Key takeaways
- Buyers rarely shop around: 44% hired their agent after a single contact[2], so the first answered call matters more than a polished fifth one.
- Software is strongest on inbound first touches at any hour; people are stronger on months-long nurture, objections and judgment calls.
- AI-generated voices count as artificial under the TCPA[3], so AI outbound dialing needs prior express written consent. Most hybrids keep outbound human.
- Compare hours, not salary against subscription: count the first-touch calls on your ISA's desk and decide what that person should do instead.
One seat, two different jobs
Most teams that hire an inside sales agent are buying two jobs in one seat. The first is speed: picking up when a buyer calls off a yard sign, ringing back a portal lead within minutes, and getting a seller's address and timeline before another agent does. The second is patience: calling the same homeowner for the fourth time in five months, noticing the change in their voice when the school year comes up, and turning that into a listing appointment. Software has become good at parts of the first job. It is still poor at most of the second.
That split is the honest answer to the AI ISA question, and it explains why a hybrid usually makes more sense than a swap. Agents are not short of AI tools. In NAR's 2026 technology survey, 23% of REALTORS said they use AI daily and 25% weekly, while only 12% said they are not using it and have no plans to[1]. The open question is which conversations to hand over.
- Daily 23%
- Weekly 25%
- Occasionally 31%
- Curious, not tried 9%
- Not using, no plans 12%
Source: National Association of REALTORS®, 2026[1]
Show the data
| Item | Value |
|---|---|
| Daily | 23% |
| Weekly | 25% |
| Occasionally | 31% |
| Curious, not tried | 9% |
| Not using, no plans | 12% |
Most of that use is writing. Among REALTORS who use AI, 75% use it for listing descriptions, 56% for social posts and 52% for emails and follow-ups[1]. Putting software on a live phone line is a different decision: the caller is a person in the middle of the biggest purchase of their life, and every answer goes out in your team's name.
Speed and coverage: where software wins
Consumers do not shop agents for long. In Zillow's 2026 buyer survey, 54% of buyers who contacted an agent contacted only one before hiring, and 44% hired their agent after a single contact[2]. Sellers say it differently but mean the same: 77% of agent-using sellers rate an agent's first impression of responsiveness as very or extremely important[4]. The first conversation is often the only audition.
A full-time human ISA covers about 40 of the 168 hours in a week. Leads do not keep those hours. Agents' own schedules often run into evenings and weekends because that is when clients are free[5], and the calls follow the same pattern. An ISA who logs off at 6 pm leaves the Sunday-evening seller call to voicemail, and the Monday callback comes from a number the seller has never seen. Eight in ten Americans say they don't generally answer their cellphone when an unknown number calls[6].
This is where software on the phone earns its place. An SI Employee answers the calls a team forwards to it, all day or only after hours and when agents are busy, in the team's name. It asks a seller for the address, timeline and plans and books a listing consultation in your AssistCall calendar. It takes a showing request with the property, preferred times and whether the buyer already has an agent, for your agent to confirm. It answers listing questions only from the listing sheet you approve and marks anything else as not confirmed. When someone on the routing rule is free it transfers the call; when nobody is, it takes a detailed message. The tenth call of a busy Saturday gets the same questions as the first.
It also does not know which sign or ad the caller saw. A forwarded call carries no listing context, so the line has to ask, and the listing sheet has to be current. A price drop entered on Tuesday morning is only useful if the sheet changes on Tuesday morning too.
Conversation depth: where people still win
The long middle of a lead's life is a human job. The median seller thinks seriously about selling for three to four months before listing, and 27% think about it for six months or more[4]. Those months are where a good ISA earns a salary: the check-in after the neighbor's house sells, the straight answer when a seller asks whether spring really is better, the decision to stop calling someone who is clearly annoyed and send a market update instead.
- Reading hesitation. A seller who says “maybe next year” in a flat voice and one who says it while asking about school boundaries need different next steps.
- Handling objections. “We have a friend who's an agent” is the start of a conversation, not a field on a form.
- Knowing when to stop. Pushing a lead who has gone quiet burns goodwill a script cannot see.
- Noticing what matters. A buyer who mentions a divorce in passing may need a careful agent and a slower pace, and a person picks that up.
- Staying inside license law. Pricing opinions, negotiation and offer advice belong with a licensee. A human ISA can be trained on that line; an AI line should be set up to hand those questions to an agent rather than answer them.
Software asks the same questions in the same order every time. That is a strength on a first call, when consistency and complete notes matter most, and a weakness on a fifth, when the lead needs someone who remembers the last four.
Outbound calling is a different legal animal
Many AI ISA pitches include software that dials your old leads for you. Read the rules first. In February 2024 the FCC ruled unanimously that calls made with AI-generated voices are “artificial” under the Telephone Consumer Protection Act[3]. Telemarketing calls to residential lines that use an artificial or prerecorded voice need the called party's prior express written consent, and such messages must state clearly at the start who is responsible for the call[7]. A web lead who typed a phone number into a portal form has not necessarily given your team that consent.
