A $47,000 hire costs you $51,000 to $68,000

Every pitch in this category runs the same comparison. A sixty thousand dollar employee against five hundred dollars a month of software.

It looks decisive. Both halves of it are wrong.

The employee doesn't cost sixty thousand dollars. The software doesn't do what the employee does. Here is how to work out the real number on both sides, from free public data, before anybody puts a proposal in front of you.

Multiply the wage by 1.10 to 1.46 to reach the true cost

Start with wages. The Bureau of Labor Statistics publishes national wage data by occupation. From the May 2025 estimates, the mean annual wage was $46,590 for a customer service representative, $49,350 for a secretary or administrative assistant, and $53,560 for a bookkeeping or accounting clerk. For property, real estate and community association managers the median was $69,990, and the mean was $83,710. Use the median for that role. The mean is pulled upward by a long tail of large operators.

Then add the part most owners carry in their head as a vague percentage. BLS also publishes what employers actually spend on benefits. In the March 2026 release, for office and administrative support workers in private industry, total compensation ran $36.42 an hour: $25.00 in wages and $11.41 in benefits. Benefits were 31.3% of total compensation.

That gives you a multiplier. Our arithmetic, not the BLS's: $36.42 divided by $25.00 is about 1.46. If you offer no health insurance and no retirement, the legally required piece alone is $2.56 an hour, which puts the floor at roughly 1.10.

So a customer service hire at the national mean costs you somewhere between about $51,000 and about $68,000, depending on what you offer. That range is the most useful number in this article, and it kills two claims at once. The vendor line about replacing a $50,000 employee understates it. The folklore that benefits are half of payroll overstates it.

Turnover costs 21% of salary by one review, 40% by another

You'll have seen the figure that replacing an employee costs 33% of their salary, or about $15,000. We couldn't find any published methodology behind it. The page it traces to cites no study, no sample and no method, and routes you to the publisher's own calculator.

Two sources do show their work, and they disagree in a way worth understanding. The Center for American Progress reviewed 30 case studies across 11 research papers and found a median cost of 21% of annual salary, falling to 20% for workers earning under $50,000 and 16% for those under $30,000. The Washington Center for Equitable Growth puts the average at 40%, and notes the underlying literature ranges from 2% to almost 150%.

One is a median, the other an average, and the honest answer is that turnover cost varies enormously by role. For an admin position in a small property management company, the low end of that range is the more defensible planning number. If you use a recruiter, add SHRM's 2025 figure of $5,475 as the average cost per non executive hire, from a survey of 2,371 members.

A person is 2,080 hours whether or not the phone rings

This is where most vendor arithmetic cheats, and the trick is worth naming precisely.

You pay a person for around 2,080 hours a year, busy or idle. AI is billed by the minute of work it actually does, or by the unit, or by the seat. Setting a $60,000 salary against a $500 monthly subscription only becomes a comparison once you know how many of those 2,080 hours the software absorbs, and what happens to the rest of them.

So the question to put to a vendor is narrower than the one they want to answer. Which specific tasks does this take. How many hours a week do those tasks consume in your business today. What does the person do with the hours that come back. If they can't answer the first part in plain language, the rest of the math is decoration.

Published prices, July 2026: what was actually on the page

These are the numbers we found printed, not quoted to us, and every one links to the page it came from. Checked 28 July 2026. Vendors change these.

Property management software. Buildium starts at $62 a month for its Essential tier. DoorLoop starts at $99 a month plus $3 per unit. Rentvine advertises as low as $1.50 per unit. TenantCloud prices by lease count, from $18 a month. AppFolio publishes no prices at all: the page states only that a minimum spend and a 50 unit minimum apply, and asks you to contact them.

Answering a phone with people. Ruby's published plans work out at roughly $3.45 to $5.00 per minute. Smith.ai's plans work out at roughly $7.00 to $11.50 per call.

