work project
AI Sales Assistant for Service Business
Services
- →AI qualification agent design and prompt engineering
- →CRM integration — automatic contact and deal creation
- →Personalised email follow-up via LLM + email API
- →Calendar booking integration (Calendly)
- →Human handoff logic and escalation routing
- →Lead scoring and intent detection
Deliverables
- ✓Website-embedded qualification agent — no human required outside business hours
- ✓CRM records auto-created with budget, timeline, and project type
- ✓Personalised follow-up email sent within 30 seconds of conversation end
- ✓Calendar link surfaced to qualified leads automatically
- ✓Full conversation transcript forwarded on escalation
Challenge
A service business was losing qualified leads because response time averaged 4–6 hours. Visitors who filled in a contact form expected fast acknowledgement; by the time a sales rep replied, many had moved on. The team also spent significant time on leads that were clearly outside budget or scope — conversations that should never have been escalated to a human.
Options Considered
- Live chat widget with human agents — immediate, but required staffing around the clock. Expensive and impractical outside business hours.
- Rule-based chatbot (Intercom, Drift) — fast to deploy, but rigid decision trees broke under unexpected questions and produced a poor experience for anything outside the happy path.
- LLM-powered qualification agent with CRM integration — chosen. Handles open-ended conversations naturally, enforces a qualification script without feeling scripted, and writes directly to the CRM on completion.
Decision
An AI agent embeds on the website and opens a conversation when a visitor shows purchase intent (time on page, scroll depth, pricing page visit). The agent asks about the project, budget range, and preferred timeline. On completion it creates a CRM contact and deal, sends a personalised email follow-up summarising the conversation, and offers a calendar link for a discovery call.
Implementation
The agent runs on GPT-4 with a structured qualification prompt that extracts budget, timeline, and project type as typed fields. A function-calling layer writes the extracted data to the CRM via API — no manual data entry. The follow-up email is drafted by the same LLM using the conversation transcript and sent via the email provider's API within seconds of the conversation ending. Calendar booking uses Calendly's embed API; the agent surfaces the link when the lead qualifies.
Human handoff is triggered when the lead score exceeds a threshold or when the visitor explicitly asks to speak to a person — the conversation and CRM record are forwarded to the sales rep with full context.
Outcome
First-response time dropped from hours to under 30 seconds. Unqualified leads reaching the sales team fell by over 60%. Qualified leads arrived in the CRM with budget, timeline, and project summary pre-filled — call preparation time cut to near zero.
Open for contract collaboration
I am available for contract-based collaboration. If you have an interesting project idea, schedule a call via Calendly.
Schedule a 30-min call