UAE real estate agency: from 4-hour response times to 60-second lead qualification
A WhatsApp AI agent now handles 200+ daily enquiries, qualifies buyers in Arabic and English, matches properties via API, and routes hot leads to agents — all synced to HubSpot CRM.
Response time reduction
−78%
From 4–6 hours to under 60 seconds
Lead-to-viewing conversion
+41%
Month 1 vs. baseline
Time to go-live
26 days
Contract signed to production
The challenge
The agency received 200+ WhatsApp enquiries per day through a team of just 3 agents. Average response time was 4–6 hours. Hot leads — buyers ready to view properties within 48 hours — were going cold before first contact. The team had no lead qualification process, no property matching automation, and no CRM integration. Arabic-speaking buyers were underserved because the existing English-only chatbot could not handle dialect variation or mid-conversation language switching.
Key pain points
- 200+ daily WhatsApp enquiries overwhelming a 3-person team
- 4–6 hour average response time killing hot lead conversion
- No lead qualification — every enquiry treated as equal priority
- No automated property matching to live inventory
- Arabic buyers underserved by English-only system
- Zero CRM sync — conversation history lost after each chat
What we built
We built a WhatsApp Business API AI agent powered by GPT-4o and LangChain. The system qualifies leads (budget, timeline, property type), matches against live inventory via the agency's property API, routes hot leads to the correct agent in under 60 seconds, and syncs the full conversation to HubSpot CRM — in Arabic and English with seamless mid-conversation language switching.
Lead qualification agent
AI asks budget range, viewing timeline, and property preferences. Scores intent and flags hot leads for immediate agent handoff.
Live inventory matching
Connected to the agency's property API. The bot suggests available properties matching the buyer's criteria in real time.
Arabic + English switch
Detects language from first message and switches seamlessly mid-conversation. Handles Modern Standard Arabic and major Gulf dialects.
The results
−78%
Average response time
From 4–6 hours to under 60 seconds
+41%
Lead-to-viewing conversion
Month 1 result
26
Days to go-live
From contract to production
200+
Daily enquiries handled
Without adding headcount
How we delivered it
Day 1–3: Discovery & API audit
We mapped the agency's property API, reviewed their HubSpot CRM structure, and identified the 12 most common buyer enquiry patterns. We also recorded sample Arabic and English conversations to build the initial training set.
Day 4–7: WhatsApp Business API approval
Applied for WhatsApp Business API access through Meta's official partner channel. Prepared the agency's Facebook Business verification documents.
Day 8–14: Core agent build
Built the lead qualification flow, property matching logic, and HubSpot CRM sync. Tested Arabic language detection and mid-conversation switching with native speakers.
Day 15–21: Integration & internal testing
Connected live property API. Ran parallel testing: AI responses vs. human agent responses on 50 real enquiries. Tuned the hot-lead routing threshold based on agent feedback.
Day 22–26: Go-live & handover
Soft launch with 20% of traffic. Monitored for 4 days, fixed two edge cases in Arabic dialect handling. Full traffic switch. Delivered handover document and trained the 3 agents on escalation protocols.
Technical implementation
The system runs on a production stack designed for reliability, not demos.
LLM: GPT-4o
Primary model for reasoning, qualification logic, and natural language generation. Fallback to GPT-4o-mini for high-volume periods.
Framework: LangChain
Orchestrates the multi-step qualification flow, API calls to property inventory, and HubSpot CRM updates.
Channel: WhatsApp Business API
Official Meta API for enterprise messaging. Handles delivery receipts, read status, and media messages.
CRM: HubSpot
Two-way sync via HubSpot API. Conversations logged as activities. Contact records auto-created or updated.
Language: Custom Arabic NLP layer
Built on top of GPT-4o with prompt engineering for dialect handling and RTL rendering. Mid-conversation language detection using a lightweight classifier.
Infrastructure: n8n + Make
Workflow automation for data pipelines, error handling, and alerting. Self-hosted for data residency compliance.
Similar to your business?
If you handle high-volume inbound enquiries in WhatsApp or web chat, we can build the same system for you — with a written go-live timeline before any code is written.
Book free AI auditWhat we learned
- The decision to keep human agents in the loop for price negotiation was correct. The AI hands off cleanly before any price discussion. Removing humans from that stage would have hurt conversion — buyers in this market expect a human for negotiation.
- Mid-conversation language switching was harder than expected. The first version required the user to explicitly say 'switch to English.' The second version detects language change automatically from message content. Conversion improved 18% after this change.
- Property API latency was the biggest performance bottleneck. We added a 3-second cache layer for popular search queries. Average response time dropped from 4.2 seconds to 1.1 seconds.
- Agent adoption was higher when they could see the AI's reasoning. We added a 'why this lead is hot' summary to every handoff. Agent follow-up speed improved 34%.
“We went from 6-hour response times to instant lead qualification. Our conversion rate on inbound WhatsApp leads went up 38% in the first month. The AI handles Arabic and English seamlessly — our buyers switch languages mid-conversation and the bot follows without missing a beat.”
Head of Digital
Real Estate Group • at UAE — name withheld by NDA
Related case studies
Ready for results like this?
Start with a free AI audit to identify your fastest path to production.
Book free 30-min audit →