What AI Actually Costs — and What It's Actually Worth
A business owner called me last week. Good-sized field service operation, about 30 people, $15M in revenue. He'd been pitched by three different AI vendors in the past six months.
Every one of them said the same thing: "AI will save you significant time and improve efficiency across your operations."
He wanted to know what that actually meant in dollars. None of them had told him.
That's the gap we're trying to close.
Why Vendors Don't Show the Math
Two reasons. First, most of them can't. They're selling a capability, not an outcome — and there's a big difference between "this agent can handle inbound calls" and "here's exactly how many hours a week that recovers and what that's worth to your P&L."
Second, showing the math is risky. If you put a number on it, you're accountable to it. Vague "efficiency gains" are much easier to defend at month six when the pilot hasn't moved anything.
We show the math anyway. It's the only honest way to have the conversation — and it's the core of what the diagnostic produces.
Here's how we build it.
The Five Inputs
Every ROI estimate we build comes down to five inputs. No proprietary formula. No black box. Just five questions we work through with the operator.
What's the task?
Specifically: what's the thing a person is doing right now that an agent could do instead? Not "scheduling" in the abstract — the actual activity. Manually entering job tickets from voicemails. Rewriting the same status update email every afternoon. Triaging every inbound call to figure out if it's a quote request, an existing customer, or a vendor. The more specific, the better. Vague tasks produce vague estimates.
How long does it take per occurrence?
Not the official answer. The real answer. "How long does it take to reschedule a job when a tech calls in sick?" The official answer is "maybe 15 minutes." The real answer — once you trace the calls, the customer notifications, the CRM updates, the dispatch board changes — is usually closer to 45. This is where the structured interviews earn their keep. The operator doesn't know the real number until you walk through it with them.
How often does it happen?
Per day. Per week. Not "it varies." A real estimate. Most operators undercount this by a factor of two — they're thinking of the memorable instances, not the background hum. A dental practice front desk fielding recall reminder calls thinks it happens "maybe 10 times a day." Log it for a week and it's usually 25–30.
What's the fully-loaded hourly cost?
For owner time: what's an hour actually worth? If you're billing at $300/hour and doing your own dispatch coordination, that's $300 of work that didn't get billed. For staff time: salary plus burden. A $45K/year admin doing rote data entry is running you about $30/hour all-in.
What's the error or miss rate?
This one gets skipped most often. When the current process is manual, what percentage of the time does something fall through? A voicemail that doesn't get logged. A callback that doesn't happen. A quote that goes out late. That miss rate has a dollar value — and AI, done right, usually cuts it close to zero.
The Calculation
Once you have those five inputs, the math is straightforward.
That's the number we put in the diagnostic. Not "significant efficiency gains." A specific dollar figure, built from inputs the operator gave us, with the assumptions visible and auditable.
What This Looks Like in Trades
Take a 20-person HVAC company. One dispatcher, six techs, running 30–40 jobs a week.
The process we typically surface first: after-hours call handling. Every call that comes in after 5pm either goes to voicemail — and maybe gets returned in the morning, maybe doesn't — or gets routed to the owner's cell.
The math: an operation that size fields roughly 15–20 after-hours calls per week. Industry data puts the voicemail callback miss rate at 30–40% — customers call a competitor if they don't hear back within an hour after hours. Average job value for a standard HVAC service call runs $400–$800.
At the conservative end — 15 calls/week, 30% miss rate, $400 average job — that's 4–5 missed jobs per week. Roughly $80K–$100K in revenue annually that never gets booked.
A voice agent that answers every after-hours call, captures the information, books the appointment, and notifies the on-call tech costs a fraction of that to build and run.
That's the math. That's the conversation.
What This Looks Like in Professional Services
Law firm, accounting practice, architecture firm — the pattern shifts, but the logic is the same.
The process we typically surface first: intake triage. Every inbound inquiry — phone call, web form, referral — requires someone to figure out if it's worth a partner's time. At most firms that size, that triage is happening at the partner level. Because nobody else has the judgment to do it.
At a 10-person firm fielding 20 inquiries a week, with partners spending 8 minutes per inquiry at $300/hour fully loaded, that's $800/week — $40K/year — of partner time spent deciding whether something is worth a partner's time.
An intake agent trained on the firm's qualification criteria doesn't replace the partner's judgment. It handles the 60–70% of inquiries that are clearly in or clearly out, and surfaces only the borderline cases for human review.
That's not a chatbot. That's the firm's qualification logic, running 24/7, without the partner.
What the Diagnostic Actually Produces
We don't walk into the diagnostic guessing at these numbers. The structured interviews we run are specifically designed to surface the real inputs, not the sanitized ones.
By the end of three days, we've built a map of the three highest-leverage AI opportunities in the business, ranked by ROI — with the full calculation for each: inputs, assumptions, projected annual value. Plus an honest go/no-go recommendation on whether the math is there and whether JNOW is the right firm to build it.
The deliverable you leave with isn't a proposal. It's an analysis. One you can take to anyone — another vendor, your COO, your board — and stress-test.
If the numbers don't work, we say so. The diagnostic fee doesn't depend on us finding something worth building.
A simple test you can run right now
- 01What specifically does a person do, step by step, in the process that bothers you most?
- 02How long does each step actually take? (Add 30% to your first answer.)
- 03How many times per week does it happen? (Log it for three days if you're not sure.)
- 04What's the hourly cost of the person doing it?
- 05What's the cost when it doesn't happen right?
If the annualized number is under $30K, it's probably not the first thing to build. If it's over $50K, you have a conversation worth having.
We'll do the rest in the diagnostic.
Three days. A prioritized map of your highest-ROI AI opportunities, with the math behind each one. $3,000. If we can't find a play worth building, we'll say so and refund the difference.
see how the diagnostic works →