Every claim,
with its working.
Ten measurements across four engagements. Measured means we timed it on a live engagement. Modelled means the arithmetic came from the client’s own task list. Nothing here is a benchmark, and where a figure is rounded, the working underneath shows the unrounded number.
Three of the four engagements below are property businesses, because that is where we started. The exception is our own operation, where 32 agents run this company. The jobs are the same in a clinic, a restaurant or a contractor’s office — what changes is the record.
Record compiled 10 Nov 2026 · 10 measurements · 4 engagements · 1 vertical held back
| Engagement | What was measured | Result | Tier | |
|---|---|---|---|---|
| 01 | Our own operation | Agents in production | 32 agents | Measured |
| 02 | Our own operation | Runs a day, from the run log | 3,781 / day | Measured |
| 03 | Commercial brokerage | Listing preparation, per property | 19.25 → 4.05 h | Modelled |
| 04 | Commercial brokerage | Hours returned, one listing desk, a year | 122 h a year | Modelled |
| 05 | Public records | Bridgeport VGSI, per parcel | 22.8 s | Measured |
| 06 | Public records | Worcester, per parcel | 19.0 s | Measured |
| 07 | Public records | Greenville, per parcel | 16.4 s | Measured |
| 08 | Property management | Screening turnaround | 3 days → 4 h | Measured |
| 09 | Property management | Maintenance resolution | 60% faster | Measured |
| 10 | Property management | Late renewal offers | 0 late | Measured |
One vertical is published with no number, on purpose — home services, because we have not measured it yet.
We run it ourselves before we sell it.
This business is operated by scheduled agents that never sleep and never forget a deadline. Every number below is taken from our own logs, not modelled.
32 agents in production
Scheduled work across both mailboxes, the deadline board, document intake and reporting. Nothing sends without a person approving it.
3,781 runs a day
Reads, recalculations, retrievals and drafts — measured from the run log, not estimated.
A report waiting 44 days
A management report had been sitting for 44 days; it closed the day the missing files landed. A payment tracked as “overdue” for 18 days turned out to be due on the day it was paid — the date on the envelope was the mailing date, not the due date.
44 days is how long one report sat before the missing files landed
The twenty-hour listing.
A commercial broker in Hartford County keeps one number for getting a single listing to market: the preparation. His own estimate was twenty hours. We modelled his task list top to bottom and landed within 5% of it, independently.
19.25 hours to 4.05
Photographs, comps, dimensions, zoning, the write-up, and the same facts typed into three portals that each want them in a different order.
15.2 hours a listing, 122 a year
The hours that go are the ones with no decision in them: retyping, reformatting and filing. The four that stay are the ones only the broker can do.
122 hours a year
On the listing desk alone, at his volume. Client anonymised at his request; the method and the full task list are available on request.
15.2 h × 8 listings a year = 121.6 h, rounded to 122
122 h is about 15 working days of one person’s year, on the listing desk alone
No API, no problem.
Most county property records are public and hostile to automation: a search box, one parcel at a time, no data feed and no plan for one. We built a routine that drives the site the way a person does — search, wait, read, record — and timed it on real parcels in three states.
Bridgeport VGSI
Ownership, assessment, appraisal and last sale, re-checked against the source page.
Worcester
Same fields, a different county form, the same run.
Greenville
One parcel is a demo. A hundred, the night before a meeting, is a different week.
Same routine, three different forms. No API was used or asked for.
200 units, and the question that never stops.
A property management firm came with a problem that did not sound like a technology problem. The portfolio was growing; operations were drowning the team.
Three days into four hours
Applications flow through background, credit and scoring. A human reviews the final call only.
60% faster resolution
Tenants submit, the job is auto-assigned by trade, and status flows back without anyone retyping it.
Zero late offers
Triggers fire 90 days out off actual lease dates. No spreadsheet. No forgetting.
Three calendar days was the old clock; four hours is the new one
We measured ourselves first.
Buyers increasingly ask an assistant for a recommendation instead of scrolling ten results. So we ran our own version of that test on ourselves — and we were missing, for reasons we could fix.
Our own site was telling AI crawlers to stay out. A page still published a contact address retired months earlier. Claimed results had no case study behind them.
Crawler permissions opened for the assistants people actually use. Contact details corrected at the source. This page — so every claim has a number and a method under it.
A report that only ever finds problems in the client’s business is not a report, it is a pitch. We publish ours because nobody else does.
The jobs travel. The record is what changes.
Four verticals are measured and live. The rest are in build — each one gets its own agent, its own record and its own front door, and none of them ships with a number until we have one.
Your job goes in this ledger next.
Built to a written scope. Price agreed before anything is built.