
Meeting Supe: How Conversational AI Is Changing Construction Estimating
Justin Van Wie
•
Most construction estimating software starts with a file. You export a PDF, you upload it, you draw over it. That works if you have drawings. The problem is that most residential work does not start with drawings — it starts with a conversation in a driveway, and no amount of takeoff software helps you price a job the owner has only described out loud.
That gap is why we built Supe.
What is an AI superintendent?
An AI superintendent is an assistant that asks the questions an experienced superintendent would ask about a job, then turns the answers into a priced, trade-divided estimate. The name is deliberate. A superintendent is not the person who measures — it is the person who knows that this job needs a dumpster, two weeks of supervision, protection for the floors, and a permit, because every job does, and because the estimate that forgets them is the one that loses money.
Supe is the conversational front end to Estim8 Pro's estimating engine. You describe a job. It asks what it still needs to know. Then it builds the scope, prices it against regional rates, applies overhead and profit, and hands back something you can send.
Can you really build a material list just by talking?
Yes, if the description carries a size. Tell Supe "a 12 by 16 covered deck with composite decking, cedar rails and a shingle roof" and it will identify the scope, expand it into assemblies — footings, framing, decking, railing, roofing — and produce quantities and a price.
The engine behind it holds 283 building assemblies and 59 trade checklists, of which 54 trades are reachable directly from the conversation. An assembly is not a single product; it is the way a trade actually buys and builds. "Wall" means studs, plates, sheathing, fasteners and labour in the right ratio, not forty loose line items you assemble by hand.
You can type it or speak it. In practice most people speak it, because they are standing in the driveway with the owner and one hand on a tape.
What does Supe need to know, and what happens when it does not?
It needs a size driver — a dimension, an area, or a count — and when it does not have one it says so rather than inventing a number. This is the honest limit of conversational estimating and the thing most worth understanding before you rely on it.
"Re-roof 2,200 square feet" works. "Four more footings" works. "The owner wants the pergola extended" does not, because there is no quantity in that sentence — and a scope expansion run against a missing dimension produces a confident-looking estimate built on zero.
We know this because it happened. An early run on exactly that pergola sentence returned carpentry priced at a few dollars and overhead in the hundreds, because the overhead does not scale off a driver and everything else did. The fix was not better pricing; it was making the software admit what it did not know. Supe now returns a count of the lines needing confirmation and flags whether the size drivers were found at all, and when they were not, it does not show you a number it cannot stand behind.
Practical rule: the conversation is good when it carries a size, and degrades honestly when it does not. Any vendor telling you their AI needs nothing from you is selling you a number you should not send to a client.
How does a conversation become real pricing?
Quantities are priced against national unit rates adjusted by a regional factor, and then marked up — not left at bare cost. Two steps matter and most estimating tools get the second one wrong.
Regional rates. A framing hour in Sarasota is not a framing hour in San Francisco. Estim8 Pro applies a location factor derived from Bureau of Labor Statistics wage data, currently spanning roughly 0.79 to 1.41 across US metros. A national average price list quietly misprices every job outside the average.
Markup. An estimate that stops at materials and labour is a cost sheet, not a bid. Overhead, profit, bonds and general conditions go on top, and the bid that omits them is the one you win and lose money on.
What does a conversation catch that a spreadsheet does not?
The lines that are on every job and therefore live in nobody's template. Estim8 Pro carries 15 cross-trade allowances — dumpsters and debris removal, floor and surface protection, supervision, consumables, the things a superintendent assumes and an estimator forgets. They are added automatically and netted out of the general-contractor markup so they are not charged twice.
The same logic runs in reverse on a change order. When you scope an amendment to a job already underway, those site-wide lines are dropped, because the dumpster is already in the driveway and the supervision is already being paid for. Charging for them again is how a change order gets rejected.
Is conversational estimating actually better, or just faster?
Faster is the visible benefit; the real one is that it removes the reason bids do not get sent. A decade of JBKnowledge Construction Technology Reports, surveying more than 2,000 construction professionals annually, finds estimating is the single most spreadsheet-dependent workflow in the industry — ahead of accounting and project management.
And it is mostly done by people with no one to delegate it to. NAHB's analysis of the 2022 Economic Census found roughly 80% of home builders and trade contractors are self-employed. No estimator, no back office. The bid gets done at the kitchen table after a ten-hour day, or it does not get done.
The appetite for a different answer is already there: Dodge Construction Network's SmartMarket research with CMiC found 87% of contractors expect AI to transform their work, while only 19% have changed a single workflow. The demand is not the bottleneck. The tooling has been.
Where does conversational estimating still fall short?
It is weakest where the job is least like a sentence. Three honest limits:
Complex commercial work. Large commercial takeoffs depend on drawing sets, schedules and dimension conventions that a spoken description cannot carry. Talking through a 40,000 square foot tilt-up is not the use case.
Vague scope. Covered above — no size driver, no reliable number.
Your own negotiated pricing. Regional rates are a good default, not your supplier's quote. Serious contractors do not pay list, and any estimate should be reviewed against what you actually buy at.
What it is genuinely good at is the thing that happens most: a remodel, an addition, a re-roof, a deck — described out loud, priced before the conversation ends, and sent as a proposal before the owner has called anyone else.
Frequently asked questions
What is an AI superintendent?
An assistant that asks the questions an experienced superintendent would ask about a job, then turns the answers into a priced, trade-divided estimate. Supe is Estim8 Pro's implementation, reachable by voice or typing from the app.
Can you create a construction estimate without blueprints?
Yes. Describing the job out loud is a supported path alongside photographing the plans and uploading a full PDF set. The description needs to carry a size — a dimension, an area or a count — for the pricing to be reliable.
Does conversational AI estimating include labour and overhead?
It should. Estim8 Pro prices materials and labour against regionally adjusted rates and then applies overhead, profit, bonds and general conditions. An estimate left at bare cost is a cost sheet, not a bid.
How does the app know local prices?
National unit rates are adjusted by a location factor derived from Bureau of Labor Statistics wage data, spanning roughly 0.79 to 1.41 across US metros.
What happens if the AI is unsure about something?
It tells you. Estim8 Pro returns the assumptions it had to make, ranked by dollar impact, and flags when it could not establish the size of the work rather than producing a number it cannot support.
Can Supe scope a change order, not just an original estimate?
Yes. Describing an amendment to a job in progress produces a priced change order, and the site-wide allowances already covered by the original contract are excluded rather than billed twice.