A framework for embedding recursive AI loops into the businesses that power local economies — and for treating the customer as the largest source of domain knowledge in the company.
The most valuable asset in any service business is the knowledge of its customers — and almost every business is throwing it away in real time.
A typical tree service generates ~40 calls a week. Every one is a customer in the moment of deciding.
Almost no local business records, transcribes, and analyzes their inbound. The signal dies on the call.
A 20% lead-quality improvement on a $1–5M business translates to six-figure annual swings.
Each layer feeds the next. Domain knowledge becomes policy. Policy drives tools. Tools execute through quality gates. Outcomes flow back as learning. The cycle compounds.
The foundation. What the business knows (services, pricing, capacity) fuses with what the customer reveals (language, objections, drivers) into a queryable representation — the kind of intelligence that used to live entirely in the owner's head.
Decision logic, made explicit. For every recurring decision — how to price a job, qualify a lead, respond to an objection, shift ad spend — the rules get written down. Versioned. Updateable. Tacit owner judgment becomes executable.
Where the system actually does work in the world. API calls to ad platforms, CRM updates, calendar events, invoice generation, GBP posts, landing-page edits. Every action is gated by an explicit policy — no rogue execution.
The layer most AI implementations skip — and the reason most fail. Before any action ships: human approval for high-stakes decisions, tone and factual checks on customer-facing copy, escalation triggers for novel situations. Where trust is earned.
Every action produces an outcome. Quotes get accepted or declined. Leads convert or don't. Owner overrides reveal policy gaps. Those signals flow back into Layer 1 — sharpening knowledge, refining policies, tightening segments.
Capture what the business knows. Capture what the customer reveals. Fuse them into a queryable asset.
Take what the business knows and turn it into rules a system can execute. Versioned, updateable, auditable.
The hands of the system. The specific tools are interchangeable — what matters is that they speak a protocol the AI can talk to. MCP is the unlock.
An AI is only as useful as the tools it can reach. Model Context Protocol (MCP) is the standard that lets an AI talk to any tool that exposes itself — your ad platform, your CRM, your inbox, your billing, your repo, your knowledge base. The specific tools change all the time. The protocol doesn't. Pick tools that can be reached. Replace them whenever something better shows up.
Every action is risk-tiered. Some auto-execute. Some queue for approval. Some always escalate. This is how trust is built.
The feedback layer. Every outcome — every conversion, every override, every objection — flows back into Layer 1 and updates the model.
Each rotation deepens the intelligence in the center. The signals get richer, the patterns sharpen, the decisions improve. The agency that walks in next year starts at zero. The self-learning business starts from year-three intelligence about itself.
Fastest feedback loop, clearest ROI. Ads, landing pages, GBP, email — all rebuilt from real customer language. This is where the foundation gets paid for.
The same intelligence layer informs pricing, dispatch, capacity, and which jobs to take. The business stops guessing about where its margin actually lives.
Owner know-how becomes a queryable asset. The business becomes legible to itself — and can be sold for a multiple it never could have commanded before.