Key takeaways
- An AI receptionist greets callers, answers simple questions and takes a message. An SI Employee is set up to finish the task: book, route, update and report.
- The gap matters because callbacks often fail; only 19% of US adults generally answer calls from unknown numbers[1].
- The employee model needs more setup: booking limits, urgency rules, an on-call path and someone who reviews the records.
- Ask every vendor to show, call by call, which tasks are completed without a person and which become messages.
Two labels, two different jobs
Vendors use the terms loosely, so start with working definitions. An AI receptionist is software that answers your phone, greets the caller in your business name, answers a few common questions and records a message for someone to act on later. An SI Employee is software given a defined set of tasks it is allowed to complete on its own, such as booking a routine visit, reaching the on-call technician or moving an existing appointment, plus a clear record of what it did.
The difference is not the voice, the speed or the number of languages. It is where the call ends. With a receptionist, most calls end in a message. With an employee, most calls end with the task done, and the message becomes a report.
Many products sit between the two. A system might book appointments but only take messages for emergencies, or route urgent calls but never touch the schedule. That is fine, as long as you know which tasks are which. The useful question to ask is not which label a vendor uses, but what share of your real calls will be finished without a person, and what happens to the rest. A product that completes the easy half of your calls and hands over the hard half cleanly can be worth more than one that claims to do everything and does it badly.
The names come from human roles
In a plumbing office the distinction already exists between people. A receptionist answers phones and passes calls on; the median US receptionist earned $18.27 an hour in 2025[2]. A customer service representative resolves the caller's issue and earned a median $21.53 an hour[3]. A dispatcher decides who goes where and earned a median $24.20 an hour[4]. Pay rises with the amount of decision-making the role carries.
AI products follow the same ladder. Answering is the easy part. Booking into a live schedule, judging urgency and routing an emergency are the parts that carry risk and value, and they are the parts you need to examine most closely.
Task by task
| Task | AI receptionist | SI Employee |
|---|---|---|
| Burst pipe at night | Takes a message marked urgent; someone checks it later | Takes the address and problem, marks it urgent and rings the on-call tech, then the backup |
| Routine drain cleaning | Takes the caller's details and promises a call back | Books a slot that fits your rules, or records a request if you prefer to confirm |
| Do you cover my ZIP code? | Gives a general answer or takes a message | Checks your approved service area and gives the approved next step |
| What do you charge to come out? | Says someone will call back | States the fee you approved, word for word, and never guesses repair prices |
| Move my appointment | Takes a message for the office | Finds the booking and moves or cancels it |
| Caller speaks Spanish | Varies by product | Continues in Spanish and sends your team an English summary |
| After the call | A message or voicemail transcript | A job summary with recording and transcript, urgent jobs flagged |
Why finishing the call matters
A message is only worth something if the callback connects. Pew found that only 19% of US adults generally pick up cellphone calls from unknown numbers, and 67% let unknown calls go to voicemail and check it if a message is left[1]. When your office returns a call from a homeowner who rang at 7 pm, there is a good chance the call goes unanswered, the homeowner has already called a competitor, or both.
Plumbing makes this sharper because demand peaks when offices are closed. Plumbers are often on call for emergencies, and evening and weekend work is common[5]. A receptionist-style product that takes a polite message at 10 pm has not solved the problem of the 10 pm leak. It has only moved it to the next morning.
Where the employee model needs more care
Giving software permission to act also gives it room to make mistakes. A receptionist that takes a bad message causes a bad callback. An employee that books the wrong job type into a full Tuesday causes a missed appointment and an unhappy customer. The employee model is only better if you set it up properly:
- Booking rules: which job types can be booked, how many per slot, which days and which ZIP codes.
- Urgency rules: the situations that count as urgent in your shop, written in plain language.
- An on-call path: who is rung first, who is next, and the honest message the caller hears if nobody answers.
- Approved answers: the exact fees, hours and policies it may share, and a firm rule to say it does not know otherwise.
- A reviewer: someone who reads the records daily for the first weeks and samples them after that.
Our guide to booking safeguards covers the first of these in detail.
Which one fits your shop
A receptionist-style product can be the right choice. It is simpler, needs less setup and suits some situations well:
- Choose receptionist-style if you have a staffed office during all the hours you want covered and only need overflow messages.
- Choose receptionist-style if you do not take same-night emergency work at all.
