Skip to content
‹ ALL CASE STUDIESConstruction

Sam's Door Shop

Sam’s Door in Chico quotes doors and windows for whole apartment buildings. Counting them used to be done by hand, on paper, from plan sets as long as 185 pages. Now the shop uploads the PDF and gets the count. In the first 11 weeks they wrote 92 quotes for 73 customers in the quote desk we built, more each month.

92
quotes written in the first 11 weeks
73
customers quoted
About 8 s
to read the door list of a 185-page plan set
2,709
old quote files imported and searchable
Sam's Door Shop schedule extraction workspace with architectural PDF regions selected.
What changed

Their estimator showed us what he does to work out the door and window schedules. It was all physical and very painful. After that he made a new Excel sheet and filled it out by hand.

The boss wanted a faster way to quote. They also need their data stored locally, so the counting app is installed on their own Windows machine.

The shop’s work before and after.
BeforeNow
Counting doors and windowsBy hand, on paper plansUpload the PDF. The door list for a 185-page set is read in about 8 seconds, then checked by a person
Finding an old priceOpen old Excel files one at a timeSearch 2,709 past quotes and 13,133 items in one place
Writing a quoteA new Excel sheet, typed by handPick items, set quantities, export in the shop’s own Excel format
From a count to a quoteRetype everythingThe estimator drafts the quote and asks about what it cannot match
A door schedule read from a plan sheet.
A door schedule read from a plan sheet.
66 marks counted and reviewed on one floor.
66 marks counted and reviewed on one floor.
The takeoff the estimator starts from.
The takeoff the estimator starts from.
In daily use

The quote desk went to the shop on 20 August 2026. By 5 November it had 92 quotes for 73 customers, written on 41 different days.

With the imported history it now holds 4,192 quotes and 31,587 quote lines.

The work grew four times with the same client: the counting app in April, the quote desk in August, the estimator in October, and supplier prices in November. Nothing was billed until we had shown the first one could be done.

“looks solid thank you for this”

From the shop, by email
Asked for, then shipped

Most of what we built after the first version came from the estimator using it on real jobs.

He got confused counting across buildings and unit types, so we added multipliers for both. The same day he sent back a finished customer quote, itemized per unit type.

An estimate run stalled in front of the client during a demo. Two weeks later a job started in 0.6 seconds, down from 5.2, and a stalled run restarts itself within 45 seconds.

Requests and when they shipped.
RequestAskedLive
Lead times on each quote line3 September8 September
A bug, with 18 log files attached16 JulyFixed 16 July
Multiply units as well as buildingsJuly22 July
Reopen an old quote and reuse it for a new customer20 August21 August
Cost from the shop’s own supplier discount sheet4 November4 November
From a PDF to a count

The estimator uploads the PDF and gets a detailed index of the sheets, then marks the schedules. Then he goes to each floor plan and adds the doors and windows, and the walls they’re attached to.

Then the takeoff summary adds the units together, then the buildings, then the total.

It used to have buildings, floors, and units. That was too complicated, so we removed floors. Units can be multiplied now too, not just buildings.

Faster and cheaper to read a plan

We tested different ways of reading the same plans. The simplest one got all 41 door rows right and cost about a quarter as much to run, so that’s the one the app uses.

Every row still gets reviewed by a person before the export.

Updates without the email

Every time we updated the app we had to upload it to Google Drive, send them an email, wait for the download, then replace the app carefully so there were not two versions on their desktop.

Since 6 August the app has a Windows installer and updates itself. There is an update button on the home screen.

The quote desk

Every quote used to be a new Excel sheet, filled out by hand. The quote desk has 13,133 active items, with prices, customers, quotes, and history.

We cleaned up the database for duplicates slowly and carefully, and standardized the naming scheme. The catalog went from 15,280 raw records to 13,133 without changing a single historical price.

He almost never looks a price up. It is almost always inferred from the old quotes. So the item search has to get the closest result to what he wants.

We added the common lead times from the old quotes to a dropdown. In the Excel export, items with the same lead time are grouped together, and item notes are red and italic.

The quote desk, with example customers.
The quote desk, with example customers.
Line items, lead times, and pricing, with example customers.
Line items, lead times, and pricing, with example customers.
From takeoff to quote

The estimator takes the takeoff and finds the items the shop has quoted before. We tested it on 4,458 quote lines it had never seen.

66% of those lines were items never quoted for an earlier customer. For those it asks a question. It does not guess a price.

Search results on held-out quotes. Median search time: 6 ms.
MeasureKeyword searchCurrent search
Correct item in first result10.3%46.2%
Correct item in top 529.4%76.8%
Correct door opening in top 53.5%80.8%

Build what’s next
with Zenit.

BOOK A CALL