What Is Voice Estimating? (And How It Actually Works)

Voice estimating means describing a job out loud while you walk it — dimensions, scope, prep, your price — and letting software turn the recording into a written estimate. Speech recognition transcribes it, an extraction model structures it into line items and terms, and a server does the math. Voice Estimate is an iPhone app built around exactly that workflow: speak the job on-site, review the numbers, send the estimate before you leave the driveway.

That's the definition. The rest of this page is how it actually works — including the parts most AI estimating tools gloss over, like who does the arithmetic and where the price comes from.

How voice estimating actually works

A voice estimate goes through six steps between the walkthrough and the PDF:

1. You record the walkthrough. Talk while you walk: "12 by 14 living room, eight-foot walls, two coats, patch six nail holes, exclude the ceiling." No forms, no typing with a tape measure in one hand.
2. Speech recognition transcribes it. The transcription model is primed with trade vocabulary — thinset, fascia, linear feet, squares of shingles — so job-site terms come through as the words you said, not near-misses.
3. An extraction model structures the transcript. It sorts your words into line items, measurements, prices, and terms. The numbers it outputs are the literal numbers you spoke. "12 by 14" is captured as 12 and 14 — the model never multiplies them.
4. The server does the math. Wall area, tile with waste, roofing squares, line-item totals, the estimate total — all computed in ordinary code, the same way a calculator app computes. Language models are good at understanding speech and bad at reliable arithmetic, so ours is never asked to do any.
5. You review the draft. Anything the app wasn't sure about is flagged with a warning right on the draft. You check quantities, fix anything the site conditions change, and confirm the price is yours.
6. You send the PDF. A clean, professional estimate with scope, line items, terms, and validity — done while the details are still fresh, not three nights later at the kitchen table.

What the AI does — and what it never does

Most AI estimating tools make the same pitch: describe the job and the software prices it. We think that's backwards. A language model that invents your total is guessing about your labor rate, your market, and your margin — and dressing the guess up in a professional-looking PDF. So Voice Estimate draws a hard line through the middle of the estimate:

The split is simple: the AI is trusted with language, the server is trusted with math, and you are trusted with money. Every number on the final page is either a number you said, a number you typed, or server arithmetic on those numbers — which means every number is one you can defend at the customer's door.

What a voice estimate sounds like

There's no command syntax to learn. You describe the job the way you'd describe it to a sub:

"Estimate an 1,800 square foot asphalt shingle roof with ten percent waste. One layer tear-off, architectural shingles, ice and water shield at the eaves, new pipe boots. Lump sum is 13,400 dollars. Fifty percent deposit to start, and the price is good for 30 days."

Here's where each phrase lands. "1,800 square foot" and "ten percent waste" become measurements. Tear-off, shingles, ice and water shield, and pipe boots become scope and line items. "13,400 dollars" is your price — spoken by you, not generated. "Fifty percent deposit to start" is filed under acceptance terms, as a condition of the job. And "good for 30 days" sets the validity date, because you actually said it.

Common mistakes when dictating an estimate

The workflow is forgiving — you're talking, not filling out a form — but a few habits make the difference between a draft you sign off in a minute and one you end up rewriting:

• Dropping the units. "Twelve by fourteen" works for a room, but "one fifty" on its own doesn't work for a fence line. Say "feet" or "linear feet" every time a number could be read two ways — that's what keeps a mixed-units warning off your draft.
• Leaving the price out. If you don't speak a rate or a lump sum, the draft comes back with scope and quantities but nothing priced — the app won't fill that in for you, by design. Say the number on-site or plan to type it in before sending.
• Describing scope without exclusions. "Exclude the ceiling and trim" is one sentence during the walkthrough and an argument avoided after the job starts. If you're not doing it, say you're not doing it.
• Keeping the terms in your head. Speak the deposit and the validity — "fifty percent deposit to start, good for 30 days" — and they land on the estimate as written terms instead of something you meant to mention.
• Recording next to the compressor. Transcription tuned for trade vocabulary still can't hear through running equipment. Step away for the ten seconds it takes to say the numbers.

What happens when a measurement is ambiguous?

It gets flagged, not guessed. When a recording leaves room for interpretation, Voice Estimate puts a named review warning on the draft instead of silently picking an answer — because a wrong quantity that looks confident is worse than a question mark you can resolve in five seconds.

Three warnings you'll see:

• Measurement inferred — the app had to read a measurement between the lines instead of hearing it stated outright. Confirm it against your tape.
• Mixed units — the recording switched units mid-stream, say feet in one phrase and inches in another where it matters. Check which one you meant.
• Calculator assumption used — a material calculator ran with a default you didn't specify, like eight-foot post spacing on a fence. Fine if that's your build; change it if it isn't.

This is the honest answer to "how accurate is it?" No transcription is perfect on a windy roof. The difference is what the software does with its own uncertainty: hide it inside a tidy draft, or show it to you before the estimate leaves your phone.

Which trades does voice estimating cover?

Any trade that can describe a job out loud can estimate by voice — the extraction step doesn't care whether you're pricing pavers or window wraps. Four trades go further in Voice Estimate, with automatic material calculators that turn spoken measurements into quantities:

• Paint — wall area from room dimensions, gallons from coats and coverage.
• Fence — posts, rails, pickets, and concrete bags from linear feet and spacing.
• Tile — tile quantity with waste from floor dimensions.
• Roof — squares from area and waste percentage.

For every other trade — decks, concrete, gutters, pressure washing, and the rest — Voice Estimate captures the job as dictated scope and line items you price yourself. Same walkthrough, same review, same PDF; the material takeoff stays in your head instead of a calculator. Either way, the quantities are planning aids: verify them against site conditions before you order anything.

Is voice estimating private?

It should be — you're describing a customer's home and your prices into a microphone. Voice Estimate works without an account: estimates, clients, and settings are stored locally on your device. When you process a recording, the audio is sent over an encrypted connection for transcription and extraction, and it isn't stored on our servers after processing. Recording works offline; the processing step needs a connection, so a no-signal site just means record now, process from the truck. The full details are in our privacy policy — it's short enough to actually read.