Case study · our own company

Wasatch Cleaners. Quote to cash, on itself.

The flagship install runs on our own cleaning company first. This is the honest version: what was broken, what went in, what failed on the way, and the numbers as they are counted on the return page. Nothing here is a projection.

No screenshots here on purpose. The screens are on the systems in production page and we share them live on a call.

Our own company · numbers as counted on the return pageScrollSalt Lake City · Mountain time
The problemLeads discarded

Two lead platforms sent contacts masked; the old parser threw them away. Quotes were priced from memory. Over a hundred applications sat unread.

The systemLead sentinel, quote engine, hiring funnel, care plans, local pages

Installed in stages, each one live before the next started, on a company we could not afford to break.

The number5 minutes

From a 9:42pm lead to a priced reply. Median across after-hours leads that month was under ten. Timestamped on the lead record.

The price if it were yours$25k to $45k, then $3k to $6k a month

The Operate tier. Two weeks longer than it took us, spent in shadow mode so your customers never see a first version. How that is priced.

What was actually broken

A person was the integration between six tools.

Leads arrived from the website, two lead platforms, missed calls and texts. The platforms mask the customer’s contact; the first parser we wrote discarded anything with a masked email, and we found out because a customer called to ask why nobody had replied. Quotes were priced by whoever answered, from memory, sometimes below what the job cost. The hiring inbox had over a hundred applications nobody had opened.

None of this was unusual. It is what a service business looks like at the size where the owner is still the operating system. The point of the install was to take the owner out of that seat without taking the owner out of the decisions.

  • Masked leadsRecovered to a real contact through seven enrichment sources tried cheapest-first: about five cents per recovered contact. Roughly one in six stays masked; those go to a person.
  • PricingA quote engine with a labor model, real costs and a floor it cannot go below. The floor was set with the owner in week one and has been changed twice since, both times in a table anyone can read.
  • Hiring137 applications scored in one pass, every one invited or declined with a reason, EEOC-guarded, a manager reviewing every invite before it sends.
  • MoneyA nightly reconciler that compares scheduled payments to what the processor actually took. In its first week it found two payment jobs that had silently stopped.
What failed first

The parser threw away the leads that mattered most.

The first version of the lead sentinel discarded masked contacts as junk, which is exactly the category lead platforms send. It ran for days before a customer phoned. That is the failure we tell every client about, because it is why the install now runs in shadow mode first: the system reads every lead alongside the human for two weeks, and we compare its answers to theirs before it is allowed to reply on its own.

How it went in

In the order it happened.

  1. Weeks 1 to 2Shadow mode

    Read everything, reply to nothing

    The sentinel parsed every lead, priced it, and logged what it would have sent. The owner kept working as before. We compared the two.

    • Recovery rate measured before anyone relied on it
    • The quote floor tuned against real jobs
    • The parser bug found and fixed here, not in front of customers
  2. Weeks 3 to 6Live in stages

    Leads, then quotes, then hiring

    Each system switched on when the previous one had a week of clean logs. The owner’s approval boundary was written before the first automatic reply.

    • After-hours leads answered in minutes
    • Quotes never below the floor; owner texted on every one above the standard band
    • Applications scored and slotted; a manager sends every invite
  3. Weeks 7 to 10Operate

    Care plans, local pages, the reconciler

    Recurring customers moved to three care tiers with a margin floor. Twenty-three location pages went up and began ranking. The reconciler ran its first night and found the dead jobs.

    • 3 care-plan tiers live
    • 23 location pages ranking locally
    • 2 dead payment jobs found in week one
    • Errors reach us before customers; monthly numbers arrive on a schedule
Why these numbers are real

It is our company, so it is the one we would never let break.

That is the honest reason the numbers here are real: we could not afford to fake them. Every figure above is counted the way the return page describes, caveats included.

Your teardown is
day zero.

Ninety minutes on a recorded call inside your tools: what to automate, what to replace, what it is costing you. Free, specific, and the recording is yours.

Book an AI teardown