Liability is counted per call. The TCPA lets a person recover $500 for each violation, and a court can triple the award for willful or knowing violations[8]. An AI dialer that works through 400 aged leads in an afternoon can create 400 problems in an afternoon. Human outbound calling has its own rules too, from calling hours to do-not-call scrubs; the TCPA guide for real estate agents and AI voice calls and the TCPA cover both.
That is the practical reason to put software on inbound calls, where the lead dialed you, and keep outbound prospecting with people working a scrubbed, consented list. The SI Employee described here answers calls that come in. It does not run outbound dialing campaigns.
Cost structure: a salary against a service
A human ISA is a fixed cost that arrives whether the phones ring or not. Inside sales pay is usually benchmarked against customer service work, and the median hourly wage for customer service representatives was $21.53 in May 2025[9]. Wages are only part of the bill: across private industry, benefits made up 30% of what employers paid per hour worked in June 2026[10]. Divide $21.53 by 0.70 and a median-wage hire costs roughly $31 an hour before the desk, phone line, CRM seat, lead-source logins and the team lead's coaching time. Ramp-up and turnover push the cost per productive hour higher; what replacing an ISA costs works through that, and ISA compensation models covers the pay plan.
An AI line is priced as a service, so salary against subscription is the wrong comparison. The useful one is which hours of the human ISA's month go to work software could absorb, and what that person would do with the time instead. Start with first-touch inbound calls: the sign call, the portal call, the “what's it listed for” call. They are short, repetitive and urgent, which makes them a poor use of a skilled ISA and a good fit for automation.
| Input | Example |
|---|---|
| Inbound lead calls per working day | 22 |
| Minutes per call, including notes | 7 |
| Share the SI Employee could take | 55% |
| Loaded hourly cost of the ISA | $31 |
| Calls a month | 477 |
| Staff hours on these calls | 56 |
| Hours handed off | 31 |
| Monthly value of that staff time | $950 |
How it is worked out: Calls a day × 21.7 working days × minutes per call ÷ 60 × share handled = hours; hours × hourly cost.
Assumptions
- Hourly cost: the $21.53 median hourly wage for customer service representatives[9], grossed up because benefits are about 30% of employer cost per hour worked[10].[9]
- Handled share: sign, portal and overflow first touches that end in a booked consultation, a showing request or a detailed message. Calls that need judgment stay with the ISA.
- Call volume and minutes are illustrative for a four-agent team; take yours from your phone system.
Example only. The result is ISA time spent on first-touch calls, not a saving you bank.
In this example about 30 hours a month of first-touch calls come off the ISA's desk. That is close to a full working week every month that can go back into nurture calls, appointment confirmations and the database nobody has called since spring. Whether that is worth paying for depends on what the ISA does with the week.
Supervision: what each one needs from you
A human ISA needs managing the way any salesperson does: a daily huddle, a scorecard, recorded calls reviewed every week and someone to escalate to. A daily ISA scorecard keeps that from turning into opinion. An AI line needs a different kind of attention. Someone owns the listing sheet and updates it the day a price drops or a listing goes pending. Someone reads a sample of transcripts each week. Someone changes the routing rule when an agent goes on vacation. Owners and team leads can also listen to a call in progress, whisper guidance the caller does not hear, or take the call over.
Disclosure belongs in the setup, not in a later fix. In Pew's 2025 survey, 76% of Americans said it is extremely or very important to be able to tell whether content was made by AI or by people[11]. A greeting that tells callers plainly they have reached an automated assistant, plus a recording notice where your state requires one, avoids a bad surprise. Disclosing AI to callers and leads has wording to start from.
| Area | Human ISA | SI Employee |
|---|---|---|
| Hours covered | Scheduled shifts, typically about 40 a week | Every hour you forward calls to it |
| First-touch speed | Fast when free; queues when already on a call | Answers each forwarded call as it rings |
| Months-long nurture | Core strength | Not its job; hands the lead to your team |
| Outbound prospecting | Yes, from a scrubbed, consented list | Not used for outbound dialing here |
| Consistency | Varies with fatigue, mood and tenure | Same approved questions on every call |
| Notes and CRM | As good as the person's habits | Recording, transcript and summary after every call, sent by webhook or Zapier |
| Judgment calls | Reads hesitation and adjusts | Follows written rules; hands off the rest |
| Your supervision | Daily huddle, call reviews, scorecard | Approve the script, keep the listing sheet current, read transcripts weekly |
| Main failure mode | Burnout, turnover, skipped follow-up | Out-of-date listing sheet or routing rules |
How a hybrid setup splits the work
In practice the split follows the direction of the call. Inbound first touches go to the line that never clocks out; nurture and outbound go to the person with judgment and a consented list. The handoff between them is the call summary, which is why it matters that the summary lands in the CRM where the ISA works, not in somebody's personal inbox.
