IC Studio
B2B E-Commerce • HVAC & Climate Equipment

AI Assistant for Lead Qualification & Customer Support

How to scale sales team capacity during peak season without proportional headcount growth — an assistant integrated into the online chat and sales funnel.

Industry
B2B E-Commerce • HVAC
Format
Project development with ongoing evolution
Our Role
Discovery · Architecture · AI Engineering · Integrations
Platform
Online Chat + Knowledge Base + CRM
Focus
Lead Qualification · 24/7 Support · Business Actions
Operational Flow Architecture

Before the project

1
Team2 sales reps
2
Peak Season×10 inquiry traffic
3
Every request reaches sales reps
4
Queues and response delays
5
Loss of warm prospects

After the project

Informational inquiry
Client
Assistant
Info / spec question
Instant 24/7 answerNo rep time spent
Commercial request
Client
Assistant
Buying intent detected
Sales rep receives qualified leadWith full context & transcript
The operational challenge

Seasonal demand exceeded team capacity by 10x

During heatwaves and peak season, inquiry volume surged almost 10-fold, far exceeding the throughput of a 2-person sales team. Speed of first response became critical.

Expert time was drained by repetitive questions

Managers spent hours answering routine questions about pricing, stock, Wi-Fi setup, and legacy models instead of consulting high-intent buyers.

Hiring didn't align with seasonal unit economics

Training a new HVAC sales rep took 5–6 months due to technical complexity (capacities, system types, refrigerants, installation limits). Premature hiring was expensive, while off-season overstaffing was inefficient.

Risk of losing warm sales leads

While reps were busy with basic queries, warm prospects waited and risked leaving for faster-responding competitors.

What we build

Catalog guidance & alternative matching

The assistant provides prices, stock, system capacity, and technical specs. If a requested model is out of stock or archived, the system recommends a compatible active alternative.

  • Instant active catalog lookups
  • Functional comparison across models
  • Automatic alternative matching for legacy items

Lead qualification & rep offloading

Separates informational questions from real sales intent. Customers get instant answers to general questions; when buying intent surfaces, the lead is handed off to a manager.

  • Filtering info requests from sales opportunities
  • Sales reps spend zero time on basic lookup routine
  • Seamless lead handoff to sales reps

Triggering business actions inside the chat

Performs real business actions in the sales funnel, delivering qualified leads with full context and contact details straight to CRM.

  • Callback request creation
  • Order placement
  • Commercial proposal (RFQ) generation
  • Full conversation history attached to lead

Server-side dialogue context storage

Custom server-side context state storage enables multi-turn dialogs. The system tracks conversation history and understands follow-up questions.

  • Server-side message history storage
  • Multi-turn follow-up handling ('Do you have this model in white?')
  • Reps start sales conversations with complete context
Why this project proved more complex than it seemed
01

The system had to learn data grounding

Early prototypes answered too confidently even when information was missing. Reducing hallucination rates and building quality controls became a top priority.

10% → 1% hallucination rate within 2 months
02

The knowledge base mattered as much as the model

Product cards alone were insufficient. We transformed scattered company knowledge into one unified operational base.

Knowledge base expanded by +300% (PDFs, shipping, specs, legacy items)
03

Customers don't speak catalog language

Users ask: 'Need an AC for a 3-room flat', not 'Gree Pular Inverter GWH12AGBXB-K6DNA4A'. The retrieval engine had to be trained on natural customer phrasing.

Handling typos, transliteration, colloquial names & intent context
04

Coherent dialogue requires state memory

Standard chat platforms didn't pass history. We engineered custom server-side context storage so turns were processed as a single conversation.

Multi-turn clarification support: 'Do you have this exact model in white?'
Before vs After Transformation
BeforeAfter
Up to 10% inaccurate answers
≈1% after grounding and quality control
Scattered product cards
Unified knowledge base (+300%)
Office hours response only
24/7 customer support & lead capture
Basic chat interface
Automated Orders, RFQs & Callbacks
The operational outcome
  • Sales reps stopped spending time on routine info queries and focus strictly on purchase-ready leads
  • Sales conversations start with prepared context — reps see client interests, questions asked, and viewed alternatives
  • Customers get instant answers to routine questions and can submit requests 24/7
  • Inquiry traffic is qualified before manager involvement, eliminating peak-season bottlenecks
  • Management gained a measurable sales channel with source attribution
  • Sales department throughput scaled without proportional headcount expansion
Key Project Takeaway

“We didn't automate sales. We automated the first 70% of the customer journey before the conversation reaches a sales rep.”

Thanks to this, the company scaled seasonal peak traffic without expanding team size proportionally.

If your operations feel harder to manage than they should — we're happy to take a look.

We usually start by identifying one area that can be improved quickly.

Discuss your situation