Why Does AI Keep Giving Me Bad Estimates?
Because it does not have your costs. It has the internet's.
What's inside:
- The one blocker that stalled a software migration, and it was not the software
- A renovation contractor who cut estimate time from three hours to 45 minutes
- Five things you can fix in under an hour, starting with a date column
"I never anticipated it would be so difficult to get our costs from a supplier."
That came from a deck builder on a call this month. She is not fighting AI. She is fighting a blank cell. Her quotes keep getting blocked because the unit costs on her line items are missing. The tool works fine. The catalog behind it does not.
Bottom line up front
AI gives you a bad estimate when it does not have your numbers. Four builder conversations this month landed on the same split. The operators who fed the tool their own documented costs got hours back. The operators who pointed it at knowledge living in one person's head got nothing, or got a confident wrong answer, which is worse.
This issue covers what separates those two groups, and the cleanup that has to happen first.
How an outdoor living contractor found the real blocker in their migration
An outdoor living contractor spent this year moving off a legacy estimating platform onto JobTread. The target was a front-end launch by early winter. The blocker turned out to have nothing to do with either system.
It was the kit library. Twenty to thirty separate paver kits, one for every combination of square footage and finish level, carried over from a decade of copy-and-paste. The owner said it plainly on the call: "we don't need 40 different paver kits. You can have one."
Then a harder question landed. Had anyone ever written down the specific questions they ask a client on a first estimate visit? Measurements, site conditions, access. The answer was "we just kind of know."
That is the whole problem in four words. A process that lives in one head cannot be handed to a tool, a new hire, or a system.
The fix was parameters instead of copies. One kit, driven by square footage, instead of thirty near-identical ones. On a test 500 square foot patio, the estimator returned a range of $20,900 to $24,500. It named the historical job it pulled from. Change the parameter from 500 to 750 and every line item recalculates on its own.
Here is the honest status. They are not done. The team is still marking priority kits inside an Excel export, and full production estimating does not switch over until Q1. The cleanup is the work. Nobody sells a course on it because it is not fun.
Start where they started. Export your estimate templates to a spreadsheet and sort by name. Count the duplicates before you decide the software is the problem.
Builder spotlight: the router table argument
A renovation contractor out of Fairfield County, Connecticut has been running his estimates through AI for the past six months. He is booked for an upcoming episode of Beyond the Bid.
What he did first is the part worth copying. Before asking the tool for a single number, he uploaded his own. Cost per square foot for flooring, insulation, and spray foam. That he self-performs his labor. His true cost of carry.
Then the results showed up. A full renovation estimate that used to take him two to three hours now takes about 45 minutes. He has a job running now with a new roof, added rakes, and a large front porch. His trim takeoff came back at 461 linear feet of Versatex. That is roughly three hours he did not spend measuring.
His read on why builders should care:
"You've got to assume that 50% of your clients are running something through AI in some way, shape or form. And guys like us, if we don't make that assumption, we're in trouble."
He compares the whole thing to a router table. Everybody on the job site can buy the same one. What separates people is how well they learn to run it.
Note the order of operations. The tool did not make him fast. His documented cost per square foot made him fast. The tool just stopped him from doing the math by hand.
Your implementation playbook
Put a date next to every price. Add a column to your cost catalog for the day you last confirmed each unit cost. Anything past 90 days gets flagged, not trusted. The deck builder above had line items carrying no unit cost at all, and nobody could see it. 15 minutes.
Write down the five questions you ask on a first site visit. Most owners answer "we just kind of know." That knowledge cannot be automated, delegated, or sold with the business. Open a note on your phone after your next estimate walk and write what you asked. 10 minutes.
Count your duplicate templates before you migrate. Export your kits or assemblies to a spreadsheet and sort by name. One contractor found up to thirty versions of the same paver kit. Consolidating to one parametric kit beats migrating thirty broken ones.
Feed the tool your numbers, not the market's. Give it your actual cost per square foot, your crew rate, your true cost of carry, before you ask for an estimate. That single step is what took a renovation contractor from three hours to 45 minutes.
Track the materials you keep on hand. Drywall, plywood, small plumbing fittings. One design-build shop with 25 employees admitted they never tie those back to a specific job. That is margin walking off the truck with no record. Pick one job and one category this week.
Resources to move faster
SkillMatch Assessment. 26 questions about how your operation runs today. Scores you across five categories and ranks the fixes by dollar recovery, adjusted for your revenue and project mix. 10 minutes.
Beyond the Bid. The podcast where residential builders share what works. New episode weekly.
JobTread. Where the cost catalog and cost code structure above lives. Read their documentation on catalog setup before you build yours.
The Association of Professional Builders. Free material on builder margin and pricing structure. This is where "know your real cost" keeps coming from.
P.S.
You are not losing money because your estimating software is bad. You are losing it in the gap between what a job really costs you and what your catalog says it costs.
Close that gap and the estimate stops being a guess. You quote in 45 minutes instead of three hours, and the number holds up when the invoices land.
The mechanism is not a new tool. It is a documented cost basis, dated and owned by someone. The tool comes after.
Do it now because your clients are already running your proposals through AI. They are showing up to the table with a second opinion, and yours needs to be the one backed by real numbers.
Start with the SkillMatch assessment. 26 questions, 10 minutes, and you will see which gap is costing you the most before you spend a dollar fixing it.
It takes ten minutes. Nothing to install, nothing to schedule.
Sources worth following
If you want to go deeper on the cost side of this, Russ Stephens and the APB team publish the clearest material on builder margin and pricing structure. Modern Craftsman covers the owner side of the same problem. Contractor Growth Network is worth following on the lead-to-close half.
References
- SkillMatch Assessment
- Beyond the Bid
- GO First Consulting
- JobTread
- The Association of Professional Builders
- Modern Craftsman
- Contractor Growth Network
Builder conversations referenced in this issue are drawn from GO First Consulting client and prospect calls recorded between July 13 and July 24, 2026. All client references are anonymized.