The raw AI underneath. Retell publishes an all-in voice range of $0.07 to $0.31 per minute, with the components broken out. Vapi sells the same layer and charges $0.05 a minute for its own platform, with the speech and model costs underneath billed separately. Twilio charges $0.0085 per minute for inbound local voice.

Offshore staffing. BruntWork publishes $4 to $8 an hour for property management virtual assistants, stated as including equipment and fees, with a 20 hour weekly minimum.

Put the phone numbers side by side and you get the most interesting comparison in this article. Retail human answering runs somewhere around eleven to seventy times the raw compute cost of the AI equivalent, and a fully loaded in-house representative sits at about $32.63 an hour by our arithmetic from the BLS figures above.

That gap is not automatically a rip-off. But you are entitled to ask exactly what you are buying inside it.

There are real things in that gap. Integration with your systems. Escalation paths that work at two in the morning. Somebody accountable when it goes wrong. Compliance discipline on the calls where the law cares. A vendor who can name those things specifically is worth your time. A vendor who changes the subject is selling margin.

Gartner's 40% cancellation figure is a forecast, not a measurement

Two numbers get quoted at you constantly, and both need deflating.

Gartner predicts that more than 40% of agentic AI projects will be canceled by the end of 2027, citing cost, unclear value and weak risk controls. That is a forecast rather than an observed failure rate, and the underlying poll was of 3,412 webinar attendees. Two other lines from the same release are more useful to you: Gartner describes widespread agent washing, meaning existing chatbots and automation rebranded as agents, and estimates that of the thousands of vendors claiming agentic AI, only around 130 genuinely have it.

The other is the claim that 95% of company AI pilots produce no return. The underlying report is real, and we read it. It describes itself as preliminary findings, it is not peer reviewed, and it is built on interviews with 52 organizations plus 153 survey responses collected at conferences. Its own definition of success is whether users or executives remarked that there had been an impact, not audited profit and loss. The 95% also refers only to custom, task specific AI tools. For general purpose assistants the same report shows roughly 80% investigating, 50% piloting and 40% implementing.

The finding underneath both does survive the methodology problems, and it is the gap between piloting and shipping. Plenty of businesses try this. Far fewer are still running it in production a year later. That is the thing to ask about.

Seven questions with something checkable behind each

Each of these is grounded in something you can verify, rather than in skepticism for its own sake.

1. What is your all-in cost per minute or per call, and what is your markup over the infrastructure underneath? Published rates from Retell, Vapi and Twilio let you sanity check whether the answer is plausible.
2. Do you publish pricing, and if not, why not? Two of the biggest names in this space do not. That asymmetry is a choice they've made.
3. Is your SOC 2 a Type 1 or a Type 2, and which criteria are in scope? The vendor chooses the scope. We are SOC 2 usually means security only. Ask for the report, not the badge. Note also that the availability criteria explicitly do not set any minimum performance level, so a SOC 2 tells you nothing about uptime.
4. On the day I leave, what do I get back, in what format, and how long do you keep a copy? Get the answer in writing before you sign, not after you've decided to go.
5. Who is liable when your system tells my tenant something wrong? Read the indemnity clause. This has already been litigated in other industries.
6. Per unit, per seat, per minute or per call, and what happens to my bill if my door count drops 20%? Per unit pricing means you pay more in a good year and stay locked in through a bad one.
7. Show me one customer my size, in my business, running this in production for twelve months. Shipped, not piloted.

Work out your own number before you take a single sales call

Take the wage you actually pay. Multiply by somewhere between 1.10 and 1.46 depending on the benefits you offer. Write it down.

That one figure does more work than any vendor comparison. It tells you what an overpriced tool looks like, and it tells you what an overpriced hire looks like. Almost nobody in this industry has it to hand, and it's a free number sitting in public data.

Once you have it, you can check the sixty thousand against five hundred comparison yourself, which is all it was ever worth.