An AI Sales Team for a Dubai Company: Three Languages, an 80-Second Reply, and Not a Single Lost Lead

In the spring of 2026 the region grew unstable and the market froze. The owner called me himself: "Moments like this are exactly when you have to work harder." And the company decided it needed to treat every single lead with real care.

I built a system that replies to every client message in three languages: English, Arabic, and Russian. It handles the client, walks them through a sequence, and brings them to a call with a manager. And it never misses a single lead.

Unlike the case at my own construction company, where the AI closes deals on its own, here the goal is to convert the lead to a call. We're talking about amounts from a few hundred thousand to several million dirhams. Deals like that close in person, at the showroom. The agent's job is to consult and qualify the client, then move them to a call.

How Things Were Before

When the market froze, the company made a tough decision: the sales department of four managers was cut. More than forty leads had piled up in the CRM with no reply. Leadership closed ranks and took the clients on themselves: one of the top managers ran sales. Meanwhile the leads didn't stop coming. Just the opposite: the cost of each one went up, and losing even one was no longer an option.

The Launch

From the company's side we needed: access to the CRM and WhatsApp, photos and videos of projects, and all the information a salesperson should know. That's what the knowledge base is built from. Here the commercial director built the knowledge base himself: he wanted to bring all of the company's knowledge into one place, not just for the agent but for every employee. And he built it beautifully. In general, the bigger the company, the more powerful its knowledge base. I connect the agent to any knowledge base and make sure it understands it. But I can also build one myself; that's a separate piece of work.

Within 5 days the company's team was already testing a working bot. For another month we kept fine-tuning it on real material: the company sent edits, and I made them the same day. Since April 30 the agent has been working on every incoming lead.

The owner wrote to me back then: "We're really counting on you." And after the tests the commercial director said: "I have no doubts and no fears. Conversion will go up. Let's launch!"

What It Looks Like for the Client

Every incoming lead instantly gets a message on WhatsApp. And that message is personalized, based on what the client entered in their inquiry. From there the agent runs the conversation in the client's language, which it figures out on its own from the messages.

Leads come in not only during business hours, but in the evening, at night, and on weekends too. Those are the leads I love watching most: the AI agent handles them entirely on its own.

The numbers over the time it's been running: 47% of first inquiries arrive outside business hours (8 PM to 9 AM Dubai time), and 23% on weekends. The agent's average reply time is 80 seconds.

AI as a Designer

The system recognizes incoming images, PDF files, and audio. It responds thoughtfully and sensibly, and it can even step into the role not just of a salesperson but of a designer for a moment.

A client sent a floor plan and wrote: "I am already lost." Forty messages later he had a chosen style and specific finishes: greige, light oak, hidden lighting. Part of that consultation happened in the middle of the night.

The Knowledge Base: ~105 Pages Behind Every Reply

The company has built up a large knowledge base: 14 documents, ~32,000 words, roughly 105 A4 pages. Prices, services, segments, objections, cultural nuances of communication, and a glossary in three languages.

On every request the agent fully analyzes the entire base and takes in each item. It effortlessly finds any price and answers the client.

A side effect I didn't expect: the company's entire sales department started using the agent themselves: they ask it about the product and prices instead of digging through documents. They love both the structured base and the agent itself. So for the team I made a separate Telegram bot with the same base: it answers instantly. For clients, though, the agent replies with a delay that depends on the length of the text, deliberately, to feel more like a human.

The Handoff to a Manager

The agent brings the client to the point of being ready for a call and then calls in a live person. So the manager doesn't miss that moment, a notification lands in a Telegram channel: a short summary of the conversation, the latest messages from the client and the agent, the phone number, and quick-access buttons to the CRM and WhatsApp.

The manager can take over the lead right in the middle of the conversation and keep talking; for the client it's the same chat.

Manager Didn't See the Message? A Reminder Follows in 10 Minutes

Once the manager takes the conversation into their own hands, we ran into a situation where the client writes and the manager doesn't see it right away. I added a watchdog: if 10 minutes pass with no reply, the manager gets another notification in Telegram. And when they do reply, the exact response time is recorded.

A simple and effective thing: the owner sees in real time how the managers are working. If they're taking more than a few hours to reply, they can ask on the spot: and where exactly were you?

Every CRM Lead in Telegram, with Delivery Status

The owner ran into a problem: leads come in, but the managers don't see them. Bitrix is a complex system with inconvenient notifications, and it has to be monitored constantly. I pushed every incoming lead into Telegram.

When a lead comes in, a WhatsApp message goes out to them immediately, and the manager sees it in real time in the card: "sent," and a few seconds later "delivered." And if there's no number, or it's invalid, that's immediately visible too.

The media buyers report how many leads they brought in, but in reality some of those leads have an invalid number or none at all. This way the owner can see whether the marketing budget is being spent effectively.

Emails That Keep the Conversation Going

It turned out that because the company runs Facebook ads, the leads come with an email in addition to a phone number. I built an email sequence: nicely designed, with photos of projects, that goes out to leads on days 3, 6, and 9 and warms them up.

The twist: one paragraph of the first email adapts to the client's WhatsApp conversation! The person can see the email was written specifically for them.

The sequence is three emails so far. Open rates: the first at 50%, the second at 45%, the third at 43%. Even the third email is opened by nearly one in two. The deal cycle is long, so the sequence will grow from here, so that a client keeps getting emails for months and years.

Holidays Without a Manager: a Test the System Passed on Its Own

At the end of May the UAE had a holiday stretch, and the single manager dropped out of the process for a few days. I open the system: no one is replying, the system is replying on its own. The agent greeted new leads around the clock throughout the holidays, quoted prices and budget ranges, and fended off spam. And with the client who sent the plan of her villa, it arranged the site-visit time itself: Tuesday, 9 AM.

If the agent hadn't been there, there would simply have been no one to reply. The leads would have gone nowhere.

There won't be a sales conversion percentage here: deals of this size close in person, and their cycle runs for months. The agent's job is to qualify the client and prepare them for the conversation so that all the manager has to do is call.

Here's how it plays out in practice. In the morning a client sent documents for her apartment; the agent went through the drawings, determined the scope of work, quoted a budget range, and clarified the details. Two and a half hours later the manager entered the conversation for the first time:

— Is now a good time for a call?
— Sure.

Three messages from the manager, and a call with a ready client is scheduled.

It also happens without a manager: the agent arranges the next step with the client itself, whether a call or a site visit, and creates a task in the CRM, so the manager just has to show up at the appointed time. Two different clients:

The first paid contract went exactly this way. A 570-square-meter villa, a lead from an ad. The inquiry landed at night. At 4:25 AM the system greeted the client right away with a personal message, then followed up three times by email. A live manager closed the deal over the phone. The system kept the lead from going cold. Exactly as designed.

An AI system like this works on an ongoing basis: it means constant refinements to increase conversion. Questions the agent didn't know the answer to get added to the base; the email sequence will keep growing. I train the agent like the best manager in the sales department, except this manager doesn't leave afterward, it works forever.

A system like this can be implemented in any business with a steady flow of leads. Check it against your own numbers: how many inquiries over the past month came in at night, on weekends, or while a manager was busy? All of those people went to whoever answered first, because they weren't writing only to you. According to an MIT study, replying within the first 5 minutes makes you 21 times more likely to qualify a lead than replying half an hour later. And 78% of customers buy from the company that responds first. An automated "we received your inquiry" doesn't count here: the client isn't waiting for a confirmation, but for an answer to their question: about price, timelines, "can you do…". In this case, that answer arrives in 80 seconds, in the client's language.