Executive Summary
A website that behaves like a technical sales engineer
The client's technical buyers arrived with feasibility questions — process selection, geometry limits, material trade-offs — that a generic contact form could not answer, and that only sat unresolved until a sales engineer manually followed up. I led the architecture, build, and governance program for a domain-aware web agent that answers those questions in real time and converts the conversation directly into a structured, RFQ-quality lead.
The Business Problem
The client's inbound funnel depended on a static inquiry form that could not handle the technical, feasibility-stage questions serious buyers actually had — process selection, tolerance limits, material trade-offs, and geometry constraints. Buyers either abandoned the form, submitted incomplete requirements, or emailed sales engineering directly, triggering multiple rounds of manual back-and-forth before a quote-ready RFQ existed. Every one of those cycles consumed senior sales engineering time on work that was fundamentally repetitive.
A baseline audit at engagement kickoff quantified the exposure: form abandonment of 54% on technical inquiry pages, an average of 3.4 email exchanges required to assemble a complete RFQ, and sales engineers spending an estimated 11 hours per week on early-stage feasibility screening that added no unique value per inquiry. Leadership had no standardized way to guarantee that every technical inquiry, regardless of who handled it, resulted in a consistent, complete, quote-ready RFQ.
The Solution
I designed and delivered a governed web agent that behaves like a technical sales engineer: it asks the right feasibility questions in buyer-friendly language, recommends viable manufacturing processes with explicit trade-offs, and produces a structured, RFQ-quality summary mapped directly to CRM fields. Every recommendation is grounded in a curated knowledge base of the client's actual capabilities, tolerances, materials, and past project precedent — not generic manufacturing knowledge.
The agent was deliberately built to stay within the client's validated capability envelope: recommendations outside known tolerances or materials are flagged rather than guessed, and ambiguous or high-complexity inquiries are routed to sales engineering with full context already captured. This correctness guardrail was the deciding factor in getting technical leadership to trust an AI system speaking on the company's behalf.
How It Works
Guidance
Process selection, constraints, and trade-offs explained in clear, buyer-friendly language rather than a static form.
Knowledge
Retrieval anchored on the client's actual capabilities, tolerances, materials, and past project precedent.
RFQ Capture
Structured intake and summarization, ready for fast sales engineering review the moment it arrives.
Architecture & Governance
The diagram below traces how a buyer's feasibility question becomes a CRM-ready RFQ, with an automated guardrail check ensuring every recommendation stays inside the client's validated capability envelope.
System Architecture — Inquiry-to-RFQ Flow
Five stages · one automated correctness checkpoint
Feasibility questions on process selection, geometry, and material trade-offs, captured conversationally
LLM agent with tool orchestration for process recommendations and trade-off explanation
Curated retrieval over the client's real capabilities, tolerances, materials, and project precedent
Recommendations outside the validated capability envelope are flagged, not guessed, and routed for review
Structured summary mapped directly to CRM fields, ready for immediate sales engineering review
Inquiry & conversion log runs beneath every stage — every feasibility question, recommendation, and RFQ outcome is recorded and reviewable.
- LLM agent with tool orchestration for capability lookup, process recommendations, and RFQ summarization.
- Curated knowledge base indexed for retrieval across processes, tolerances, materials, and constraints.
- Frontend web chat interface backed by a scalable API layer built for production traffic.
- Outputs mapped to standardized CRM/RFQ fields for consistent lead qualification across every sales engineer.
Business Impact & ROI
- Cut RFQ turnaround time by 58% by capturing complete, structured requirements in a single guided session instead of multiple email cycles.
- Increased complete-on-first-submission RFQs by 3.1x, eliminating most of the manual back-and-forth previously required to assemble a quote-ready inquiry.
- Freed an estimated 11 hours per week of senior sales engineering time previously spent on repetitive early-stage feasibility screening.
- Delivered a fully costed ROI model presented to the executive team: total program investment recovered in approximately 6 weeks of production use, driven by sales engineering time reclaimed and higher conversion, with a 4.9x return in year one.
- Improved technical-inquiry conversion by replacing a static form with a guided, premium intake experience that kept buyers engaged instead of bouncing.
- Gave sales leadership a standardized qualification bar for every inbound technical inquiry, regardless of which engineer eventually handled it.
My Role
I owned this engagement end-to-end as the client's AI/ML product leader, operating at CTO scope: I built the business case and ROI model presented to the executive team, defined the buyer journey and knowledge representation, and set the correctness guardrails that kept the agent's recommendations bounded to the client's real, validated manufacturing capabilities.
I also owned the agent and integration architecture end-to-end, from the knowledge base design through CRM field mapping, and aligned sales engineering leadership on how edge cases would be routed for human review. The result was a scalable technical sales capability the client's leadership could trust with real buyer-facing conversations, not a chatbot experiment.