Who owns a virtual employee
Give an automation one named owner, a budget line in money and in hours, a 20-minute weekly review and a retirement date, all set on launch day.
On this page
An automation that runs unattended still needs someone who answers for it, the way a manager answers for a team. Ownership is four things written down on the day it ships: a named owner, a budget line, a review cadence and a retirement date. Without them an automation runs on the momentum of its launch, and nobody is placed to say whether it still earns its cost.
"Owned by sales"
Picture the agency's lead handler, the workflow priced in how to use this course, shipped with a one-line handover: owned by sales. Sales is the founder and two account managers. Five months on:
- The prompt had been edited four times by three people. None of the edits ran against the eval set, so when replies began offering a free audit the agency does not sell, nobody could say which edit did it.
- The platform plan had gone from $20 to $50 a month for a feature needed for one week, billed to the founder's card.
- The launch one-in-five sample, never revisited, was read by whichever account manager had time: in a busy May, 30 replies of the 80 planned.
- Asked whether the handler still paid, nobody could say, because nobody held the sums.
None of it was expensive; each was a duty with no name on it. A team is not an owner: a duty nobody notices gets done by nobody.
Four fields, set on launch day
The virtual employee frame gave the role a manager and a budget. Ownership turns those into a person, a number, a meeting and a date, written before the first lead runs.
| Field | The lead handler |
|---|---|
| Owner | one named account manager; the other covers holidays |
| Budget line | $305 a month, in money and in hours, each with an alarm on the rate |
| Review | 20 minutes, Thursdays at 16:00, a fixed agenda, a written log |
| Retirement date | 12 months after launch; moved only at a review |
The owner
One person, with one named backup. The owner runs the weekly review, signs every change to the prompt, the gates or the model, holds the budget line, takes the page when a connector fails, and decides at the retirement date. The job is knowing what a correct output looks like, so the owner is the person who would do the work by hand, or who manages whoever would. That is why the eval set in what "working" means is labelled by the owner and nobody else.
Two defaults are wrong. The builder knows how it works, not what right looks like, and a contractor leaves at handover. The founder knows what right looks like but costs $100 an hour on a $30 task: the review alone, 20 × 52 ÷ 12 ≈ 87 minutes a month, is $43 at an account manager's rate and $144 at the founder's. A founder hour belongs at the second hop of an escalation, where escalation paths and the manager puts it.
The owner's time is a cost the first sums left out: the lead handler's $262 a month becomes $305, a saving of $1,695, and the $1,200 build still pays back in about three weeks.
The budget line
A number on the owner's own cost centre, so the owner sees the bill first. It has two parts, because most of what an automation costs is people.
Money. Tokens and fees: an expected figure, an alarm on the rate, a hard stop. The lead handler's is $22 a month, alarmed on cost per lead, $0.0054, where a prompt that doubled in length shows first. The online retailer's support triage from module 8, a different company from the agency, expects $382 a month, about $12.70 a day; its line is $450, the alarm fires on any day over $16 or a month-to-date running 20% ahead, and the $450 is the stop, not the alarm. The arithmetic is in cost control.
Hours. The people line: 8 hours a month of escalations and sampling plus 1.4 hours of review, 9.4 in all, alarmed like tokens. Escalations move first: 40 a month is about 9 a week, so the alarm fires on any week over 14. No invoice ever arrives for hours, so nothing else tells the owner they grew.
The review cadence
Twenty minutes in the same slot every week, with the agenda in the weekly review. A launch week of the lead handler is about 92 leads, 18 sampled replies and 9 escalations: few enough to read in the slot, recent enough to trace to a cause. A monthly review would face 40 escalations and skim them. A daily review repeats the alarms' job.
At the retailer's 3,000 tickets a day the slot is the same 20 minutes, spent on rates rather than items.
The retirement date
Twelve months after launch, set on launch day, in the owner's calendar and the founder's. It makes payback a test at build time: a build that pays back after month 12 does not ship, which is why the 17-month board report in how to use this course stays manual. It schedules a question nobody asks while things run quietly: is this still worth its cost? And it keeps the documents alive, because at the date the owner needs the SOP, the two sums at today's volume and the latest eval score. A missing one is the review's first finding.
The date moves only at a review, by up to 12 months, with the sums rewritten, the eval passing and the reason logged. A date pushed back without the sums is not a date. When the sums say no, the next step is retiring an automation.
One row per automation
Every automation gets one row the founder can read in a minute:
| Lead handler | Invoice run | |
|---|---|---|
| Owner, backup | an account manager, the other account manager | the operations manager, the founder |
| Sum 1 at launch | $2,000 a month | $520 a month |
| Monthly cost with the review | $305, $785 with the gate from month 3 | $151 |
| Payback | 0.7 months | 3.3 months |
| Review slot, retirement | Thursday 16:00, month 12 | Friday 11:00, month 12 |
| Off switch | the trigger, and who takes the queue | the trigger, and who invoices by hand |
The invoice run from the arithmetic of a task costs $108 a month to run and check; the review adds $43 and moves its payback from 2.9 months to $1,200 ÷ $369 ≈ 3.3. The off-switch row is the one most often left blank, and an automation with nobody named to take its queue can only be retired by deleting it.
In practice
When a contractor or a vendor hands over an automation, the handover is not finished until the four fields carry a name and a number: who owns it, its budget line in money and in hours, when it is reviewed, and when it retires. "We'll keep an eye on it" is an offer to stay the builder, not to be the owner. Inside the company, announce all four in one sentence on launch day, with the owner's name in it.
The mistake that costs money is naming a team, or the builder, as the owner. The team splits every duty until nobody does it, and the builder either leaves or is the founder spending $100 hours on a $30 task.
Check yourself
1. The founder built the lead handler and wants to own it "because I know how it works". Price the choice and decide.
An account manager owns it. The review costs $144 a month at the founder's rate against $43, about $1,200 a year, and the account manager answers leads by hand, so she knows what a right reply is and labels the eval set. The founder stays the builder and the second hop for escalations.
2. Fifteen days into the month, the retailer's triage has spent $240 against its $450 line. Raise the line?
Not yet. The pro-rated line is 15 × $12.70 ≈ $190, so spend is 26% ahead and the alarm has fired. Read cost per ticket first. Flat cost per ticket on a quarter more tickets is growth, and the line rises with volume at the next review. A rising cost per ticket means something changed, often a deploy that broke the cached prefix; fix that, not the budget.
3. At the month-12 review the lead handler escalates 15% of leads instead of 10%, the platform plan is $50, and nothing else has changed. Move the date?
Rewrite the sums first. Run: $2 + $50 = $52. People: 60 escalations × 6 minutes = $180, plus the gate's $600, every reply read since month 3, and the review's $43. Total $875 against Sum 1's $2,000, a saving of $1,125: move the date 12 months and log the sums. Take the escalation rise to next week's review: a repeated reason is a rule to write.