Guide 03 of 04 · 4 min read
Measuring ROI in week one: what an AI employee's first seven days can prove
Week one answers the speed and quality questions. Money questions need the month. This guide shows what to baseline, which targets the cabinet checks, and how to turn hours into a monthly figure.
Updated · from the product's own rules
Baseline the desk before launch
For one week before release, write down how the desk runs today. Without a baseline every improvement is a guess. Four numbers are enough:
- Average first-reply time on web chat and email, including evenings and weekends.
- Enquiries per week, and how many got no answer at all.
- Hours a week a person spends answering them.
- That person's loaded hourly cost: pay plus benefits and paid leave.
The weekly card: results against your trade's targets
Each week the cabinet compares the results with the targets of the job description. Every trade shares four of them: errors under 5% of replies (10% for consulting, legal and tax), thumbs-down under 20% of rated replies, zero missed handovers, and the AI disclosure in 100% of conversations. The first-reply target depends on the trade:
| Job description | First-reply target |
|---|---|
| Salons and barbershops | 10 seconds |
| Real estate | 60 seconds (plus three qualification fields per lead and a 70% show rate on requested showings) |
| HVAC, plumbing and electrical | 5 minutes |
| Moving | 5 minutes |
| Consulting, legal and tax | 5 minutes |
| Restaurants and catering | 15 minutes |
First-reply time is measured from the customer's first message to the first reply actually sent, by the employee or by a person. Handovers count requests for a person, complaints and conversations you took over yourself. Speed-to-lead targets and the qualification questions per trade are on one page.
What week one proves, and what needs the month
Seven days show speed, coverage outside business hours, handover discipline and the tone of replies. Bookings that turn into revenue, show rates and repeat customers need the full 30 days of probation. Read the week's three sample conversations (one rated down, one handed over, one random) before you read the totals.
| System proposal | When the cabinet makes it |
|---|---|
| Retrain | Thumbs-down above 20% of rated replies, errors above 15% of replies, or two or more missed handovers |
| Let go | Exam reliability below 0.5, or three or more hits on mandatory rules |
| Promote | Probation over, at least 10 conversations, exam reliability of 0.9 or more, errors under 5%, at most one thumbs-down |
| Keep | Everything else |
The system proposes; you decide. A week-one card never reads “promote”, because promotion waits for the end of probation.
The ROI formula, with the arithmetic shown
- Hours25 h a weekWork the desk takes off a person
- × Rate$35 an hourLoaded cost of that person's time
- × Weeks4.33Weeks in an average month
- = Human side$3,789 a monthWhat those hours cost you now
- − AI fee$0 in the pilotThen the flat fee agreed on the review
- = NetYour monthly ROIBefore faster replies and after-hours leads
The $35 input sits near the US BLS median total compensation of $34.78 an hour (March 2026). In Ottawa, Job Bank Canada puts an in-house receptionist at about C$3,870 a month before benefits (NOC 14101). The ROI calculator runs the same arithmetic on your hours and rate; its fee field is an example input and no price. AI employee vs hiring breaks the human side down line by line.
Decide at the end of the month
At the end of probation the cabinet shows the month next to your baseline and asks for a decision: accept onto the staff, extend probation, or let go. The fee after the pilot is agreed on the same review, from the conversations, leads and handovers it handled. Rolling out to a team covers what changes for your staff once it stays.
Questions owners ask
Baseline the desk for a week, then compare: first-reply time, enquiries answered, hours a person no longer spends. Monthly ROI = hours handed over a week × loaded hourly rate × 4.33, minus the AI fee. With 25 hours at $35 the human side is $3,789 a month.
A first-reply target for its trade (10 seconds for a salon up to 15 minutes for a restaurant), errors under 5% of replies (10% for consulting, legal and tax), thumbs-down under 20% of rated replies, zero missed handovers and AI disclosure in every conversation.
Speed and coverage show in the first week. Revenue effects such as show rates and closed bookings need the full 30-day probation.
After probation, with at least 10 conversations, exam reliability of 0.9 or more, errors under 5% and at most one thumbs-down. You make the final call.
It is an example input for the arithmetic. The pilot is free for 30 days, and the flat monthly fee after it is agreed on the pilot review.
Run this guide on your own desk, free for 30 days
Bring your price list, hours and service area. The Unistaff team sets up the web chat and the email address with you, the employee sits its exam, and you approve its first replies. The team replies within one business day.