How a US Hospitality Firm Built an AI Sales Pipeline with Kylon

The Challenge

The company sells premium event packages to corporate decision-makers, brand managers, and event planners across the United States. Their sales team was small. One coordinator could handle about 20 personalized emails a day. At that rate, reaching 1,000 qualified prospects would take months. And the follow-up sequences that actually convert were falling through the cracks.

What They Built with Kylon

Lead Sourcing with Apollo

The head of business development asked a Kylon agent to pull a targeted list from Apollo. The agent searched for decision-makers matching their ideal customer profile and built a list of 1,000 contacts with verified email addresses, company names, titles, and firmographic signals. The list lives inside Kylon as a structured dataset. No export-import dance between Apollo and a spreadsheet.

Automated Email Sequences

The agent sends 30 personalized cold emails per day, spaced throughout the day to avoid spam filters. Each email is tailored based on the prospect's company and role. The system follows a multi-step sequence:

• Initial email with a personalized hook based on company signals
• Follow-up emails if no reply
• Break-up email after a set number of attempts, ending the sequence respectfully

The agent handles the entire sequence autonomously. For high-volume cold outreach, the team chose to let the agent send without manual approval on each message, trading per-email control for throughput.

AI Reply Scoring and Routing

When prospects reply, the agent scores each response:

• Hot: The prospect is interested and wants to learn more
• Warm: They asked a question (pricing, availability, details)
• Cold: Not interested, out of office, or unsubscribe requests

For hot and warm replies, the agent checks whether it has the information to respond. If someone asks "How much is a weekend package?", the agent looks up the answer from its knowledge base and replies directly. If the prospect says "I'm interested, tell me more," the agent sends a calendar link to book a call with the sales coordinator.

For replies the agent can't handle confidently, it escalates by CC'ing the human sales coordinator, who picks it up manually.

See It in Action

The chat panel shows real-time agent updates. The dashboard tracks every lead's status, score, and last action. No tab-switching. No copy-paste between tools.

Continuous Improvement

The head of BD periodically reviews the agent's sent replies. When he spots a response that doesn't match the brand voice or misses a nuance, he screenshots it and sends it to the agent in the channel. The agent updates its internal guidelines and improves over time.

The team also uses channel guidelines to set behavioral rules the agent follows: tone, escalation thresholds, and which topics require human sign-off.

Results

• 30 personalized emails sent daily without any human effort
• 80%+ of replies handled by AI, with only edge cases escalated
• Sales call bookings happen end-to-end without manual scheduling
• The agent improves over time through screenshot feedback and guideline updates
• One sales coordinator manages what previously would have required a team of 3-4 SDRs

What Makes This Work

This isn't a chatbot bolted onto a CRM. The agent lives in the same workspace as the sales team. It reads the same guidelines, operates on the same data, and gets corrected in real time. The feedback loop is measured in minutes, not sprint cycles.

• Apollo integration for lead data without manual imports
• Email connection for sending and receiving directly from the company's domain
• Channel guidelines as a living playbook the agent follows
• Database App as a shared source of truth for the pipeline
• Human-in-the-loop where it matters, fully autonomous where it doesn't