Case Study Nxt Level Roofing
One 4-person roofing company. The entire operation, automated.
How Nxt Level Roofing replaced a pile of breaking flows and an Excel "CRM" with Mission Control — one system that runs every job from estimate to collected payment, operated by the owner with no in-house developers.
Nxt Level Roofing is a 4-person insurance-restoration roofing contractor near Louisville, Kentucky, covering about 20 counties. We replaced their fragile no-code flows and spreadsheet "CRM" with a custom operations platform that automates the full job lifecycle — lead, estimate, claim, production, invoice, and collection — with AI agents handling each stage and a human approving anything that touches money, contracts, or customer communications.
The before
Like most small contractors, their back office ran on duct tape:
- Excel doubling as the CRM, with duplicate rows and race conditions corrupting job data
- A tangle of fragile no-code flows that sent duplicate emails and failed silently
- The same job re-keyed into three different tools by hand
- Claims and adjuster follow-ups tracked from memory and sticky notes
- No safe way to let automation touch anything, because nothing was tested
What we built: one system for the whole job
A custom, self-hosted operations platform — Mission Control — running the full job lifecycle: lead → estimate → claim → production → invoice → collection. Here's what got automated at each stage.
Every lead captured once, from every source
Storm canvassing, referrals, inbound calls, and web leads flow into one place. Automated ingestion of NOAA hail and storm-report data, scoped to their territory and rendered on a live map, tells reps where to focus. Every record is written with idempotent, duplicate-proof logic — so a flaky signal or a retry never creates a second copy of the same job.
Contracts out and tracked to signature
Estimates move into e-signature dispatch (Adobe Sign), triggered automatically by the job's status. Everything is tracked to signature, and nothing sends without approval — contracts are a hard-gated, human-approved step.
The insurance grind, handled
For restoration work, the platform runs the adjuster follow-up cadence, tracks supplements (RCV vs. ACV, depreciation release), and automatically assembles a denial kit when a claim is denied. Carrier-facing communications are architected in the homeowner's voice to comply with Kentucky law — the system structurally cannot send contractor-voiced claim negotiations.
A pipeline that can't double-send
Every project moves through an explicit lifecycle — new → contract sent → signed → started → materials ordered → install scheduled → completed → NOC sent → collected — and status changes are the only place messages fire. Welcome notes, materials-on-the-way, install-scheduled, and review-request updates all send automatically, each checked against a communications log so it goes out exactly once. No more duplicate or missed updates.
Closeout and getting paid, tracked
Project closeout packages the completion photos, runs the Notice-of-Completion pipeline, and tracks payment milestones through to collected. Anything touching money is hard-gated for human approval — the system does the chasing and the tracking, a person releases the money moves.
One source of truth, so nothing corrupts
We inverted the system of record: an Azure SQL database is now the truth, and Excel, SharePoint, and Trello became read-only views synced from it. Job photos (CompanyCam) drive project creation, deduped on their external ID. This ended duplicate rows, ID race conditions, and spreadsheet corruption as business risks — and because the old tools still work as views, the team's day-to-day didn't break on day one.
The system watches itself
A watchdog health-checks every flow, webhook, e-signature agreement, and database queue every 15 minutes — self-healing where it's safe and escalating to a human when it isn't. Silent failure stopped being possible.
Seven AI agents, each owning one part of the operation
The trust architecture (the part that matters)
Every agent action falls into one of three tiers — which is why a non-technical owner can safely run it:
| Tier | Behavior |
|---|---|
| 🟢 Auto-run | Fires automatically, fully logged — storm scans, health checks, data sync. |
| 🟡 Logged review | AI drafts; a human approves with one tap before anything sends — outreach, claims follow-ups, marketing. |
| 🔴 Hard gate | Money, e-signatures, and all insurer communications require explicit human approval. No bypass. |
Compliance isn't a policy memo — it's in the code. Kentucky insurance law, TCPA consent, and financial controls are enforced at the architecture level. The platform structurally cannot send communications that would violate them.
Built to be run by an owner, not an engineer
Every change ships through automated tests and a deploy pipeline — one feature per push, a green build required before the next, and a dry-run before any bulk data change. Both rules exist because violating them once caused real breakage. The result is a production platform a 4-person team operates with no in-house developers.
The results
- The whole job runs in one system — estimate to collected payment, instead of nine apps and a spreadsheet
- Every client message sent exactly once — duplicate communications eliminated by design
- Duplicate data entry gone — one database is the source of truth; the old tools are read-only views
- Claims, closeout, and payment milestones tracked automatically, with a human releasing anything involving money
- The system monitors and heals itself, escalating to a person only where law and money require it
- A 4-person team operating with the automation depth of a company several times its size
Your business has a version of this.
Every business has the same lifecycle underneath — work comes in, gets done, money gets collected. A free 20-minute audit shows you where automation pays off first.
Book a Free Audit