Two weeks of looking. Then we build.
Most AI consulting fails the same way: ship a tool, watch nobody use it. We don't have an opinion on the tool until we've watched the day. The first two weeks are diligence. Build starts on week three, on the highest-ROI automation of three we've already ranked together.
Watch the day. Read the messages. Count the gaps.
A half-day watching whoever is actually doing the work: front desk, dispatcher, owner answering the phone at 7pm. We read the inbox. We listen to the after-hours voicemails. We open the booking calendar and count the gaps.
In metro Atlanta that half-day is on site, at your counter. Everywhere else it is two recorded screen-share sessions where you do the actual task while we watch, not a call where you describe it. Plus a 60–90 minute interview with you or your ops lead, and real sample records including the ugly ones. We will say which one you are getting, and we will not imply we have people in your city, because we don’t.
The output is a notebook of friction, not a slide deck. Specific moments where time leaks, where a customer falls through a crack, where a number nobody is watching is moving in the wrong direction. We send the notebook to you the same week.
No NDA gate, no procurement song-and-dance. We sign yours if you have one. We don't have ours to push.
Read-only access first, wherever the API allows it. You approve every rule before anything gets built.
Three opportunities, ranked. Each one has a number next to it, and a check on whether you already own the fix.
We rank by the only metric your accountant recognizes: money that comes in, hours that go out. Every line item is concrete. The build cost is what it costs us to ship it, the run cost is what your stack costs to keep it running, and the hours-saved came from real numbers we watched in week one.
You read the doc. You push back. We don't have a quota of "AI automations sold per quarter," which means cutting an opportunity from the list isn't a problem for us. If two of the three are wrong, we go back to the notebook.
Before any of it gets a build price, we run it against what you already own. QuickBooks sends invoice reminders on every plan. Jobber sends two more from Connect up. Jobber Plus includes an AI receptionist; Housecall Pro has one too. Each opportunity gets a line saying what your current software already covers, what just needs switching on, and only then where the real gap is. We build on top of that layer rather than replacing it. The work worth paying for is the coordination those tools cannot do, not a second copy of the features you already have. If configuration closes the gap on its own, that is what we recommend, and the engagement gets smaller.
The deliverable is one short doc. Three line items. No appendix.
Sample data from a fictional business, not a client result. An example ranked.json: after-hours intake saves 11.5 hours a week and recovers $3,200 a month at a $9,500 build cost; dispatcher dedup saves 2.6 hours a week; a weekly digest saves 3.5 hours.
Build the first one. Run it in pilot alongside production.
Highest-ROI automation first. We ship the smallest version that solves the actual problem, and we ship it live, not a demo, not a test setup.
For one week the new system runs alongside your existing one. The AI handles half of the after-hours intake. Your old voicemail handles the other half. We watch the difference in real time, on real customers, before it's the only thing you're running.
Edge cases surface here, in the engagement, while we're still paid to fix them. Not on day 17 after we've already moved on to the next shop.
Sample data from a fictional business, not a client result. In an example pilot week, the AI handled 47 after-hours calls and booked 31 of them (66 percent), while voicemail handled 49 and booked 14 (29 percent). Four edge cases surfaced and were fixed.
Cutover. Hand off. We can be fired in an afternoon.
Once the pilot is clean, we cut over. Old process retires.
For seven days post-cutover we're reachable for actual breakage: not feature requests, just bugs. Day 7 we hand off: runbooks, accounts in your name, and the code in an account you own.
You don't owe us a retainer. You don't depend on any software only we control. Your accounts, your code, a documented handoff: nothing is held hostage, and we built it that way on purpose.
Sample data from a fictional business, not a client result. An example cutover report: 99.94 percent uptime, 218 calls handled, 142 booked (65 percent), and $14,200 in captured revenue, with the test set passing and accounts handed off in the owner's name.
What happens after the first build.
-
return
We come back for the next opportunity on the ranked list. Usually 4–6 weeks after the first ships. Same approach: scoped, numbered, deliverable. cadence · per-engagement, not retainercycle · 2-4w
-
verified
Before we switch you over, we ship a test set the AI has to pass, a log of every decision it makes (kept 90 days), and written rules for when it hands a call to a person, not buried in a prompt. test set · decision log · handoff rulesproofs
-
handoff
Everything we build is handed off with documentation, runbooks, and the bills in your name. Your stack, your accounts, your ownership. No platform lock-in, nothing held hostage. docs · runbooks · ownership in your accountsdeliverable
-
no-retainer
We don't do monthly retainers. Every engagement has a fixed scope, fixed price, and a number to hit. If nothing else needs building, we don't invoice for nothing. fixed scope · fixed price · numeric outcometerms
Our guarantee isn't a refund policy. It's a gate.
Your automation doesn't go live until it passes a test set we show you, and we don't leave until the rollback is proven: your old process comes back in under fifteen minutes if it ever has to. If it can't pass, it doesn't ship, and you don't owe the balance.
There's no lock-in on the other side of that gate. Everything runs in accounts you own, so if you fire us, it all keeps working.