stpierre.ai (Akash’s company) runs Facebook ads that get homeowners asking for concrete-repair estimates. An automated chat assistant (Jordan) collects each homeowner’s details and phone number, and those ready-to-call homeowners (“leads”) are handed to Dan, a Dallas concrete-leveling contractor — the paying customer. Selling that same service to more contractors is the second half of the business. AI assistants (the agents) do all the software work; Akash sells and approves. This page maps both halves.
Money goes in at the top. It splits into two businesses that share one engine. Click either lane to jump to its full journey.
Homeowner sees an ad → chats with the bot → gets handed to Dan’s team → Dan wins a job. This is what the company sells. Right now it runs for one client (Dan, concrete leveling, Dallas).
Getting clients Contractor sideA contractor hears about the company (ads or Akash’s calls) → sees a demo → tries it for a week → pays monthly. This is how the company gets more Dans. Right now it runs on phone calls; its ads are paused.
One homeowner in Dallas, start to finish. Click any step to see what happens, which tools run it, who (or what) does the work, and what’s wrong with it today.
⚠ Honest note: past “qualified,” the homeowner side has no real stage list — the bot invented five labels on the fly across 54 conversations. Building a proper homeowner pipeline is scoped and ready whenever someone says go.
This is the sales pipeline: 4 cold-outreach steps, 15 selling steps, 4 after-the-sale steps. It’s fully built in the CRM with automations firing on stage changes. What feeds it today is Akash making calls — the contractor-facing ads are paused.
Every step from money spent to money collected, in order. It’s tracked week-over-week on its own dashboard. Green = measured and trusted. Amber = measured but flawed. Red = not measured at all.
Two breaks matter most: Leads — Facebook receives them but can’t credit them to the ad that produced them, so its ad system can’t learn which ads work; and Booked / Showed — nothing reports back once a lead is handed to the client. Both fixes are scoped; the first just needs a yes (Decision 1 on the Execution Plan).
Seven layers of software run the business. Each card lists what’s inside and whether it’s healthy today. (Product names live in the Technical notes; here every tool is described by what it does.)
| Who | Owns | Does NOT do |
|---|---|---|
| Akash | Sales, full-time: calls, demos, closes. Every approval gate: ads going live, bot wording, anything a real customer sees. Sets the targets. | Building pages, pulling data, drafting copy — all delegated. |
| The AI agents (Claude/Codex) | Build everything: ads, pages, reports, automations. Draft every reply. Run 9 approved background agents (data pulls, quality checks, the morning read, sweeps). Log every work session to the shared memory. | Never send to a real customer without a yes. Never turn on paused ad campaigns. Never touch Akash’s landing page. |
| Jordan (the homeowner bot) | The entire homeowner conversation, 24/7, no human in the loop until the hand-off. | Doesn’t send the hand-off text itself — a separate automation does; the bot’s auto-send stays off. |
| Dan’s team (the client) | Calling handed-off homeowners, quoting, winning jobs. | Reports nothing back automatically — that’s the blind spot in section 2. |
| Collins (contract builder) | Six delegated build projects, tracked on a shared board. | Gets nothing sent to him without Akash’s explicit go (scope, pay, and go-ahead all individually gated). |
The $1M version of the machine, and what each piece looks like when it’s done. (The full worked math, including the three ways it breaks, is in the 100-day plan.)
Strategy (locked Jul 31): leads first. Generate homeowner leads as proof, inventory, and assets. A free chatbot as the front door for contractors. The real business model on the back end: a 10% growth partnership where stpierre.ai covers the marketing. Year one = concrete leveling only, US-wide. Other trades are year two.
| Piece | End state |
|---|---|
| Homeowner journey | Ad → bot → hand-off → outcome reported back, fully automatic, follow-ups armed, every lead credited to the exact ad that produced it. |
| Contractor journey | Inbound (ads + the free chatbot) fills the calendar; Akash only does demos and closes; trial → paid is a routine, not a project. |
| Data | One brain, every number trusted, nothing hand-typed, every dashboard shows its own age. |
| Reporting | 8 pages total. One daily front door, one weekly scorecard, one client report. Everything else retired. |
| Client experience | Dan sees his real numbers weekly without asking; onboarding is a 15-step checklist someone else runs. |
| Akash’s time | Mornings = calls and demos. Approvals batched once a day. Zero building. |
Every area of the business in one table. “State” tells you whether the fix is already scoped, ready to execute, or waiting on a decision.
| Area | Today | Goal | State |
|---|---|---|---|
| Ad → lead credit | Facebook can’t tell which ad made which lead; its system can’t learn. | Every lead credited; Facebook optimizes itself. | fix ready — needs a yes |
| Lead → outcome | Blind after hand-off; “jobs won” is one hand-typed number. | Outcomes flow back automatically; the proof pack builds itself. | scoped, not built |
| Bot quality | Asks ZIP before phone; loses people at the last step. | Phone first, city never blocks, photos on. | fix written — needs a yes |
| Follow-up | Nothing chases a quiet lead. | Text + email ladders fire automatically. | scripts staged — need a read + yes |
| Homeowner pipeline | Five improvised labels, no order, no rules. | A real stage list like the contractor side has. | scoped, ready to build |
| Contractor inbound | Ads paused; only Akash’s dials feed the pipeline. | Ads + the free chatbot fill the calendar. | relight decision + wrong-goal fix staged |
| Cold email | Can’t send at all — no sending domain exists. | A warmed-up domain sending sequences. | blocked — buy the domain now or slip |
| Reporting | ~27 pages, 4 zombies showing June data, numbers disagree across pages. | 8 pages, one truth, every number dated. | prune plan ready — needs a yes |
| Data trust | Ads data-pull dead since Jul 24; pre-Jul-22 funnel data quarantined; sales page stale. | All pulls green; the watchdog catches failures same-day. | fixes scoped |
| Task management | 2,279 open tasks, 7 tied to a number; the clean 44 live on the project board. | Only number-tied work exists; the rest deleted, not “someday.” | cut proposed |
| Time management | No system — attention goes to 100 things. | Calls and demos in the morning, one approval batch daily. | nothing built — this page’s Top 3 is the start |
| Scorecard | Live, but 2 of the key numbers have no target. | All key numbers targeted; predicted vs. actual weekly. | 2 numbers needed |
| Client experience | One client; his report has made-up tiles; onboarding barely exercised. | Weekly proof without asking; onboarding runs itself. | partial |
Akash’s own diagnosis: the missing ingredient is focus. Every gap above collapses into three moves. Everything else is parked until these are done.
The product is “we send you jobs.” Prove it, weekly, with real numbers. This is the 25-hand-offs-a-week goal.
Facebook can’t learn and nobody can steer while ad credit and job outcomes are broken. Fix the measuring stick once, benefit forever.
Nothing above matters without more Dans. Akash is the only revenue engine until inbound is switched back on.