- Choose employee-style if you run an on-call rota and urgent calls currently go to voicemail or a personal cell.
- Choose employee-style if a meaningful share of calls are routine bookings or reschedules that follow clear rules.
- Choose employee-style if you lose jobs because callbacks happen too late.
We built Ava to be an SI Employee for this reason: the jobs that matter most to a plumbing company, from the night emergency to the routine booking, only pay off if they are completed on the first call. Whatever you choose, judge it on completed tasks, not on how natural the voice sounds.
Costs to compare beyond the subscription
Comparing the two models on the vendor's fee alone misses most of the cost. A receptionist-style product looks cheaper, but every message it takes creates follow-up work for a person. An employee-style product needs more setup time but removes some of that work. List these costs for each option before you decide:
- Callback time: minutes your office spends returning messages, including calls that are not answered and need a second or third try.
- Lost jobs: calls where the caller booked someone else before your callback reached them.
- Setup time: hours spent writing rules, approved answers, the on-call path and the service area, plus time to update them each season.
- Review time: minutes per day reading records, more in the first weeks and less once results are stable.
- Error cost: the cost of a wrong booking or a missed urgent call, which is why limits and reviews matter.
Our article on measuring return from AI call handling shows how to put numbers on each line with your own call data.
A two-week setup for the employee model
Days 1 to 3: write the rules
List bookable job types, slot limits, hours, ZIP codes, the urgent-call definition, the on-call order and the exact fees you will share.
Days 4 to 5: test calls
Make at least twenty scripted calls covering emergencies, routine work, reschedules, price questions and out-of-area callers. Fix every rule that produced a wrong result.
Days 6 to 10: after-hours only
Let the system handle nights and weekends. Keep bookings as requests that the office confirms each morning, and read every record.
Days 11 to 14: widen carefully
If records are accurate, allow direct booking for one or two simple job types and add daytime overflow calls. Keep reading a daily sample.
Example: the same Saturday call, two ways
Example: at 4:40 pm on a Saturday, a homeowner calls because the water heater is leaking into the garage. With a receptionist-style product, the caller hears a friendly greeting, gives a name and number, and is told someone will call back. The on-call tech sees the message at 6:15, calls twice and gets voicemail; the homeowner booked another company at 5:30. With an employee-style setup, the call is marked urgent from the description, the on-call tech's phone rings at 4:43, the tech speaks to the homeowner at 4:45 and is on the way by 5:10. The office sees the full record on Monday morning.
Questions that show what you are being sold
- Of your current customers' calls, what share end with a task completed rather than a message?
- Show me the record of a call where the on-call tech did not answer. What did the caller hear?
- Which tasks can I switch on or off myself, and which need your team?
- Can bookings be held as requests for my office to confirm while we build trust?
- Which features in this proposal are live, and which are planned?
Frequently asked questions
Is an SI Employee just a more expensive AI receptionist?
Not necessarily. The difference is the scope of tasks it may complete, such as booking, routing urgent calls and updating appointments. Compare products on the tasks they finish, then discuss pricing.
Can an SI Employee replace my office staff?
In most plumbing shops it takes over repetitive intake and after-hours coverage. People still handle judgment calls, complaints, complex quotes and anything the rules do not cover.
Is it risky to let AI book jobs?
It can be if there are no limits. Start with a short list of bookable job types, capacity limits per slot, and an option to keep bookings as requests until you trust them.
What should happen if the on-call tech does not answer?
Follow a written no-answer step: try the backup, then tell the caller honestly what happens next and when. Never let the system promise an arrival time it cannot guarantee.
References
Pew Research Center (2020). Most Americans don't answer cellphone calls from unknown numbers (opens in a new tab) โ
U.S. Bureau of Labor Statistics (2025). Occupational Outlook Handbook: Receptionists (opens in a new tab) โ
U.S. Bureau of Labor Statistics (2025). Occupational Outlook Handbook: Customer Service Representatives (opens in a new tab) โ
O*NET OnLine (U.S. Department of Labor), BLS 2025 wage data (2025). National Wages 43-5032.00 - Dispatchers, Except Police, Fire, and Ambulance (opens in a new tab) โ
U.S. Bureau of Labor Statistics (2025). Occupational Outlook Handbook: Plumbers, Pipefitters, and Steamfitters (opens in a new tab) โ