Every broker who has ever hired an inside sales agent remembers the math that justified it. One good ISA on the phones, qualifying inbound and chasing your aged leads, and suddenly you are not the bottleneck anymore. The trouble is that the math has changed. An AI ISA now does a chunk of the same job for a fraction of the monthly cost, and the question for a 2026 mortgage team is not “human or machine” so much as “where does each one actually earn its keep.”
This piece walks the real cost-per-qualified-lead math for both. Not the demo-stage hype, the line items. What a human ISA runs, what an AI ISA runs, where the AI quietly loses, and how to decide without lighting money on fire.
What an ISA actually does, and why qualified leads are the unit that matters
An ISA, human or AI, exists to do one thing: turn raw leads into qualified appointments your LOs can close. A real estate ISA is a specialized sales professional who works the phones, not the field, whose sole job is to convert leads into appointments, handling inbound lead follow-up, outbound prospecting, lead qualification, and appointment setting. Swap “listing appointment” for “loan application” and you have the mortgage version.
The reason cost-per-qualified-lead is the right yardstick (and not cost-per-lead or salary) is that qualification is where the speed game is won or lost. The research here is old, consistent, and brutal. The MIT Lead Response Management study found the odds of contacting a lead in 5 minutes versus 30 minutes drop by 100 times, and the odds of qualifying a lead in 5 minutes versus 30 minutes drop by 21 times. That study examined three years of data across six companies, over fifteen thousand leads and over one hundred thousand call attempts.
A human ISA cannot hit five minutes at 9pm on a Sunday. An AI ISA can. That single fact is what makes this comparison worth running.
What a human ISA really costs
Start with payroll, because that is the number people quote and the one that undersells the true cost.
For mortgage and real estate inside sales, the salary band is fairly settled. As of January 2026, ZipRecruiter put the average annual pay for a real estate inside sales agent in the United States at $69,398 a year, roughly $33.36 an hour. Independent coaching sources land in a similar place. An inside sales agent with little to no experience will normally make between $40,000 and $65,000 a year, while more experienced ISAs usually make between $60,000 and $80,000. Both ranges include a base salary of between $24,000 and $30,000 a year, with the rest coming from bonus.
But salary is not the cost. The real cost stacks up like this:
| Line item | What it adds |
|---|---|
| Base salary | $40,000 to $80,000 depending on experience |
| Commission / bonus | Commonly 5% to 15% of GCI, or per-appointment bonuses around $50 to $150 |
| Payroll taxes, benefits, equipment | Real money on top of base; varies by shop |
| Ramp time | Commissions for an ISA usually do not flow in during the first six months |
| Turnover risk | An ISA you took time to hire and train can cost thousands or tens of thousands of dollars if they leave and you have to replace them |
On commission structure, the sources are consistent. The most popular and effective way to pay an inside sales agent tends to be a low salary with a 5% to 15% commission at closing. Some teams pay on conducted appointments instead, which is the smarter structure if you want qualified-appointment volume rather than just bodies on a calendar.
The fully loaded number a lot of operators quote is higher than payroll alone. One 2026 write-up pegged it at a full-time ISA running $4,000 to $6,000 per month. Treat that as a directional figure rather than gospel; your real number depends on your market, your benefits, and how you structure bonus. The point stands either way: the all-in monthly is well north of the base salary divided by twelve.
There is an offshore lane that changes the math. A virtual inside sales agent can do everything an in-office ISA does, from managing new leads to cold calling, performed remotely at much lower rates. Virtual ISA services exist precisely to convert fixed payroll into a monthly fee. Vendor marketing in this space leans hard on savings claims (one provider markets saving “up to 70%” versus an in-house hire, which is the vendor’s framing, not an audited figure), so verify the actual scope and per-lead cost before you sign.
What an AI ISA really costs
Here the picture is messier, because pricing in the AI voice space is genuinely all over the map and a lot of it is not public.
What is public, as of mid-2026, clusters into a few bands. AI voice agent pricing in 2026 ranges from roughly $0.10 to $0.30 per minute for thin voicebot platforms, to $300 to $1,500 per month for real-estate-specific voicebots with included usage, to per-conversation pricing on depth-first tools. A separate 2026 breakdown landed in the same neighborhood: most platforms run between $500 and $1,500 per month depending on call volume, while some like Retell AI charge by the minute at around $0.07 per minute, which works well for lower volume.
So the AI ISA monthly, for a typical mortgage team’s volume, tends to sit in the few-hundred to roughly $1,500 range. Against a human ISA’s all-in monthly, that is the comparison in one line.
A warning on the “complete platform” pitch. Several mortgage AI voice vendors name a single product as “the best” (usually their own) and quote impressive conversion lift, like delivering 2.5x lead conversion and 57% faster follow-ups without adding headcount. Those are vendor-claimed outcomes from vendor marketing, not independently audited results for your pipeline. Treat them as the starting point for a demo question (“show me that on accounts like mine”), not a number to plug into your spreadsheet.
