A smart home security company was losing leads every night. Their old chatbot couldn't answer real questions — and every edge case fell back on a human. We replaced it with a conversational AI receptionist that handles the entire pre-sales conversation, end-to-end, without anyone lifting a finger.
Sentinel Smart Homes was getting around 40 inquiries a month from homeowners and small businesses interested in smart security. On paper, they had a chatbot. In practice, it was a liability.
The bot could handle "What are your hours?" and not much else. The moment a prospect asked about pricing tiers, coverage areas, or which package fit their property — the conversation stalled. A human had to step in. And if that human wasn't available? The lead went cold.
Worse, the leads that did make it through were often half-baked. Sales staff would show up to consultations not knowing the property size, the client's budget range, or even what they were trying to achieve — security, automation, or both. It was inefficient, frustrating, and costing them business.
Every unusual question required a staff member to intervene — even outside business hours. Leads that came in at night often didn't get a response until the next morning.
The sales team had no structured info before consultations. They were flying blind — wasting time on prospects who weren't a fit.
Any AI upgrade risked the bot inventing prices or services that didn't exist — promising "solar cameras" when none were on offer — destroying trust instantly.
Lead details were scattered across WhatsApp chats, email threads, and mental notes. Nothing structured. Nothing in a CRM. Nothing actionable.
The goal wasn't to build another chatbot. It was to build a system with the discipline of a trained senior receptionist: capable of answering real questions, guiding a natural conversation, and never making things up. We called her Ada.
Ada was engineered in n8n across three distinct layers — each solving a different failure mode of the old system.
Rather than letting the AI rely on its general training data, we built a Retrieval-Augmented Generation (RAG) pipeline. All of Sentinel's actual service documents, pricing guides, and policies were ingested, chunked, embedded via Google Gemini, and stored in a Pinecone vector database. When a prospect asks "What's included in the Full Smart Home Package?" — Ada pulls the exact answer from that database. She cannot invent an answer that doesn't exist there.
This is the core innovation. Ada operates under a strict business logic framework: she cannot recommend a package until she has collected four specific data points from the prospect — property type, approximate size, primary goal (security, automation, or both), and preference level. If someone asks "What should I buy?" in the first message, she politely pivots back to discovery. She also remembers everything said earlier in the conversation, so she never asks the same question twice. The result is a qualification flow that feels natural — not like filling out a form.
Once a prospect agrees to a consultation, Ada switches from conversation mode to data entry mode. She collects name, email, phone, and preferred time — then converts relative dates ("Next Tuesday at 2pm") into structured timestamps and saves the full qualified lead profile directly to Airtable via API. When the sales team opens their CRM in the morning, they're not looking at a name and a phone number. They have the prospect's property type, size, goal, preference level, and availability — everything needed to have a productive first call.
The system went live handling the full pre-sales conversation for Sentinel's incoming inquiries — approximately 40 per month. Here's what changed.
| Metric | Before | After |
|---|---|---|
| Response time to first message | 2—4 hours (or next day) | Under 3 seconds, 24/7 |
| Staff time spent on pre-sales chat | ~3—4 hours/week | Near zero — only edge cases |
| Lead data quality in CRM | Partial or missing info | 100% of leads pre-qualified with 4 data points |
| Wasted consultations (unqualified leads) | Frequent — no filtering | ~70% reduction |
| After-hours lead capture | Lost until morning | Captured and qualified instantly |
| AI-invented answers (hallucinations) | Risk with any AI tool | Eliminated via RAG + Gatekeeper logic |
Every component was chosen for reliability, cost-efficiency, and ease of handover.
Ada was built for Sentinel, but the underlying system is domain-agnostic. Swapping the knowledge documents and qualification rules redeploys the entire stack for a new industry — without rebuilding from scratch.
Qualify emergency vs. maintenance calls. Route urgent jobs to on-call staff automatically while capturing non-urgent leads for scheduling.
Run intake conversations for new clients. Identify case type, urgency, and jurisdiction before a lawyer spends a single minute on a call.
Pre-screen tenants or buyers before scheduling viewings. Collect budget, timeline, and property preferences in a natural chat — not a form.
If you're losing leads after hours, wasting time on unqualified prospects, or manually managing pre-sales conversations — this system was built for exactly that problem.
Book a Free Automation Audit// 30 minutes — No pitch — Just your situation and whether we can solve it