End-to-end architecture for replacing a human SDR function with a fully automated AI system. Covers lead intake, territory-specific pricing, AI conversation agent, cold email infrastructure, and deal packaging. Scoped and designed for a residential solar company serving PG&E and SMUD territories.
The client was a residential solar company relying on human SDRs, dealer networks, and fragmented vendors. High overhead, inconsistent close rates, and a growth ceiling tied directly to headcount. Leads went cold overnight and on weekends. Objection handling and pricing varied by rep. No single system connected prospecting, qualification, pricing, and deal handoff.
The goal was a 24/7 sales engine capable of handling thousands of simultaneous conversations, commission-free, with human involvement only at final review and signature.
Self-hosted n8n instance as the central orchestration engine. Ingests leads from prospecting tools and the client's existing Airtable bases (active leads, aged data, cancelled accounts). Handles deduplication, enrichment, validation, and error handling with retry logic throughout. Pipeline orchestration managed entirely in Airtable.
Automatically identifies each homeowner's utility provider from their address or ZIP code. Built for PG&E vs. SMUD routing in the Sacramento market with the architecture to expand to additional territories. Pricing rules, incentives, and net metering logic are applied per lead before any conversation begins.
Territory-specific calculation engine built on the client's actual pricing inputs. Runs initial load and roof estimation from available lead data, applies PG&E vs. SMUD cost modeling, and handles base pricing pass-through with adders for transformer upgrades, MPU work, and other project-specific variables. Pricing is pre-calculated before the agent contacts any lead.
GPT-4 agent trained on the client's sales scripts, objection responses, and closing methodology. Runs multi-turn structured email conversations: screens leads against qualification criteria, runs an education sequence covering solar value, financing, and incentives, handles objections using the client's actual techniques, and references territory-specific pricing throughout. Triggers handoff only when a lead is contract-ready.
High-volume cold email targeting homeowners by ZIP code within the client's service radius. Dedicated sending domains, full SPF/DKIM/DMARC authentication, inbox warmup via GSuite, suppression lists, opt-out handling, and CAN-SPAM compliance. Conservative send limits to protect deliverability as volume scales. Also reactivates dormant leads from existing client databases as day-one targets before new outreach ramps up.
On qualification, the system assembles a structured deal package: conversation transcript, lead data, system sizing output, and pricing. Formatted for a closer to review in a single pass and make the call on whether to move to signature. Pipeline visibility maintained across all active deals throughout.
The volume and complexity of this pipeline (multi-source lead intake, conditional territory routing, API integrations, error handling with retry logic) required a tool that could handle branching workflow logic without per-task pricing at scale. Self-hosted n8n gives full control over data flow and keeps operating costs flat regardless of lead volume.
Territory detection and system sizing run as upstream pipeline steps before any outreach begins. By the time the agent sends a first email, a territory-specific price range is already attached to that lead record. The agent references it directly rather than triggering a live calculation during the conversation. This removes a mid-conversation failure point and ensures pricing is consistent regardless of whether downstream tools are available at that moment.
Residential solar homeowners respond on their own schedule. Email-based multi-turn conversations remove the friction of a cold call and produce a built-in transcript that feeds directly into the deal package without extra processing. The agent can run thousands of conversations simultaneously without the concurrency limits of a voice system.
Rather than waiting for new outreach leads to flow in, the pipeline ingests the client's existing Airtable databases immediately on go-live: active leads, aged data, and cancelled accounts. This generates qualified conversations within days of launch while the cold outreach infrastructure ramps up over weeks.
The agent does not attempt to educate and close every lead that enters the pipeline. Qualification criteria run first: ownership status, property type, utility territory, and basic fit signals. Leads that do not pass are routed out before the multi-turn conversation sequence begins. This keeps the agent focused on leads worth the investment and prevents the pipeline from clogging with unqualified volume.