The Challenge
A property rental company managed a portfolio of properties with guests reaching out across multiple communication channels — email, SMS, messaging platforms. Virtual assistants (VAs) were fielding every inquiry manually, and the process was breaking down.
The problems:
- Inconsistent responses across VAs — different staff gave different answers to the same question, with no standardised SOP for guest communication
- Slow response times (~30 minutes average) — VAs had to manually look up property details, past incidents, and booking specifics before they could respond
- Information scattered across systems — property details, guest history, and past interactions lived in Hostfully but VAs were pulling it up manually for every query
- No context awareness — VAs couldn't efficiently reference past incidents or property-specific nuances, leading to generic or incomplete answers
- Multi-channel complexity — guests came in from different channels, and each required the same manual lookup process
They needed a system that could deliver consistent, property-aware responses instantly across all channels — something VAs simply couldn't do at scale.
The Solution
Built an intelligent multi-channel communication system that intercepts guest inquiries, enriches them with property and guest context, and routes them to specialised AI agents — all with human oversight.
Multi-Channel Integration & Workflow Engine
- Hostfully integration as the communication hub — the system connects to Hostfully (their communication management platform) to intercept incoming guest messages across all channels
- Automated workflow triggers — when a guest message arrives, a workflow fires that reads the message and begins the intelligence pipeline
- Channel-agnostic processing — regardless of where the guest reaches out, the same AI-powered workflow handles the inquiry
Deep Hostfully API Integration (5 APIs)
- 5 Hostfully API integrations to pull comprehensive context — guest details, property information, booking data, communication history, and property-specific configurations
- Automatic context assembly — for each inquiry, the system pulls the relevant guest profile and maps them to their property, gathering all the context a VA would normally spend minutes looking up
- Real-time data access — no stale information, every response is grounded in the current state of the property and booking
Property-Specific AI Agents (OpenAI GPT-4 + Assistants + Vector Store)
- OpenAI Assistants with property-based mapping — each property maps to its own OpenAI Assistant with a custom system prompt tailored to that property's specifics — amenities, house rules, check-in procedures, local guides, and historical incident data
- OpenAI Vector Store for knowledge retrieval — property data is stored in OpenAI's vector store, powering each assistant with semantic search over property-specific documents and past interactions
- Thread continuity — guest communication threads are synced to OpenAI Assistant chat threads, maintaining full conversation context across interactions
- Intelligent routing — based on the property identified from the guest's booking, the inquiry is routed to the correct property-specific assistant
- Context-aware responses — the agent doesn't just answer the question, it factors in past incidents, property quirks, and guest history to deliver responses that a well-briefed human would give
Human-in-the-Loop via Trello
- Trello-based review workflow — AI-generated responses are pushed to Trello cards for human approval before sending, giving VAs a clear queue to review and approve
- Confidence-based escalation — queries the AI can't confidently answer are flagged for manual handling
- Continuous improvement — human feedback on AI drafts improves response quality over time
Data Store & Google Sheets Integration
- Centralised data store — property configurations, response templates, and operational data maintained in a structured store for the AI agents to reference
- Google Sheets integration — logging and tracking of all interactions, providing visibility into response patterns and operational metrics
Fallbacks & Failure Alerts
- Fallback handling on every integration — each of the 6 system integrations has built-in fallback logic so failures in one system don't break the entire pipeline
- Failure alerts — automated notifications when any integration encounters an error, ensuring issues are caught and resolved quickly rather than silently failing
The Impact
- 30x faster response times — from ~30 minutes average down to ~1 minute, because the system compiles property details, guest history, and past incidents automatically
- Consistent, SOP-driven responses — every guest gets the same quality of answer regardless of which VA is on shift, creating a standardised communication SOP that wasn't possible with manual processes
- VAs freed from manual lookups — staff no longer need to dig through Hostfully to find property-specific details or past incident references — it's compiled out of the box
- Property-aware intelligence — responses account for property-specific nuances and historical context that VAs would miss or forget
- Scales without adding headcount — new properties are onboarded into the RAG system, not assigned to additional VAs
Testimonial
"It was an absolute pleasure working with Yatharth. He really helped us build automation for our guest communication. Highly recommended!"
— Founder, UK property rental company