An AI assistant that qualifies leads 24/7—and forwards the completed lead to the manager

A real estate agency receives an inquiry at 10:47 PM.

The manager is asleep. The “we will get back to you” form stays silent.

The client goes to a competitor who replied in 8 seconds.

We built a system that solves exactly this problem and, in the very first month, validated the hypothesis with a real client.

The problem we saw

Most small and mid-sized business websites lose 30 to 50% of potential customers not because of bad marketing and not because of a weak product.

They lose them because of the delay.

A user comes in with a question at 11:00 PM. Or on Saturday. Or simply when the manager is busy. They see a contact form and leave.

“Leave a request and we will get back to you.”

That does not work in 2026. People want an answer now, and if not from you, then from your competitor.

What we built and how it works

We did not put ChatGPT into an iframe and we did not buy a ready-made SaaS widget.

We built our own system from scratch: server-side logic in PHP, PostgreSQL, Claude AI integration via API, a Telegram channel for the manager, and an admin panel for lead management.

What the conversation looks like for the client

The person opens the chat on the site. The bot greets them within seconds and starts a conversation — not “fill out this form,” but a real dialog tailored to the business niche.

For a real estate agency, it looks like this:

• “What type of property are you looking for?”
• “What is your budget and city?”
• “For yourself or as an investment?”

No more than two questions at a time. No feeling of a questionnaire.

What the manager receives

After the client finishes the conversation, the manager receives a lead card in Telegram with all the details: name, phone, email, property type, budget, city, purchase goal, urgency, and notes the bot prepared in Ukrainian.

The manager opens Telegram and already knows everything about the client before the first call.

Escalation to a human

If the client writes “I want to talk to a manager,” the bot does not try to hold them in the chat. It instantly notifies the team and tells the client that someone will get back to them. The manager closes the chat from the admin panel, and the client sees confirmation.

First client: KB Consulting

The first implementation was for the real estate agency KB Consulting, which works with clients in Ukraine and abroad.

Before the implementation, managers handled the first-line questions manually and regularly lost clients who wrote late at night or on weekends.

After launch:

• The bot was tuned to the niche: bilingual clients, international real estate, qualification by budget and purpose
• The manager receives a ready-made lead card in Telegram instead of a raw message stream
• Setup took less than an hour: one line of code on the site

The first month is free. The client can test the system in real conditions with no risk.

Technical stack

Everything you see in action is built on:

• Claude API (Haiku) — a fast and cost-efficient model for conversations
• PHP + PostgreSQL — server logic and lead storage
• Railway — hosting with automatic deploys from GitHub
• Telegram Bot API — instant notifications for the manager
• Custom JS widget — embedded into any site with a single <script> tag

What’s next

• Consultation booking directly inside the chat — no Calendly, no external links
• A WhatsApp channel for clients who do not use Telegram
• Multi-tenant architecture — each business gets its own branded bot