- Counted by
- Monthly provider invoices before and after routing each task to the cheapest model that passes its test. August 2026 against May 2026.
- Caveat
- Volume grew in the same period, so the per-task figure is the honest one; the total bill fell less.
- Counted by
- Enrichment provider charges divided by contacts recovered, cleaning company, August to September 2026. Seven sources tried in order, cheapest first.
- Caveat
- Recovery rate is not 100%. Roughly one in six stays masked and goes to a human.
- Counted by
- Hiring queue before and after the funnel went live, cleaning company, September 2026. Every application scored, invited or declined with a reason.
- Caveat
- Scoring is EEOC-guarded and a manager reviews every invite. The system moves the queue; it does not hire.
- Counted by
- Timestamps on the lead record and the outbound text, cleaning company, the after-hours lead shown on the home page, which is a real one. Median across after-hours leads that month was under ten minutes.
- Caveat
- Median, not guarantee. Leads that need a human decision wait for the human.
- Counted by
- Media brand, September to December 2026 calendar. Episodes sync from the host, products live in the store, clips cut from the long episode.
- Caveat
- This counts a schedule, not outcomes. At the snapshot one episode has gone out on its date; the rest are a plan until each date passes.
- Counted by
- Reconciler log, cleaning company, September 2026. Two scheduled payment tasks had silently stopped; one unmatched charge was reconciled the first night it ran.
- Caveat
- Finding them is the system. A person still fixed the code underneath.
Three numbers we tried to move and could not, yet.
Ninety days of scheduled posting on a service company did not change booked jobs in a way we could attribute. We stopped and put the effort into search and referrals, which did.
Own-inbox cold outreach stalled on provider gating and deliverability. Domains are warm and waiting; the number is still zero.
We built pricing intelligence that could re-price jobs on its own. It has never been switched on. A human approves every change, so that number did not move, by choice.
We publish these because a return page with only wins is an advertisement. Yours will have a column like this too, and we will tell you before we start what we think will land in it.
The teardown sets the baseline. Day ninety measures against it.
Before any install we write down the numbers the system is meant to move and how each will be counted. That is the spec the guarantee refers to. At the ninety-day review you get the same table as this page, for your company, with the same caveat column. Read one install, start to finish to see the shape of it, or run the calculator for a first estimate.