- Caller
- SI Employee
- System
- Your team
01Caller
Seller calls at 8:10 pm
A homeowner calls the team line about selling in the spring.
02SI Employee
Answers and qualifies
Asks for the address, timeline and plans, then offers consultation times.
03SI Employee
Books the consultation
Puts a Tuesday 10 am valuation visit in the team calendar.
04System
Summary reaches the CRM
Recording, transcript and summary go out by email and signed webhook.
05Your team
ISA calls Monday morning
Confirms the visit, asks what prompted the move and notes it for the agent.
06Your team
Agent runs the appointment
Arrives with comps and the seller's own words from the transcript.
Example: a fictional team. Routing, calendar and CRM fields are whatever you approve.
The solo listing agent example shows the same pattern at the smallest scale: sign calls during showings go to the line, neighbors who mention selling get a valuation visit booked, and the agent keeps every pricing conversation. On a team, the human ISA sits where the solo agent sits in that example, picking up each lead the morning after with the transcript in hand.
Before you decide, read the full list of what it does today and what is still planned. Listing feed sync from the MLS and instant showing booking against agent calendars are on the planned list, so showing requests still go to an agent to confirm.
Run a 30-day hybrid pilot
Week 1: choose the calls
Forward only after-hours and overflow calls at first. Keep the human ISA on everything during the day so you have a comparison.
Week 1: approve the words
Sign off the greeting, the disclosure, the seller and buyer questions, the listing sheet and the routing rule. Test a live listing, a pending one and a showing request before any real call arrives; testing an AI phone assistant before launch has a test script.
Weeks 2 to 3: read the transcripts
Have the ISA read every AI-handled summary before calling that lead. Note anything wrong, missing or awkward, and fix the script the same day.
Week 4: compare outcomes
For each channel, count calls answered, consultations booked, showing requests passed on and appointments that actually happened. Ask the ISA where the freed hours went.
Decide the split
If the numbers support it, widen the AI hours and give the ISA a bigger nurture and outbound list. If they don't, roll it back. Either result is useful.
Judge the pilot on appointments held, not calls answered. A line that answers everything but books consultations the agents cannot keep has moved the problem, not solved it.
Frequently asked questions
Can an AI ISA call my old leads for me?
Not without care. The FCC treats AI-generated voices as artificial under the TCPA[3], and artificial-voice telemarketing to residential lines needs prior express written consent[7]. The SI Employee described here answers inbound calls; outbound prospecting stays with people working a scrubbed list.
Will callers know they are talking to software?
They should. The greeting can say so and include a recording notice, and you approve the wording. Most people want to know: 76% of Americans say being able to tell AI from human work is extremely or very important[11].
Does an AI ISA replace a human ISA?
For first-touch inbound calls it can cover hours a person cannot. It does not replace months of nurture calls, objection handling or judgment about when to stop calling, which is why the usual answer is a split, not a swap.
What reaches my CRM after an AI-handled call?
A recording, transcript and summary for every call, emailed to the person you choose and sent on by signed webhook or through Zapier, Make or n8n into the CRM you already use.
What should I measure in a pilot?
Calls answered by hour, consultations booked, showing requests passed on, appointments held, and the hours your ISA spent on first-touch calls before and after. Compare the same lead sources over the same weeks.
References
National Association of REALTORS® (2026). REALTORS® Adopt Technology to Save Time and Improve the Client Experience, NAR Report Finds (2026 REALTORS® Technology Report) (opens in a new tab) ↑
Zillow Research (2026). Buyers: Results from the Consumer Housing Trends Report 2026 (opens in a new tab) ↑
Federal Communications Commission (2024). FCC Makes AI-Generated Voices in Robocalls Illegal (opens in a new tab) ↑
Zillow Research (2026). Sellers: Results from the Zillow Consumer Housing Trends Report 2026 (opens in a new tab) ↑
U.S. Bureau of Labor Statistics (2025). Occupational Outlook Handbook: Real Estate Brokers and Sales Agents (opens in a new tab) ↑
Pew Research Center (2020). Most Americans don't answer cellphone calls from unknown numbers (opens in a new tab) ↑
Legal Information Institute, Cornell Law School (eCFR text) (2026). 47 CFR § 64.1200 - Delivery restrictions (TCPA rules) (opens in a new tab) ↑
Legal Information Institute, Cornell Law School (U.S. Code text) (2026). 47 U.S. Code § 227 - Restrictions on use of telephone equipment (opens in a new tab) ↑
U.S. Bureau of Labor Statistics (2025). Occupational Employment and Wage Statistics: Customer service representatives, hourly median wage, May 2025 (opens in a new tab) ↑
U.S. Bureau of Labor Statistics (2026). Employer Costs for Employee Compensation - June 2026 (opens in a new tab) ↑
Pew Research Center (2025). How Americans View AI and Its Impact on Human Abilities, Society (opens in a new tab) ↑