Client Case Study: A Dubai Real Estate Agency Cuts Lead Processing Time by 3x with AI

Real estate in Dubai is a market of volume and speed. An agency receives on average 50 to 80 leads per day across Bayut, Property Finder, direct portals and WhatsApp referrals. First-contact time has become a stronger conversion driver than the price offered. In a 2024 Knight Frank study, 60% of Emirati buyers chose the agent who replied within an hour.

For Sara, who runs a 12-person agency in Business Bay, that delay had become unsustainable. Here is how 90 days of support restructured her lead handling.

The starting point

The agency covered two segments: long-term rentals and premium residential sales. Weekly volume: 180 to 220 inbound leads in French, English and Arabic.

Before the project, processing relied on a shared Google Sheet and three agents dedicated to qualification. Leads landed via email (Bayut and Property Finder forms), WhatsApp (public numbers) and occasional direct phone calls. An agent picked up each lead, asked qualification questions (budget, preferred area, property type, buying horizon), then assigned to a specialised broker.

Three recurring problems:

• Average first-contact time: 4 hours 12 minutes. During peak season, up to 9 hours.
• Incomplete qualification rate: 38%. Many leads moved to brokers without a clear budget or area.
• Pressure on the team. Qualification agents lasted on average 8 months before requesting a role change.

Sara wanted a solution that would not replace her brokers' human expertise, but would offload the initial qualification.

The approach over 90 days

Phase 1: Diagnostic and scoping (weeks 1 to 2)

4-day field audit on-site in Dubai. Mapping of lead sources, analysis of historical WhatsApp conversations (sample of 800 messages), one-on-one interviews with the 3 qualification agents and 4 brokers.

Output: a roadmap with 3 priority projects:

1. Lead centralisation in a shared Notion via API.
2. AI qualification pipeline with Claude.
3. Automated first contact via WhatsApp Business API.

Phase 2: Implementation (weeks 3 to 6)

Pipeline build. Make orchestrates the flow between lead sources (Bayut and Property Finder forms, WhatsApp) and Claude, which:

• Detects the lead's language.
• Extracts structured information (budget, area, property type, horizon).
• Generates a personalised first message in 3 to 5 lines in the lead's language.
• Enriches the Notion record with a heat score (1 to 10).

The message is sent automatically via the WhatsApp Business API within 90 seconds of receiving the lead. For scores above 7, a broker is notified immediately on Slack.

Phase 3: Pilot and adjustments (weeks 7 to 10)

Supervised launch. For 4 weeks, every generated reply went through a quick human validation before being sent. The goal: fine-tune tone, fix detection errors, enrich the Claude prompt examples.

End of week 10: 91% of generated replies are sent as is. The remaining 9% are atypical cases (VIP clients with personal handling, requests in less-recognised Egyptian dialect, declared urgency).

Phase 4: Autonomy (weeks 11 to 12)

Documentation and full transfer to the in-house team. Sara trains an AI lead who becomes the contact for future tuning.

The measured results

Comparison between month M-1 (pre-project) and month M+3 (90 days after launch):

| Indicator | Before | After | Change | | --- | --- | --- | --- | | Average first-contact time | 4h12 | 1h24 | -67% | | Complete qualification rate | 62% | 89% | +44% | | Lead to appointment conversion | 18% | 28% | +56% | | Appointment to signed deal conversion | 31% | 38% | +23% | | Lead to signed deal conversion | 5.6% | 10.6% | +89% |

The gain on final conversion (lead to signed deal) comes mainly from two effects:

• Speed. Leads contacted in under 2 hours convert 3.2 times better than those contacted beyond 4 hours.
• Qualification quality. Brokers receive enriched leads with clear budget, area and horizon, which makes the first conversation immediately productive.

On the team side, the 3 qualification agents were redeployed. Two became junior brokers, one became the AI lead. No layoffs.

What worked in the implementation

The initial field audit. Understanding the team's actual day-to-day before proposing a tool avoided several classic mistakes: for instance, replacing the human WhatsApp voice with an overly formal message, or forcing a default language when the lead writes in Moroccan darija.

Short feedback loops. Every week of the pilot phase produced a debrief with the team. Adjustments shipped in 24 to 48 hours.

Invisible AI. No lead realised they were initially exchanging with an AI system. Conversations stayed human in appearance and substance because the generated messages were calibrated on the best historical conversations.

What could have gone smoother

Detecting Arabic dialects required 3 extra iterations. Moroccan darija and colloquial Egyptian sometimes escaped Claude in the early weeks. Fix: a dictionary of local terms added to the system prompt and an immediate human fallback when the confidence score is low.

Integration with the existing CRM (PropSpace) took a week longer than planned because of poorly documented APIs. For similar projects, plan a 5-day buffer on this type of integration.

What this case study teaches

AI does not replace brokers. It frees them from time wasted on incomplete qualification and late replies. Value stays in the human relationship after first contact.

This project combines two phases: implementation (flow automation) and autonomy (in-house training to avoid external dependency).

Get in touch