Some of the most relevant pricing simply is not public. In the broader AI voice and CRM space, several platforms publish nothing. One 2026 roundup flatly noted certain tools’ pricing as not disclosed publicly. When pricing is hidden, that is itself useful information: it usually means it is sales-negotiated and volume-dependent, and you should get the per-lead and per-minute numbers in writing before committing.
The honest cost-per-qualified-lead comparison
Put the two side by side on the dimensions that actually drive cost per qualified lead.
| Factor | Human ISA | AI ISA |
|---|---|---|
| Monthly cost | Often $4,000 to $6,000 all-in (varies; verify for your shop) | Commonly a few hundred to ~$1,500, or ~$0.07 to $0.30/min |
| Speed to lead | Minutes when on shift; hours nights/weekends | Most call within 30 to 90 seconds; some within 10 seconds |
| Coverage | One shift, one human, sick days, turnover | 24/7/365, no ramp, no attrition |
| Ramp | No commissions for ~6 months | Live in days, no ramp curve |
| Depth of qualification | High: reads tone, handles curveballs, builds rapport | Improving fast, but scripted-flow tools still flatten nuance |
| Compliance exposure | Trained human follows your script | You own consent, calling rules, and disclosures the AI makes |
The cost-per-qualified-lead math falls out of that table. If your AI ISA costs roughly $1,000 a month and qualifies, say, 40 leads it would otherwise have taken you hours to reach, your cost per qualified lead is in the low tens of dollars and your speed-to-lead is near zero. A human ISA at five or six grand a month has to qualify a lot of leads to match that unit cost, and physically cannot cover the nights-and-weekends window where so many mortgage inquiries land.
That is the part the salary comparison misses entirely. The expensive lead is not the one you pay for. It is the one that came in at 9:47pm and sat untouched until Monday, by which point a faster competitor already has the application. AI closes that specific gap.
The expensive lead is not the one you pay for. It is the one that came in Saturday night and was cold by Monday.
Where the AI ISA loses, and where you still want a human
This is the section the vendors skip, so here it is plainly. The smarter framing in the market is not replacement. The teams seeing the biggest wins are the ones that pair AI voice agents with redesigned ISA roles, not the ones that try to eliminate the ISA function.
A few real limits:
- Depth on the messy conversations. Many AI voice tools are built for routing and scheduling. Real-estate-specific voicebots run scripted slot-filling flows optimized for tour booking, which is great for “book me with a loan officer” and weaker for a borrower with a thin file, a recent job change, and a co-signer question. A sharp human ISA improvises. A scripted bot may not.
- Authenticity and trust. A borrower expecting a callback from a person can feel the handoff. Inbound leads might expect to hear from the agent or team leader, not a scripted ISA, and while a highly trained handoff can be seamless, it will never be perfect. That applies to AI too.
- The complex pipeline. Mortgage is a long, relationship sale. A lot of deals are long-term transactions; it is a relationship sale where you have to nurture the lead and follow up multiple times. AI is excellent at the relentless follow-up cadence. The trust-building, objection-heavy conversation that converts a fence-sitter is still where a good human earns the commission.
The practical answer most strong teams land on: AI ISA owns instant response and tireless follow-up, human ISA or LO owns the deep qualification and the close.
So which one do you buy?
Run it against your own pipeline, not a vendor’s slide.
Lean toward an AI ISA first if: your biggest leak is speed-to-lead and after-hours coverage, your inbound is high-volume and fairly repetitive, and you want predictable cost without a six-month ramp. The cost-per-qualified-lead math is hard to argue with when the alternative is leads dying overnight.
Keep (or hire) a human ISA if: your leads are complex, your conversion hinges on rapport and objection-handling, and you have the volume to justify the loaded monthly. Many shops run both, with AI on the front edge and a human on the deals that matter.
The third option is to stop assembling this yourself. Buying an AI ISA tool means you are now the one wiring it into your CRM, writing the qualification scripts, managing the consent and A2P registration, and tuning it when contact rates sag. That is a legitimate choice if you like operating systems. If you would rather have the speed-to-lead win without becoming the person who maintains the dialer, having it built and run for you is the other legitimate fork. Both are real choices; the only wrong move is paying for a tool and then not running it.
If you want to see the actual numbers for your shop before you decide, the quiz below walks your lead volume and current response time and shows you a real cost-per-qualified-lead comparison.
See your real cost per funded loan
For more on the front-edge piece of this stack, see our breakdown of AI voice agents for mortgage leads, and if you are weighing the whole done-for-you question, done-for-you mortgage marketing: when it’s worth it covers that fork in detail.
One last bit of context on why any of this matters now. Margins are thin enough that wasted leads hurt more than they used to. Per-loan production costs rose to $12,579 per loan in the first quarter of 2025, up from $11,230 in the fourth quarter of 2024, per the MBA. When it costs that much to originate, the leads you already paid to generate are too expensive to let go cold. Whichever ISA you pick, that is the math that should drive the decision.
Pricing and product details above are drawn from vendor materials and third-party sources as of June 2026 and labeled as vendor claims where applicable; figures change, so verify current pricing and terms directly before buying. We have no affiliate relationship with the vendors named. This article is general information, not legal advice; confirm any TCPA, consent, or A2P 10DLC questions with your own compliance team or counsel.