I Quoted 15 Jobs Last Month Using AI. Here's What Happened.
I was skeptical that AI could write a useful contractor estimate. After running it on 15 real jobs, here is what it got right, what it got wrong, and whether I would keep using it.
The skeptic's take
When AI estimating tools started getting talked about in the contractor forums I follow, my reaction was somewhere between skeptical and dismissive. I've been doing HVAC work for eleven years. I know how to write an estimate. I know what a 4-ton Carrier installation costs in labor and materials. What is an AI going to tell me?
But I kept seeing contractors I respected mention that they were using it, and shaving real time off their estimating. So I decided to run a real experiment: use AI to generate the first draft on every estimate I sent for a month and track the results.
Here's what I found.
The test
I used Riveta's AI estimating feature on 15 jobs over 30 days. The jobs ranged from small service calls ($400) to full system replacements ($8,500). Before each estimate, I typed a plain English description of the job — something like:
"Full HVAC replacement, 2-story house, 2,400 sq ft, removing old Carrier 3.5 ton split system from 2004, installing new Carrier 4-ton split with 80,000 BTU gas furnace, new refrigerant lines, new programmable thermostat, new disconnect, haul away old equipment."
The AI generated a structured line item breakdown in under 30 seconds.
What it got right
The structure was perfect every time. Equipment, labor, line set, refrigerant, electrical, haul-away, permits — it consistently pulled out all the categories I would have broken out myself. I never had to add a category it missed.
Material descriptions were specific. It didn't write "HVAC unit." It wrote something closer to "Carrier 4-ton central air conditioner, 16 SEER2, Model 24ACC636A003" — specific enough to be defensible on a change order if the customer tries to substitute.
It got unit-cost estimates in the right ballpark for standard items. Thermostat, refrigerant per pound, haul-away — these were close enough to my actual costs that I was usually adjusting 10–15%, not rebuilding from scratch.
It was significantly faster on unusual jobs. On a mixed job I hadn't done in a few years — duct sealing plus a mini-split addition to a converted garage — I would have spent 15–20 minutes building that estimate from scratch. The AI draft took 30 seconds and was 80% of the way there.
What needed fixing every time
Labor hours. The AI consistently underestimated labor complexity. A 4-ton replacement in an attic with restricted access is not the same labor cost as one in a first-floor closet. The AI couldn't know that. I adjusted labor on every estimate.
Local material pricing. My cost for a Carrier 4-ton unit is different from what someone in a different market pays. The AI was ballpark, but I verified every equipment price against my actual distributor pricing before sending.
Site-specific conditions. Anything that required knowing the actual job — unusual duct runs, older homes with asbestos concerns flagged during the walkthrough, a customer who specifically requested a Saturday installation — had to be added manually.
Permit costs. These vary so much by jurisdiction that the AI was essentially guessing. I always replaced the permit line with my actual local number.
The time savings
In my estimation (no pun intended), the AI draft saved me an average of 12–15 minutes per estimate on standard jobs. On unusual jobs, more.
Over 15 estimates, that's 3–4 hours. In my market, 3–4 hours of billable time is real money. Even as unbillable admin time, it's three fewer hours staring at a screen instead of being on a job or with my family.
On the smaller jobs — the $400–$800 service calls — the time savings were proportionally even bigger. I used to half-skip writing detailed estimates on small jobs because it felt like overkill. Now I do it every time because it takes less than five minutes.
Would I keep using it?
Yes, without question. But with the right expectations.
AI estimating is not a replacement for pricing knowledge. If you don't know what a job should cost, an AI first draft will not save you from underpricing it. The knowledge still has to come from you.
What it is is a starting point. A very fast, consistently structured starting point that handles the administrative scaffolding of an estimate — the line item list, the categories, the descriptions — while leaving the judgment calls to you.
The contractors who will get the most out of it are the ones who already know their pricing cold and just want to get out of the first-draft phase faster.
That's me. And it's probably you.
Win the job. Lock the deposit. Move on.
Riveta is rolling out by invite. Join the waitlist and we'll reach out when your spot is ready.