a version that may or may not be current.
This is a composite example. The company, the people and every number in this piece are illustrative, built from patterns in my work, and none of it describes a real client.
The saving has a second half
When a company puts an AI worker on a repeated task, it reports a saving, and the figure is usually honest. By an e-mployee I mean an AI worker given a defined role in a team: a named task, an owner, a standard for done. In my composite, the quoting e-mployee produces about 40 quotes a day, and a person spends roughly six minutes reviewing each one.
What the time-saved figure leaves out is the second half of the saving. The hand skill that used to sit in the team is quietly leaving it. Nobody decides to let it go. The task simply stops being practised. The exception rules that lived in people's heads move into the tool's setup, and on a normal day nobody needs to say them aloud.
My claim is this: an AI worker with a defined role is only as dependable as the human who can still do its job on the day it is gone. A normal day hides that. A pause reveals it, whether the cause is a power cut, an internet outage, a vendor change, a policy change on the model side or a paused account. The causes differ. The Tuesday is the same.
So the saving has two halves. One is the time returned to the team. The other is the small, regular cost of keeping someone able to take over. A company that books only the first is borrowing against the second.
What 70% cover versus 33% cover looks like
I use one number to make this visible, the cover ratio: items a person can finish by hand in a day, divided by items finished on a normal day. It answers one question. If the e-mployee is gone, what share of the work still gets done?
Here are the illustrative figures. Normal volume is 40 quotes a day. A practised person writing a quote by hand needs about 25 minutes, so in six productive hours that is about 14 quotes a day. Two people available gives 28 a day, which is 70% of 40. The company is inconvenienced, not stopped.
Now move six months on. Nobody has written a quote by hand since. The first attempt takes about 55 minutes, because the sheet must be found, the rules recalled and the format rebuilt. That is about 6.5 quotes a day each, 13 for the pair, a cover ratio of about 33%.
| Situation | Minutes per quote by hand | Quotes a day, two people | Cover | Waiting after a 3-day pause |
|---|---|---|---|---|
| Practised hand | 25 | 28 | 70% | 36 |
| Six months unpractised | 55 | 13 | 33% | 81 |
| With The Unplugged Hour | 30 | 24 | 60% | 48 |
Composite example, illustrative figures. Three days of demand is 3 x 40 = 120 quotes.
Take a three-day pause. Demand does not stop, so 120 quotes arrive. At 70% cover, 84 get done and 36 wait. At 33% cover, 39 get done and 81 wait. Same team, same tool, same pause, and the backlog is more than twice the size.
The honest caveat: this is a worked illustration, and the 55 minutes is an assumption chosen to show the shape, not a measured rate. Redo it with your own minutes, hours and people. What I would hold on to is the shape: the skill fades, invisibly, until the day it matters.
The Unplugged Hour in three moves
The Unplugged Hour is a weekly practice with three moves: Name, Shrink, Run.
- Name the cover. For each e-mployee role, write one primary cover person and one deputy on the role card, by name. The role card is the one-page record of what the e-mployee does, who owns it and what done looks like. Add a single line: Cover: Layla. Deputy: Karim. "The sales office" is not a name.
- Shrink the job. Write a minimum version of the task that a human can finish at reduced volume, on one page. For the quoting role: the top 10 customers, the standard price list, no exception handling. Whatever is not on the page waits until the e-mployee is back. This is what makes a 40-quote job realistic for two people.
- Run the hour. Once a week, each cover person does the real task by hand for 60 minutes, on live work, with a timer running. They note every point where they got stuck, update the one-page sheet afterwards, and log the hour.
Live work matters, because a practice exercise tests nothing. The timer gives you the minutes per item, which is the input to your cover ratio.
In the composite, two people spend one hour each: 2 hours a week, about 104 hours a year. Hand time stays near 30 minutes per quote, which is 12 quotes a day each and 24 for the pair. Cover holds at 60%. After the same three-day pause, 72 quotes are done and 48 wait, instead of 81.
What to avoid, and what it costs
Three things to avoid. First, it is not a fire drill. An unannounced shutdown teaches fear, not skill; the hour is scheduled and a little boring, which is the point. Second, it is not a punishment. Someone slow in week one is giving you data about the gap, and the sheet is the fix. Third, it is not a duplicate team. You are keeping a smaller version of the job alive, not paying twice for the same output.
The cost in the example is 2 hours a week. The return is 33 fewer quotes waiting after a three-day pause (81 against 48), or about 11 more quotes finished on each paused day (24 against 13). I suggest a target of 50% or more for roles that touch revenue, and about 25% for back-office roles.
Rotate the deputy now and then, so the skill spreads and one resignation does not remove your cover. And drop the hour entirely for roles where cover is not valuable. If a task can wait a week, do not spend two hours a week keeping it warm.
Start this week
Pick the one e-mployee role whose pause would hurt most. Name two people for it. Book the first hour in both calendars, this week rather than next quarter. During that hour, time the work and count how many items get finished, then calculate your first cover ratio.
This piece starts where onboarding ends. If you are still bringing a new AI worker into a team, my two-week ramp plan comes first, and a clear job description is what the role card is built from.
Which role on your team would you test first, and who would you name as its cover?
Related evidence: NIST's contingency-planning guide sets out how organisations evaluate an information system and its operations to determine what it takes to keep the service available when a dependency fails, and ties that planning to organisational resiliency and the system development life cycle. (NIST's contingency planning guide for information systems)
Google's SRE book describes how a system with insufficient capacity for its steady-state load saturates its queues, so latency and error rates rise at the point of measurement while the actual cause is contention somewhere upstream. (Google's SRE book on saturated queues and cascading failure)
