The Challenge: 14,000 Enquiries and an 8-Hour Response Gap
An estate agency chain with 28 branches across the Midlands receives 14,000 web and portal enquiries per month. Analytics show that 43% of these arrive outside business hours — evenings, weekends, and bank holidays. The average first-response time during business hours is 4.2 hours; for out-of-hours enquiries, it stretches to 14 hours (the next morning). Industry research consistently shows that responding to a property enquiry within 5 minutes increases the probability of booking a viewing by 7x compared to responding after one hour. The agency estimates they lose 1,200 potential viewings per month — approximately £180,000 in commission opportunity — purely because of response delay.
Previous chatbot implementations failed because rigid decision trees could not handle the variety of property questions: “Does this house have a south-facing garden?”, “What are the neighbours like?”, “Can I offer below asking?”, “Is there parking for two cars?” These require property-specific knowledge and conversational flexibility that rules-based bots cannot deliver.
AI Solution: Property-Aware Conversational Chatbot
A self-hosted AI chatbot powered by an open-source LLM via vLLM handles property enquiries with genuine conversational intelligence. The chatbot is connected to the agency’s property database via RAG, enabling it to answer specific questions about individual listings — garden orientation, parking arrangements, nearby schools, EPC ratings, council tax bands. When a buyer asks “Does 42 Maple Drive have a south-facing garden?”, the chatbot retrieves the property details and either answers directly or transparently states what information is available.
Beyond question-answering, the chatbot qualifies leads (budget, timeline, mortgage status) and books viewings directly into the agency’s calendar system. Running on a dedicated GPU server, the chatbot responds in under one second, 24 hours a day, 7 days a week.
GPU Requirements
The chatbot handles concurrent conversations across 28 branches, with evening peaks of 200-300 simultaneous sessions. A 7B model fine-tuned for estate agency conversations provides the quality-speed balance needed.
| GPU Model | VRAM | Response Latency | Concurrent Conversations |
|---|---|---|---|
| NVIDIA RTX 5090 | 24 GB | ~0.8 seconds | ~300 |
| NVIDIA RTX 6000 Pro | 48 GB | ~1.0 seconds | ~450 |
| NVIDIA RTX 6000 Pro | 48 GB | ~0.7 seconds | ~500 |
| NVIDIA RTX 6000 Pro 96 GB | 80 GB | ~0.5 seconds | ~800 |
An RTX 5090 handles evening peaks comfortably. Private AI hosting ensures buyer personal information and agency sales data remain within GDPR-compliant UK infrastructure.
Recommended Stack
- vLLM serving Mistral 7B fine-tuned on 20,000 historical enquiry conversations from the agency’s CRM.
- LlamaIndex RAG connecting the chatbot to the property database for listing-specific answers.
- Calendar API integration (Reapit, Alto) for real-time viewing availability and booking.
- Lead scoring model prioritising hot leads (mortgage-ready, chain-free, urgent timeline) for immediate negotiator follow-up.
- Handoff protocol transferring complex enquiries to the duty negotiator with full conversation context.
For handling phone enquiries, add Whisper for voicemail transcription and phone-based chatbot interaction. Deploy a vision model to help buyers ask questions about specific features visible in listing photographs. Use document AI to process buyer offer documents and proof of funds.
Cost Analysis
Converting 35% more enquiries into booked viewings — an additional 1,200 viewings per month — at the agency’s historical viewing-to-sale conversion rate of 12% and average commission of £4,500, generates approximately £648,000 in additional annual commission revenue. The out-of-hours coverage alone, eliminating the 14-hour response gap, is projected to contribute £180,000 of that figure. The dedicated GPU server investment is recouped within the first week of operation.
The chatbot also reduces the negotiator team’s time spent on routine enquiries by an estimated 40%, freeing them for high-value activities: market appraisals, viewings, and offer negotiations.
Getting Started
Export 20,000 resolved enquiry conversations from your CRM, including the property discussed and the outcome (viewing booked, no further action, referred to branch). Fine-tune the LLM on this dataset. Connect to your property database and calendar system. Deploy on your website for three branches initially, measuring viewing booking rate, lead quality, and customer satisfaction against branches using traditional enquiry handling.
GigaGPU provides UK-based dedicated GPU servers optimised for real-time chatbot workloads. Scale GPU capacity as branch coverage expands.
GigaGPU offers dedicated GPU servers in UK data centres with full GDPR compliance. Deploy property chatbots on private infrastructure today.
View Dedicated GPU Plans
